replication.c

/* Asynchronous replication implementation.
 *
 * Copyright (c) 2009-Present, Redis Ltd.
 * All rights reserved.
 *
 * Copyright (c) 2024-present, Valkey contributors.
 * All rights reserved.
 *
 * Licensed under your choice of the Redis Source Available License 2.0
 * (RSALv2) or the Server Side Public License v1 (SSPLv1).
 *
 * Portions of this file are available under BSD3 terms; see REDISCONTRIBUTIONS for more information.
 */

/*
 * replication.c - Replication Management
 *
 * This file contains the implementation of Redis's replication logic, which
 * enables data synchronization between master and replica instances.
 * It handles:
 * - Master-to-replica synchronization
 * - Full and partial resynchronizations
 * - Replication backlog management
 * - State machines for replica operations
 * - RDB Channel for Full Sync  (lookup "rdb channel for full sync")
 */

#include "server.h"
#include "cluster.h"
#include "bio.h"
#include "functions.h"
#include "connection.h"

#include <memory.h>
#include <sys/time.h>
#include <unistd.h>
#include <fcntl.h>
#include <sys/socket.h>
#include <sys/stat.h>

void replicationDiscardCachedMaster(void);
void replicationResurrectCachedMaster(connection *conn);
void replicationSendAck(void);
int replicaPutOnline(client *slave);
void replicaStartCommandStream(client *slave);
int cancelReplicationHandshake(int reconnect);
static void rdbChannelFullSyncWithMaster(connection *conn);
static int rdbChannelAbortRdbTransfer(void);
static void rdbChannelBufferReplData(connection *conn);
static void rdbChannelReplDataBufInit(void);
static void rdbChannelSuccess(void);

/* We take a global flag to remember if this instance generated an RDB
 * because of replication, so that we can remove the RDB file in case
 * the instance is configured to have no persistence. */
int RDBGeneratedByReplication = 0;


/* A reference to diskless loading rio to abort it asynchronously. It's needed
 * for rdbchannel replication. While loading from rdbchannel connection, we may
 * yield back to eventloop. If main channel connection detects a network problem
 * we want to abort loading. It calls rioAbort() in this case, so next rioRead()
 * from rdbchannel connection will return error to cancel loading safely. */
static rio *disklessLoadingRio = NULL;

/* --------------------------- Utility functions ---------------------------- */

/* Returns 1 if the replica is rdbchannel and there is an associated main
 * channel slave with that. */
int replicationCheckHasMainChannel(client *replica) {
    if (!(replica->flags & CLIENT_REPL_RDB_CHANNEL) ||
        !replica->main_ch_client_id ||
        lookupClientByID(replica->main_ch_client_id) == NULL)
    {
        return 0;
    }
    return 1;
}

/* During rdb channel replication, replica opens two connections. From master
 * POV, these connections are distinct replicas in server.slaves. This function
 * counts associated replicas as one and returns logical replica count. */
unsigned long replicationLogicalReplicaCount(void) {
    unsigned long count = 0;
    listNode *ln;
    listIter li;

    listRewind(server.slaves,&li);
    while ((ln = listNext(&li))) {
        client *replica = listNodeValue(ln);
        if (!replicationCheckHasMainChannel(replica))
            count++;
    }
    return count;
}

static ConnectionType *connTypeOfReplication(void) {
    if (server.tls_replication) {
        return connectionTypeTls();
    }

    return connectionTypeTcp();
}

/* Return the pointer to a string representing the slave ip:listening_port
 * pair. Mostly useful for logging, since we want to log a slave using its
 * IP address and its listening port which is more clear for the user, for
 * example: "Closing connection with replica 10.1.2.3:6380". */
char *replicationGetSlaveName(client *c) {
    static char buf[NET_HOST_PORT_STR_LEN];
    char ip[NET_IP_STR_LEN];

    ip[0] = '\0';
    buf[0] = '\0';
    if (c->slave_addr ||
        connAddrPeerName(c->conn,ip,sizeof(ip),NULL) != -1)
    {
        char *addr = c->slave_addr ? c->slave_addr : ip;
        if (c->slave_listening_port)
            formatAddr(buf,sizeof(buf),addr,c->slave_listening_port);
        else
            snprintf(buf,sizeof(buf),"%s:<unknown-replica-port>",addr);
    } else {
        snprintf(buf,sizeof(buf),"client id #%llu",
            (unsigned long long) c->id);
    }
    return buf;
}

/* Plain unlink() can block for quite some time in order to actually apply
 * the file deletion to the filesystem. This call removes the file in a
 * background thread instead. We actually just do close() in the thread,
 * by using the fact that if there is another instance of the same file open,
 * the foreground unlink() will only remove the fs name, and deleting the
 * file's storage space will only happen once the last reference is lost. */
int bg_unlink(const char *filename) {
    int fd = open(filename,O_RDONLY|O_NONBLOCK);
    if (fd == -1) {
        /* Can't open the file? Fall back to unlinking in the main thread. */
        return unlink(filename);
    } else {
        /* The following unlink() removes the name but doesn't free the
         * file contents because a process still has it open. */
        int retval = unlink(filename);
        if (retval == -1) {
            /* If we got an unlink error, we just return it, closing the
             * new reference we have to the file. */
            int old_errno = errno;
            close(fd);  /* This would overwrite our errno. So we saved it. */
            errno = old_errno;
            return -1;
        }
        bioCreateCloseJob(fd, 0, 0);
        return 0; /* Success. */
    }
}

/* ---------------------------------- MASTER -------------------------------- */

void createReplicationBacklog(void) {
    serverAssert(server.repl_backlog == NULL);
    server.repl_backlog = zmalloc(sizeof(replBacklog));
    server.repl_backlog->ref_repl_buf_node = NULL;
    server.repl_backlog->unindexed_count = 0;
    server.repl_backlog->blocks_index = raxNew();
    server.repl_backlog->histlen = 0;
    /* We don't have any data inside our buffer, but virtually the first
     * byte we have is the next byte that will be generated for the
     * replication stream. */
    server.repl_backlog->offset = server.master_repl_offset+1;
}

/* This function is called when the user modifies the replication backlog
 * size at runtime. It is up to the function to resize the buffer and setup it
 * so that it contains the same data as the previous one (possibly less data,
 * but the most recent bytes, or the same data and more free space in case the
 * buffer is enlarged). */
void resizeReplicationBacklog(void) {
    if (server.repl_backlog_size < CONFIG_REPL_BACKLOG_MIN_SIZE)
        server.repl_backlog_size = CONFIG_REPL_BACKLOG_MIN_SIZE;
    if (server.repl_backlog)
        incrementalTrimReplicationBacklog(REPL_BACKLOG_TRIM_BLOCKS_PER_CALL);
}

void freeReplicationBacklog(void) {
    serverAssert(listLength(server.slaves) == 0);
    if (server.repl_backlog == NULL) return;

    /* Decrease the start buffer node reference count. */
    if (server.repl_backlog->ref_repl_buf_node) {
        replBufBlock *o = listNodeValue(
            server.repl_backlog->ref_repl_buf_node);
        serverAssert(o->refcount == 1); /* Last reference. */
        o->refcount--;
    }

    /* Replication buffer blocks are completely released when we free the
     * backlog, since the backlog is released only when there are no replicas
     * and the backlog keeps the last reference of all blocks. */
    freeReplicationBacklogRefMemAsync(server.repl_buffer_blocks,
                            server.repl_backlog->blocks_index);
    resetReplicationBuffer();
    zfree(server.repl_backlog);
    server.repl_backlog = NULL;
}

/* To make search offset from replication buffer blocks quickly
 * when replicas ask partial resynchronization, we create one index
 * block every REPL_BACKLOG_INDEX_PER_BLOCKS blocks. */
void createReplicationBacklogIndex(listNode *ln) {
    server.repl_backlog->unindexed_count++;
    if (server.repl_backlog->unindexed_count >= REPL_BACKLOG_INDEX_PER_BLOCKS) {
        replBufBlock *o = listNodeValue(ln);
        uint64_t encoded_offset = htonu64(o->repl_offset);
        raxInsert(server.repl_backlog->blocks_index,
                  (unsigned char*)&encoded_offset, sizeof(uint64_t),
                  ln, NULL);
        server.repl_backlog->unindexed_count = 0;
    }
}

/* Rebase replication buffer blocks' offset since the initial
 * setting offset starts from 0 when master restart. */
void rebaseReplicationBuffer(long long base_repl_offset) {
    raxFree(server.repl_backlog->blocks_index);
    server.repl_backlog->blocks_index = raxNew();
    server.repl_backlog->unindexed_count = 0;

    listIter li;
    listNode *ln;
    listRewind(server.repl_buffer_blocks, &li);
    while ((ln = listNext(&li))) {
        replBufBlock *o = listNodeValue(ln);
        o->repl_offset += base_repl_offset;
        createReplicationBacklogIndex(ln);
    }
}

void resetReplicationBuffer(void) {
    server.repl_buffer_mem = 0;
    server.repl_buffer_blocks = listCreate();
    listSetFreeMethod(server.repl_buffer_blocks, zfree);
}

int canFeedReplicaReplBuffer(client *replica) {
    /* Don't feed replicas that only want the RDB. */
    if (replica->flags & CLIENT_REPL_RDBONLY) return 0;

    /* Don't feed replicas that are still waiting for BGSAVE to start. */
    if (replica->replstate == SLAVE_STATE_WAIT_BGSAVE_START ||
        replica->replstate == SLAVE_STATE_WAIT_RDB_CHANNEL) return 0;

    /* Don't feed replicas that are going to be closed ASAP. */
    if (replica->flags & CLIENT_CLOSE_ASAP) return 0;

    return 1;
}

/* Create the replication backlog if needed. */
void createReplicationBacklogIfNeeded(void) {
    if (listLength(server.slaves) == 1 && server.repl_backlog == NULL) {
        /* When we create the backlog from scratch, we always use a new
         * replication ID and clear the ID2, since there is no valid
         * past history. */
        changeReplicationId();
        clearReplicationId2();
        createReplicationBacklog();
        serverLog(LL_NOTICE,"Replication backlog created, my new "
                            "replication IDs are '%s' and '%s'",
                            server.replid, server.replid2);
    }
}
/* Similar with 'prepareClientToWrite', note that we must call this function
 * before feeding replication stream into global replication buffer, since
 * clientHasPendingReplies in prepareClientToWrite will access the global
 * replication buffer to make judgements. */
int prepareReplicasToWrite(void) {
    listIter li;
    listNode *ln;
    int prepared = 0;

    listRewind(server.slaves,&li);
    while((ln = listNext(&li))) {
        client *slave = ln->value;
        if (!canFeedReplicaReplBuffer(slave)) continue;
        if (prepareClientToWrite(slave) == C_ERR) continue;
        prepared++;
    }

    return prepared;
}

/* Wrapper for feedReplicationBuffer() that takes Redis string objects
 * as input. */
void feedReplicationBufferWithObject(robj *o) {
    char llstr[LONG_STR_SIZE];
    void *p;
    size_t len;

    if (o->encoding == OBJ_ENCODING_INT) {
        len = ll2string(llstr,sizeof(llstr),(long)o->ptr);
        p = llstr;
    } else {
        len = sdslen(o->ptr);
        p = o->ptr;
    }
    feedReplicationBuffer(p,len);
}

/* Generally, we only have one replication buffer block to trim when replication
 * backlog size exceeds our setting and no replica reference it. But if replica
 * clients disconnect, we need to free many replication buffer blocks that are
 * referenced. It would cost much time if there are a lots blocks to free, that
 * will freeze server, so we trim replication backlog incrementally. */
void incrementalTrimReplicationBacklog(size_t max_blocks) {
    serverAssert(server.repl_backlog != NULL);

    size_t trimmed_blocks = 0;
    while (server.repl_backlog->histlen > server.repl_backlog_size &&
           trimmed_blocks < max_blocks)
    {
        /* We never trim backlog to less than one block. */
        if (listLength(server.repl_buffer_blocks) <= 1) break;

        /* Replicas increment the refcount of the first replication buffer block
         * they refer to, in that case, we don't trim the backlog even if
         * backlog_histlen exceeds backlog_size. This implicitly makes backlog
         * bigger than our setting, but makes the master accept partial resync as
         * much as possible. So that backlog must be the last reference of
         * replication buffer blocks. */
        listNode *first = listFirst(server.repl_buffer_blocks);
        serverAssert(first == server.repl_backlog->ref_repl_buf_node);
        replBufBlock *fo = listNodeValue(first);
        if (fo->refcount != 1) break;

        /* We don't try trim backlog if backlog valid size will be lessen than
         * setting backlog size once we release the first repl buffer block. */
        if (server.repl_backlog->histlen - (long long)fo->size <=
            server.repl_backlog_size) break;

        /* Decr refcount and release the first block later. */
        fo->refcount--;
        trimmed_blocks++;
        server.repl_backlog->histlen -= fo->size;

        /* Go to use next replication buffer block node. */
        listNode *next = listNextNode(first);
        server.repl_backlog->ref_repl_buf_node = next;
        serverAssert(server.repl_backlog->ref_repl_buf_node != NULL);
        /* Incr reference count to keep the new head node. */
        ((replBufBlock *)listNodeValue(next))->refcount++;

        /* Remove the node in recorded blocks. */
        uint64_t encoded_offset = htonu64(fo->repl_offset);
        raxRemove(server.repl_backlog->blocks_index,
            (unsigned char*)&encoded_offset, sizeof(uint64_t), NULL);

        /* Delete the first node from global replication buffer. */
        serverAssert(fo->refcount == 0 && fo->used == fo->size);
        server.repl_buffer_mem -= (fo->size +
            sizeof(listNode) + sizeof(replBufBlock));
        listDelNode(server.repl_buffer_blocks, first);
    }

    /* Set the offset of the first byte we have in the backlog. */
    server.repl_backlog->offset = server.master_repl_offset -
                              server.repl_backlog->histlen + 1;
}

/* Free replication buffer blocks that are referenced by this client. */
void freeReplicaReferencedReplBuffer(client *replica) {
    if (replica->ref_repl_buf_node != NULL) {
        /* Decrease the start buffer node reference count. */
        replBufBlock *o = listNodeValue(replica->ref_repl_buf_node);
        serverAssert(o->refcount > 0);
        o->refcount--;
        incrementalTrimReplicationBacklog(REPL_BACKLOG_TRIM_BLOCKS_PER_CALL);
    }
    replica->ref_repl_buf_node = NULL;
    replica->ref_block_pos = 0;
}

/* Append bytes into the global replication buffer list, replication backlog and
 * all replica clients use replication buffers collectively, this function replace
 * 'addReply*', 'feedReplicationBacklog' for replicas and replication backlog,
 * First we add buffer into global replication buffer block list, and then
 * update replica / replication-backlog referenced node and block position. */
void feedReplicationBuffer(char *s, size_t len) {
    static long long repl_block_id = 0;

    if (server.repl_backlog == NULL) return;

    while(len > 0) {
        size_t start_pos = 0; /* The position of referenced block to start sending. */
        listNode *start_node = NULL; /* Replica/backlog starts referenced node. */
        int add_new_block = 0; /* Create new block if current block is total used. */
        listNode *ln = listLast(server.repl_buffer_blocks);
        replBufBlock *tail = ln ? listNodeValue(ln) : NULL;

        /* Append to tail string when possible. */
        if (tail && tail->size > tail->used) {
            start_node = listLast(server.repl_buffer_blocks);
            start_pos = tail->used;
            /* Copy the part we can fit into the tail, and leave the rest for a
             * new node */
            size_t avail = tail->size - tail->used;
            size_t copy = (avail >= len) ? len : avail;
            memcpy(tail->buf + tail->used, s, copy);
            tail->used += copy;
            s += copy;
            len -= copy;
            server.master_repl_offset += copy;
            server.repl_backlog->histlen += copy;
        }
        if (len) {
            /* Create a new node, make sure it is allocated to at
             * least PROTO_REPLY_CHUNK_BYTES */
            size_t usable_size;
            /* Avoid creating nodes smaller than PROTO_REPLY_CHUNK_BYTES, so that we can append more data into them,
             * and also avoid creating nodes bigger than repl_backlog_size / 16, so that we won't have huge nodes that can't
             * trim when we only still need to hold a small portion from them. */
            size_t limit = max((size_t)server.repl_backlog_size / 16, (size_t)PROTO_REPLY_CHUNK_BYTES);
            size_t size = min(max(len, (size_t)PROTO_REPLY_CHUNK_BYTES), limit);
            tail = zmalloc_usable(size + sizeof(replBufBlock), &usable_size);
            /* Take over the allocation's internal fragmentation */
            tail->size = usable_size - sizeof(replBufBlock);
            size_t copy = (tail->size >= len) ? len : tail->size;
            tail->used = copy;
            tail->refcount = 0;
            tail->repl_offset = server.master_repl_offset + 1;
            tail->id = repl_block_id++;
            memcpy(tail->buf, s, copy);
            listAddNodeTail(server.repl_buffer_blocks, tail);
            /* We also count the list node memory into replication buffer memory. */
            server.repl_buffer_mem += (usable_size + sizeof(listNode));
            add_new_block = 1;
            if (start_node == NULL) {
                start_node = listLast(server.repl_buffer_blocks);
                start_pos = 0;
            }
            s += copy;
            len -= copy;
            server.master_repl_offset += copy;
            server.repl_backlog->histlen += copy;
        }

        /* For output buffer of replicas. */
        listIter li;
        listRewind(server.slaves,&li);
        while((ln = listNext(&li))) {
            client *slave = ln->value;
            if (!canFeedReplicaReplBuffer(slave)) continue;

            /* Update shared replication buffer start position. */
            if (slave->ref_repl_buf_node == NULL) {
                slave->ref_repl_buf_node = start_node;
                slave->ref_block_pos = start_pos;
                /* Only increase the start block reference count. */
                ((replBufBlock *)listNodeValue(start_node))->refcount++;
            }

            /* Check output buffer limit only when add new block. */
            if (add_new_block) closeClientOnOutputBufferLimitReached(slave, 1);
        }

        /* For replication backlog */
        if (server.repl_backlog->ref_repl_buf_node == NULL) {
            server.repl_backlog->ref_repl_buf_node = start_node;
            /* Only increase the start block reference count. */
            ((replBufBlock *)listNodeValue(start_node))->refcount++;

            /* Replication buffer must be empty before adding replication stream
             * into replication backlog. */
            serverAssert(add_new_block == 1 && start_pos == 0);
        }
        if (add_new_block) {
            createReplicationBacklogIndex(listLast(server.repl_buffer_blocks));

            /* It is important to trim after adding replication data to keep the backlog size close to
             * repl_backlog_size in the common case. We wait until we add a new block to avoid repeated
             * unnecessary trimming attempts when small amounts of data are added. See comments in
             * freeMemoryGetNotCountedMemory() for details on replication backlog memory tracking. */
            incrementalTrimReplicationBacklog(REPL_BACKLOG_TRIM_BLOCKS_PER_CALL);
        }
    }
}

/* Propagate write commands to replication stream.
 *
 * This function is used if the instance is a master: we use the commands
 * received by our clients in order to create the replication stream.
 * Instead if the instance is a replica and has sub-replicas attached, we use
 * replicationFeedStreamFromMasterStream() */
void replicationFeedSlaves(list *slaves, int dictid, robj **argv, int argc) {
    int j, len;
    char llstr[LONG_STR_SIZE];

    /* In case we propagate a command that doesn't touch keys (PING, REPLCONF) we
     * pass dbid=-1 that indicate there is no need to replicate `select` command. */
    serverAssert(dictid == -1 || (dictid >= 0 && dictid < server.dbnum));

    /* If the instance is not a top level master, return ASAP: we'll just proxy
     * the stream of data we receive from our master instead, in order to
     * propagate *identical* replication stream. In this way this slave can
     * advertise the same replication ID as the master (since it shares the
     * master replication history and has the same backlog and offsets). */
    if (server.masterhost != NULL) return;

    /* If there aren't slaves, and there is no backlog buffer to populate,
     * we can return ASAP. */
    if (server.repl_backlog == NULL && listLength(slaves) == 0) {
        /* We increment the repl_offset anyway, since we use that for tracking AOF fsyncs
         * even when there's no replication active. This code will not be reached if AOF
         * is also disabled. */
        server.master_repl_offset += 1;
        return;
    }

    /* We can't have slaves attached and no backlog. */
    serverAssert(!(listLength(slaves) != 0 && server.repl_backlog == NULL));

    /* Update the time of sending replication stream to replicas. */
    server.repl_stream_lastio = server.unixtime;

    /* Must install write handler for all replicas first before feeding
     * replication stream. */
    prepareReplicasToWrite();

    /* Send SELECT command to every slave if needed. */
    if (dictid != -1 && server.slaveseldb != dictid) {
        robj *selectcmd;

        /* For a few DBs we have pre-computed SELECT command. */
        if (dictid >= 0 && dictid < PROTO_SHARED_SELECT_CMDS) {
            selectcmd = shared.select[dictid];
        } else {
            int dictid_len;

            dictid_len = ll2string(llstr,sizeof(llstr),dictid);
            selectcmd = createObject(OBJ_STRING,
                sdscatprintf(sdsempty(),
                "*2\r\n$6\r\nSELECT\r\n$%d\r\n%s\r\n",
                dictid_len, llstr));
        }

        feedReplicationBufferWithObject(selectcmd);

        if (dictid < 0 || dictid >= PROTO_SHARED_SELECT_CMDS)
            decrRefCount(selectcmd);

        server.slaveseldb = dictid;
    }

    /* Write the command to the replication buffer if any. */
    char aux[LONG_STR_SIZE+3];

    /* Add the multi bulk reply length. */
    aux[0] = '*';
    len = ll2string(aux+1,sizeof(aux)-1,argc);
    aux[len+1] = '\r';
    aux[len+2] = '\n';
    feedReplicationBuffer(aux,len+3);

    for (j = 0; j < argc; j++) {
        long objlen = stringObjectLen(argv[j]);

        /* We need to feed the buffer with the object as a bulk reply
         * not just as a plain string, so create the $..CRLF payload len
         * and add the final CRLF */
        aux[0] = '$';
        len = ll2string(aux+1,sizeof(aux)-1,objlen);
        aux[len+1] = '\r';
        aux[len+2] = '\n';
        feedReplicationBuffer(aux,len+3);
        feedReplicationBufferWithObject(argv[j]);
        feedReplicationBuffer(aux+len+1,2);
    }
}

/* This is a debugging function that gets called when we detect something
 * wrong with the replication protocol: the goal is to peek into the
 * replication backlog and show a few final bytes to make simpler to
 * guess what kind of bug it could be. */
void showLatestBacklog(void) {
    if (server.repl_backlog == NULL) return;
    if (listLength(server.repl_buffer_blocks) == 0) return;
    if (server.hide_user_data_from_log) {
        serverLog(LL_NOTICE,"hide-user-data-from-log is on, skip logging backlog content to avoid spilling PII.");
        return;
    }

    size_t dumplen = 256;
    if (server.repl_backlog->histlen < (long long)dumplen)
        dumplen = server.repl_backlog->histlen;

    sds dump = sdsempty();
    listNode *node = listLast(server.repl_buffer_blocks);
    while(dumplen) {
        if (node == NULL) break;
        replBufBlock *o = listNodeValue(node);
        size_t thislen = o->used >= dumplen ? dumplen : o->used;
        sds head = sdscatrepr(sdsempty(), o->buf+o->used-thislen, thislen);
        sds tmp = sdscatsds(head, dump);
        sdsfree(dump);
        dump = tmp;
        dumplen -= thislen;
        node = listPrevNode(node);
    }

    /* Finally log such bytes: this is vital debugging info to
     * understand what happened. */
    serverLog(LL_NOTICE,"Latest backlog is: '%s'", dump);
    sdsfree(dump);
}

/* This function is used in order to proxy what we receive from our master
 * to our sub-slaves. */
#include <ctype.h>
void replicationFeedStreamFromMasterStream(char *buf, size_t buflen) {
    /* There must be replication backlog if having attached slaves. */
    if (listLength(server.slaves)) serverAssert(server.repl_backlog != NULL);
    if (server.repl_backlog) {
        /* Must install write handler for all replicas first before feeding
         * replication stream. */
        prepareReplicasToWrite();
        feedReplicationBuffer(buf,buflen);
    }
}

void replicationFeedMonitors(client *c, list *monitors, int dictid, robj **argv, int argc) {
    /* Fast path to return if the monitors list is empty or the server is in loading. */
    if (monitors == NULL || listLength(monitors) == 0 || server.loading) return;
    listNode *ln;
    listIter li;
    int j;
    sds cmdrepr = sdsnew("+");
    robj *cmdobj;
    struct timeval tv;

    gettimeofday(&tv,NULL);
    cmdrepr = sdscatprintf(cmdrepr,"%ld.%06ld ",(long)tv.tv_sec,(long)tv.tv_usec);
    if (c->flags & CLIENT_SCRIPT) {
        cmdrepr = sdscatprintf(cmdrepr,"[%d lua] ",dictid);
    } else if (c->flags & CLIENT_UNIX_SOCKET) {
        cmdrepr = sdscatprintf(cmdrepr,"[%d unix:%s] ",dictid,server.unixsocket);
    } else {
        cmdrepr = sdscatprintf(cmdrepr,"[%d %s] ",dictid,getClientPeerId(c));
    }

    for (j = 0; j < argc; j++) {
        if (argv[j]->encoding == OBJ_ENCODING_INT) {
            cmdrepr = sdscatprintf(cmdrepr, "\"%ld\"", (long)argv[j]->ptr);
        } else {
            cmdrepr = sdscatrepr(cmdrepr,(char*)argv[j]->ptr,
                        sdslen(argv[j]->ptr));
        }
        if (j != argc-1)
            cmdrepr = sdscatlen(cmdrepr," ",1);
    }
    cmdrepr = sdscatlen(cmdrepr,"\r\n",2);
    cmdobj = createObject(OBJ_STRING,cmdrepr);

    listRewind(monitors,&li);
    while((ln = listNext(&li))) {
        client *monitor = ln->value;
        /* Do not show internal commands to non-internal clients. */
        if (c->realcmd && (c->realcmd->flags & CMD_INTERNAL) && !(monitor->flags & CLIENT_INTERNAL)) {
            continue;
        }
        addReply(monitor,cmdobj);
        updateClientMemUsageAndBucket(monitor);
    }
    decrRefCount(cmdobj);
}

/* Feed the slave 'c' with the replication backlog starting from the
 * specified 'offset' up to the end of the backlog. */
long long addReplyReplicationBacklog(client *c, long long offset) {
    long long skip;

    serverLog(LL_DEBUG, "[PSYNC] Replica request offset: %lld", offset);

    if (server.repl_backlog->histlen == 0) {
        serverLog(LL_DEBUG, "[PSYNC] Backlog history len is zero");
        return 0;
    }

    serverLog(LL_DEBUG, "[PSYNC] Backlog size: %lld",
             server.repl_backlog_size);
    serverLog(LL_DEBUG, "[PSYNC] First byte: %lld",
             server.repl_backlog->offset);
    serverLog(LL_DEBUG, "[PSYNC] History len: %lld",
             server.repl_backlog->histlen);

    /* Compute the amount of bytes we need to discard. */
    skip = offset - server.repl_backlog->offset;
    serverLog(LL_DEBUG, "[PSYNC] Skipping: %lld", skip);

    /* Iterate recorded blocks, quickly search the approximate node. */
    listNode *node = NULL;
    if (raxSize(server.repl_backlog->blocks_index) > 0) {
        uint64_t encoded_offset = htonu64(offset);
        raxIterator ri;
        raxStart(&ri, server.repl_backlog->blocks_index);
        raxSeek(&ri, ">", (unsigned char*)&encoded_offset, sizeof(uint64_t));
        if (raxEOF(&ri)) {
            /* No found, so search from the last recorded node. */
            raxSeek(&ri, "$", NULL, 0);
            raxPrev(&ri);
            node = (listNode *)ri.data;
        } else {
            raxPrev(&ri); /* Skip the sought node. */
            /* We should search from the prev node since the offset of current
             * sought node exceeds searching offset. */
            if (raxPrev(&ri))
                node = (listNode *)ri.data;
            else
                node = server.repl_backlog->ref_repl_buf_node;
        }
        raxStop(&ri);
    } else {
        /* No recorded blocks, just from the start node to search. */
        node = server.repl_backlog->ref_repl_buf_node;
    }

    /* Search the exact node. */
    while (node != NULL) {
        replBufBlock *o = listNodeValue(node);
        if (o->repl_offset + (long long)o->used >= offset) break;
        node = listNextNode(node);
    }
    serverAssert(node != NULL);

    /* Install a writer handler first.*/
    prepareClientToWrite(c);
    /* Setting output buffer of the replica. */
    replBufBlock *o = listNodeValue(node);
    o->refcount++;
    c->ref_repl_buf_node = node;
    c->ref_block_pos = offset - o->repl_offset;

    return server.repl_backlog->histlen - skip;
}

/* Return the offset to provide as reply to the PSYNC command received
 * from the slave. The returned value is only valid immediately after
 * the BGSAVE process started and before executing any other command
 * from clients. */
long long getPsyncInitialOffset(void) {
    return server.master_repl_offset;
}

/* Send a FULLRESYNC reply in the specific case of a full resynchronization,
 * as a side effect setup the slave for a full sync in different ways:
 *
 * 1) Remember, into the slave client structure, the replication offset
 *    we sent here, so that if new slaves will later attach to the same
 *    background RDB saving process (by duplicating this client output
 *    buffer), we can get the right offset from this slave.
 * 2) Set the replication state of the slave to WAIT_BGSAVE_END so that
 *    we start accumulating differences from this point.
 * 3) Force the replication stream to re-emit a SELECT statement so
 *    the new slave incremental differences will start selecting the
 *    right database number.
 *
 * Normally this function should be called immediately after a successful
 * BGSAVE for replication was started, or when there is one already in
 * progress that we attached our slave to. */
int replicationSetupSlaveForFullResync(client *slave, long long offset) {
    char buf[128];
    int buflen;

    slave->psync_initial_offset = offset;
    slave->replstate = SLAVE_STATE_WAIT_BGSAVE_END;
    /* We are going to accumulate the incremental changes for this
     * slave as well. Set slaveseldb to -1 in order to force to re-emit
     * a SELECT statement in the replication stream. */
    server.slaveseldb = -1;

    /* Don't send this reply to slaves that approached us with
     * the old SYNC command. */
    if (!(slave->flags & CLIENT_PRE_PSYNC)) {
        if (slave->slave_req & SLAVE_REQ_RDB_CHANNEL) {
            /* This slave is rdbchannel. Find its associated main channel and
             * change its state so we can deliver replication stream from now
             * on, in parallel to rdb. */
            uint64_t id = slave->main_ch_client_id;
            client *c = lookupClientByID(id);
            if (c && c->replstate == SLAVE_STATE_WAIT_RDB_CHANNEL) {
                c->replstate = SLAVE_STATE_SEND_BULK_AND_STREAM;
                serverLog(LL_NOTICE, "Starting to deliver RDB and replication stream to replica: %s",
                          replicationGetSlaveName(c));
            } else {
                serverLog(LL_WARNING, "Starting to deliver RDB to replica %s"
                                      " but it has no associated main channel",
                                      replicationGetSlaveName(slave));
            }
        }
        buflen = snprintf(buf,sizeof(buf),"+FULLRESYNC %s %lld\r\n",
                          server.replid,offset);
        if (connWrite(slave->conn,buf,buflen) != buflen) {
            freeClientAsync(slave);
            return C_ERR;
        }
    }
    return C_OK;
}

/* This function handles the PSYNC command from the point of view of a
 * master receiving a request for partial resynchronization.
 *
 * On success return C_OK, otherwise C_ERR is returned and we proceed
 * with the usual full resync. */
int masterTryPartialResynchronization(client *c, long long psync_offset) {
    long long psync_len;
    char *master_replid = c->argv[1]->ptr;
    char buf[128];
    int buflen;

    /* Is the replication ID of this master the same advertised by the wannabe
     * slave via PSYNC? If the replication ID changed this master has a
     * different replication history, and there is no way to continue.
     *
     * Note that there are two potentially valid replication IDs: the ID1
     * and the ID2. The ID2 however is only valid up to a specific offset. */
    if (strcasecmp(master_replid, server.replid) &&
        (strcasecmp(master_replid, server.replid2) ||
         psync_offset > server.second_replid_offset))
    {
        /* Replid "?" is used by slaves that want to force a full resync. */
        if (master_replid[0] != '?') {
            if (strcasecmp(master_replid, server.replid) &&
                strcasecmp(master_replid, server.replid2))
            {
                serverLog(LL_NOTICE,"Partial resynchronization not accepted: "
                    "Replication ID mismatch (Replica asked for '%s', my "
                    "replication IDs are '%s' and '%s')",
                    master_replid, server.replid, server.replid2);
            } else {
                serverLog(LL_NOTICE,"Partial resynchronization not accepted: "
                    "Requested offset for second ID was %lld, but I can reply "
                    "up to %lld", psync_offset, server.second_replid_offset);
            }
        } else {
            serverLog(LL_NOTICE,"Full resync requested by replica %s %s",
                replicationGetSlaveName(c),
                c->flags & CLIENT_REPL_RDB_CHANNEL ? "(rdb-channel)" : "");
        }
        goto need_full_resync;
    }

    /* We still have the data our slave is asking for? */
    if (!server.repl_backlog ||
        psync_offset < server.repl_backlog->offset ||
        psync_offset > (server.repl_backlog->offset + server.repl_backlog->histlen))
    {
        serverLog(LL_NOTICE,
            "Unable to partial resync with replica %s for lack of backlog (Replica request was: %lld).", replicationGetSlaveName(c), psync_offset);
        if (psync_offset > server.master_repl_offset) {
            serverLog(LL_WARNING,
                "Warning: replica %s tried to PSYNC with an offset that is greater than the master replication offset.", replicationGetSlaveName(c));
        }
        goto need_full_resync;
    }

    /* If we reached this point, we are able to perform a partial resync:
     * 1) Set client state to make it a slave.
     * 2) Inform the client we can continue with +CONTINUE
     * 3) Send the backlog data (from the offset to the end) to the slave. */
    c->flags |= CLIENT_SLAVE;
    c->replstate = SLAVE_STATE_ONLINE;
    c->repl_ack_time = server.unixtime;
    c->repl_start_cmd_stream_on_ack = 0;
    listAddNodeTail(server.slaves,c);
    /* We can't use the connection buffers since they are used to accumulate
     * new commands at this stage. But we are sure the socket send buffer is
     * empty so this write will never fail actually. */
    if (c->slave_capa & SLAVE_CAPA_PSYNC2) {
        buflen = snprintf(buf,sizeof(buf),"+CONTINUE %s\r\n", server.replid);
    } else {
        buflen = snprintf(buf,sizeof(buf),"+CONTINUE\r\n");
    }
    if (connWrite(c->conn,buf,buflen) != buflen) {
        freeClientAsync(c);
        return C_OK;
    }
    psync_len = addReplyReplicationBacklog(c,psync_offset);
    serverLog(LL_NOTICE,
        "Partial resynchronization request from %s accepted. Sending %lld bytes of backlog starting from offset %lld.",
            replicationGetSlaveName(c),
            psync_len, psync_offset);
    /* Note that we don't need to set the selected DB at server.slaveseldb
     * to -1 to force the master to emit SELECT, since the slave already
     * has this state from the previous connection with the master. */

    refreshGoodSlavesCount();

    /* Fire the replica change modules event. */
    moduleFireServerEvent(REDISMODULE_EVENT_REPLICA_CHANGE,
                          REDISMODULE_SUBEVENT_REPLICA_CHANGE_ONLINE,
                          NULL);

    return C_OK; /* The caller can return, no full resync needed. */

need_full_resync:
    /* We need a full resync for some reason... Note that we can't
     * reply to PSYNC right now if a full SYNC is needed. The reply
     * must include the master offset at the time the RDB file we transfer
     * is generated, so we need to delay the reply to that moment. */
    return C_ERR;
}

/* Start a BGSAVE for replication goals, which is, selecting the disk or
 * socket target depending on the configuration, and making sure that
 * the script cache is flushed before to start.
 *
 * The mincapa argument is the bitwise AND among all the slaves capabilities
 * of the slaves waiting for this BGSAVE, so represents the slave capabilities
 * all the slaves support. Can be tested via SLAVE_CAPA_* macros.
 *
 * Side effects, other than starting a BGSAVE:
 *
 * 1) Handle the slaves in WAIT_START state, by preparing them for a full
 *    sync if the BGSAVE was successfully started, or sending them an error
 *    and dropping them from the list of slaves.
 *
 * 2) Flush the Lua scripting script cache if the BGSAVE was actually
 *    started.
 *
 * Returns C_OK on success or C_ERR otherwise. */
int startBgsaveForReplication(int mincapa, int req) {
    int retval;
    int socket_target = 0;
    listIter li;
    listNode *ln;

    /* We use a socket target if slave can handle the EOF marker and we're configured to do diskless syncs.
     * Note that in case we're creating a "filtered" RDB (functions-only, for example) we also force socket replication
     * to avoid overwriting the snapshot RDB file with filtered data. */
    socket_target = (server.repl_diskless_sync || req & SLAVE_REQ_RDB_MASK) && (mincapa & SLAVE_CAPA_EOF);
    /* `SYNC` should have failed with error if we don't support socket and require a filter, assert this here */
    serverAssert(socket_target || !(req & SLAVE_REQ_RDB_MASK));

    serverLog(LL_NOTICE,"Starting BGSAVE for SYNC with target: %s%s",
        socket_target ? "replicas sockets" : "disk",
        (req & SLAVE_REQ_RDB_CHANNEL) ? " (rdb-channel)" : "");

    rdbSaveInfo rsi, *rsiptr;
    rsiptr = rdbPopulateSaveInfo(&rsi);
    /* Only do rdbSave* when rsiptr is not NULL,
     * otherwise slave will miss repl-stream-db. */
    if (rsiptr) {
        if (socket_target)
            retval = rdbSaveToSlavesSockets(req,rsiptr);
        else {
            /* Keep the page cache since it'll get used soon */
            retval = rdbSaveBackground(req, server.rdb_filename, rsiptr, RDBFLAGS_REPLICATION | RDBFLAGS_KEEP_CACHE);
        }
        if (server.repl_debug_pause & REPL_DEBUG_AFTER_FORK)
            debugPauseProcess();
    } else {
        serverLog(LL_WARNING,"BGSAVE for replication: replication information not available, can't generate the RDB file right now. Try later.");
        retval = C_ERR;
    }

    /* If we succeeded to start a BGSAVE with disk target, let's remember
     * this fact, so that we can later delete the file if needed. Note
     * that we don't set the flag to 1 if the feature is disabled, otherwise
     * it would never be cleared: the file is not deleted. This way if
     * the user enables it later with CONFIG SET, we are fine. */
    if (retval == C_OK && !socket_target && server.rdb_del_sync_files)
        RDBGeneratedByReplication = 1;

    /* If we failed to BGSAVE, remove the slaves waiting for a full
     * resynchronization from the list of slaves, inform them with
     * an error about what happened, close the connection ASAP. */
    if (retval == C_ERR) {
        serverLog(LL_WARNING,"BGSAVE for replication failed");
        listRewind(server.slaves,&li);
        while((ln = listNext(&li))) {
            client *slave = ln->value;

            if (slave->replstate == SLAVE_STATE_WAIT_BGSAVE_START) {
                slave->replstate = REPL_STATE_NONE;
                slave->flags &= ~CLIENT_SLAVE;
                listDelNode(server.slaves,ln);
                addReplyError(slave,
                    "BGSAVE failed, replication can't continue");
                slave->flags |= CLIENT_CLOSE_AFTER_REPLY;
            }
        }
        return retval;
    }

    /* If the target is socket, rdbSaveToSlavesSockets() already setup
     * the slaves for a full resync. Otherwise for disk target do it now.*/
    if (!socket_target) {
        listRewind(server.slaves,&li);
        while((ln = listNext(&li))) {
            client *slave = ln->value;

            if (slave->replstate == SLAVE_STATE_WAIT_BGSAVE_START) {
                /* Check slave has the exact requirements */
                if (slave->slave_req != req)
                    continue;
                replicationSetupSlaveForFullResync(slave, getPsyncInitialOffset());
            }
        }
    }

    return retval;
}

/* SYNC and PSYNC command implementation. */
void syncCommand(client *c) {
    /* ignore SYNC if already slave or in monitor mode */
    if (c->flags & CLIENT_SLAVE) return;

    /* Check if this is a failover request to a replica with the same replid and
     * become a master if so. */
    if (c->argc > 3 && !strcasecmp(c->argv[0]->ptr,"psync") && 
        !strcasecmp(c->argv[3]->ptr,"failover"))
    {
        serverLog(LL_NOTICE, "Failover request received for replid %s.",
            (unsigned char *)c->argv[1]->ptr);
        if (!server.masterhost) {
            addReplyError(c, "PSYNC FAILOVER can't be sent to a master.");
            return;
        }

        if (!strcasecmp(c->argv[1]->ptr,server.replid)) {
            if (server.cluster_enabled) {
                clusterPromoteSelfToMaster();
            } else {
                replicationUnsetMaster();
            }
            sds client = catClientInfoString(sdsempty(),c);
            serverLog(LL_NOTICE,
                "MASTER MODE enabled (failover request from '%s')",client);
            sdsfree(client);
        } else {
            addReplyError(c, "PSYNC FAILOVER replid must match my replid.");
            return;            
        }
    }

    /* Don't let replicas sync with us while we're failing over */
    if (server.failover_state != NO_FAILOVER) {
        addReplyError(c,"-NOMASTERLINK Can't SYNC while failing over");
        return;
    }

    /* Refuse SYNC requests if we are a slave but the link with our master
     * is not ok... */
    if (server.masterhost && server.repl_state != REPL_STATE_CONNECTED) {
        addReplyError(c,"-NOMASTERLINK Can't SYNC while not connected with my master");
        return;
    }

    /* SYNC can't be issued when the server has pending data to send to
     * the client about already issued commands. We need a fresh reply
     * buffer registering the differences between the BGSAVE and the current
     * dataset, so that we can copy to other slaves if needed. */
    if (clientHasPendingReplies(c)) {
        addReplyError(c,"SYNC and PSYNC are invalid with pending output");
        return;
    }

    /* Fail sync if slave doesn't support EOF capability but wants a filtered RDB. This is because we force filtered
     * RDB's to be generated over a socket and not through a file to avoid conflicts with the snapshot files. Forcing
     * use of a socket is handled, if needed, in `startBgsaveForReplication`. */
    if (c->slave_req & SLAVE_REQ_RDB_MASK && !(c->slave_capa & SLAVE_CAPA_EOF)) {
        addReplyError(c,"Filtered replica requires EOF capability");
        return;
    }

    serverLog(LL_NOTICE,"Replica %s asks for synchronization",
        replicationGetSlaveName(c));

    /* Try a partial resynchronization if this is a PSYNC command.
     * If it fails, we continue with usual full resynchronization, however
     * when this happens replicationSetupSlaveForFullResync will replied
     * with:
     *
     * +FULLRESYNC <replid> <offset>
     *
     * So the slave knows the new replid and offset to try a PSYNC later
     * if the connection with the master is lost. */
    if (!strcasecmp(c->argv[0]->ptr,"psync")) {
        long long psync_offset;
        if (getLongLongFromObjectOrReply(c, c->argv[2], &psync_offset, NULL) != C_OK) {
            serverLog(LL_WARNING, "Replica %s asks for synchronization but with a wrong offset",
                      replicationGetSlaveName(c));
            return;
        }

        if (masterTryPartialResynchronization(c, psync_offset) == C_OK) {
            server.stat_sync_partial_ok++;
            return; /* No full resync needed, return. */
        } else {
            char *master_replid = c->argv[1]->ptr;

            /* Increment stats for failed PSYNCs, but only if the
             * replid is not "?", as this is used by slaves to force a full
             * resync on purpose when they are not able to partially
             * resync. */
            if (master_replid[0] != '?') server.stat_sync_partial_err++;
            if (c->slave_capa & SLAVE_CAPA_RDB_CHANNEL_REPL) {
                int len;
                char buf[128];
                /* Replica is capable of rdbchannel replication. This is
                 * replica's main channel. Let replica know full sync is needed.
                 * Replica will open another connection (rdbchannel). Once rdb
                 * delivery starts, we'll stream repl data to the main channel.*/
                c->flags |= CLIENT_SLAVE;
                c->replstate = SLAVE_STATE_WAIT_RDB_CHANNEL;
                c->repl_ack_time = server.unixtime;
                listAddNodeTail(server.slaves, c);
                createReplicationBacklogIfNeeded();

                serverLog(LL_NOTICE,
                          "Replica %s is capable of rdb channel synchronization, and partial sync isn't possible. "
                          "Full sync will continue with dedicated rdb channel.",
                          replicationGetSlaveName(c));

                /* Send +RDBCHANNELSYNC with client id. Rdbchannel of replica
                 * will call 'replconf set-main-ch-id <client-id>' so we can
                 * associate replica connections on master.*/
                len = snprintf(buf, sizeof(buf), "+RDBCHANNELSYNC %llu\r\n",
                               (unsigned long long) c->id);
                if (connWrite(c->conn, buf, strlen(buf)) != len)
                    freeClientAsync(c);

                return;
            }
        }
    } else {
        /* If a slave uses SYNC, we are dealing with an old implementation
         * of the replication protocol (like redis-cli --slave). Flag the client
         * so that we don't expect to receive REPLCONF ACK feedbacks. */
        c->flags |= CLIENT_PRE_PSYNC;
    }

    /* Full resynchronization. */
    server.stat_sync_full++;

    /* Setup the slave as one waiting for BGSAVE to start. The following code
     * paths will change the state if we handle the slave differently. */
    c->replstate = SLAVE_STATE_WAIT_BGSAVE_START;
    if (server.repl_disable_tcp_nodelay)
        connDisableTcpNoDelay(c->conn); /* Non critical if it fails. */
    c->repldbfd = -1;
    c->flags |= CLIENT_SLAVE;
    listAddNodeTail(server.slaves,c);

    /* Create the replication backlog if needed. */
    createReplicationBacklogIfNeeded();

    /* CASE 1: BGSAVE is in progress, with disk target. */
    if (server.child_type == CHILD_TYPE_RDB &&
        server.rdb_child_type == RDB_CHILD_TYPE_DISK)
    {
        /* Ok a background save is in progress. Let's check if it is a good
         * one for replication, i.e. if there is another slave that is
         * registering differences since the server forked to save. */
        client *slave;
        listNode *ln;
        listIter li;

        listRewind(server.slaves,&li);
        while((ln = listNext(&li))) {
            slave = ln->value;
            /* If the client needs a buffer of commands, we can't use
             * a replica without replication buffer. */
            if (slave->replstate == SLAVE_STATE_WAIT_BGSAVE_END &&
                (!(slave->flags & CLIENT_REPL_RDBONLY) ||
                 (c->flags & CLIENT_REPL_RDBONLY)))
                break;
        }
        /* To attach this slave, we check that it has at least all the
         * capabilities of the slave that triggered the current BGSAVE
         * and its exact requirements. */
        if (ln && ((c->slave_capa & slave->slave_capa) == slave->slave_capa) &&
            c->slave_req == slave->slave_req) {
            /* Perfect, the server is already registering differences for
             * another slave. Set the right state, and copy the buffer.
             * We don't copy buffer if clients don't want. */
            if (!(c->flags & CLIENT_REPL_RDBONLY))
                copyReplicaOutputBuffer(c,slave);
            replicationSetupSlaveForFullResync(c,slave->psync_initial_offset);
            serverLog(LL_NOTICE,"Waiting for end of BGSAVE for SYNC");
        } else {
            /* No way, we need to wait for the next BGSAVE in order to
             * register differences. */
            serverLog(LL_NOTICE,"Can't attach the replica to the current BGSAVE. Waiting for next BGSAVE for SYNC");
        }

    /* CASE 2: BGSAVE is in progress, with socket target. */
    } else if (server.child_type == CHILD_TYPE_RDB &&
               server.rdb_child_type == RDB_CHILD_TYPE_SOCKET)
    {
        /* There is an RDB child process but it is writing directly to
         * children sockets. We need to wait for the next BGSAVE
         * in order to synchronize. */
        serverLog(LL_NOTICE,"Current BGSAVE has socket target. Waiting for next BGSAVE for SYNC");

    /* CASE 3: There is no BGSAVE is in progress. */
    } else {
        if (server.repl_diskless_sync && (c->slave_capa & SLAVE_CAPA_EOF) &&
            server.repl_diskless_sync_delay)
        {
            /* Diskless replication RDB child is created inside
             * replicationCron() since we want to delay its start a
             * few seconds to wait for more slaves to arrive. */
            serverLog(LL_NOTICE,"Delay next BGSAVE for diskless SYNC");
        } else {
            /* We don't have a BGSAVE in progress, let's start one. Diskless
             * or disk-based mode is determined by replica's capacity. */
            if (!hasActiveChildProcess()) {
                startBgsaveForReplication(c->slave_capa, c->slave_req);
            } else {
                serverLog(LL_NOTICE,
                    "No BGSAVE in progress, but another BG operation is active. "
                    "BGSAVE for replication delayed");
            }
        }
    }
    return;
}

/* REPLCONF <option> <value> <option> <value> ...
 * This command is used by a replica in order to configure the replication
 * process before starting it with the SYNC command.
 * This command is also used by a master in order to get the replication
 * offset from a replica.
 *
 * Currently we support these options:
 *
 * - listening-port <port>
 * - ip-address <ip>
 * What is the listening ip and port of the Replica redis instance, so that
 * the master can accurately lists replicas and their listening ports in the
 * INFO output.
 *
 * - capa <eof|psync2|rdb-channel-repl>
 * What is the capabilities of this instance.
 * eof: supports EOF-style RDB transfer for diskless replication.
 * psync2: supports PSYNC v2, so understands +CONTINUE <new repl ID>.
 *
 * - ack <offset> [fack <aofofs>]
 * Replica informs the master the amount of replication stream that it
 * processed so far, and optionally the replication offset fsynced to the AOF file.
 * This special pattern doesn't reply to the caller.
 *
 * - getack <dummy>
 * Unlike other subcommands, this is used by master to get the replication
 * offset from a replica.
 *
 * - rdb-only <0|1>
 * Only wants RDB snapshot without replication buffer.
 *
 * - rdb-filter-only <include-filters>
 * Define "include" filters for the RDB snapshot. Currently we only support
 * a single include filter: "functions". Passing an empty string "" will
 * result in an empty RDB.
 *
 * - main-ch-client-id <client-id>
 * Replica's main channel informs master that this is the main channel of the
 * rdb channel identified by the client-id. */
void replconfCommand(client *c) {
    int j;

    if ((c->argc % 2) == 0) {
        /* Number of arguments must be odd to make sure that every
         * option has a corresponding value. */
        addReplyErrorObject(c,shared.syntaxerr);
        return;
    }

    /* Process every option-value pair. */
    for (j = 1; j < c->argc; j+=2) {
        if (!strcasecmp(c->argv[j]->ptr,"listening-port")) {
            long port;

            if ((getLongFromObjectOrReply(c,c->argv[j+1],
                    &port,NULL) != C_OK))
                return;
            c->slave_listening_port = port;
        } else if (!strcasecmp(c->argv[j]->ptr,"ip-address")) {
            sds addr = c->argv[j+1]->ptr;
            if (sdslen(addr) < NET_HOST_STR_LEN) {
                if (c->slave_addr) sdsfree(c->slave_addr);
                c->slave_addr = sdsdup(addr);
            } else {
                addReplyErrorFormat(c,"REPLCONF ip-address provided by "
                    "replica instance is too long: %zd bytes", sdslen(addr));
                return;
            }
        } else if (!strcasecmp(c->argv[j]->ptr,"capa")) {
            /* Ignore capabilities not understood by this master. */
            if (!strcasecmp(c->argv[j+1]->ptr,"eof"))
                c->slave_capa |= SLAVE_CAPA_EOF;
            else if (!strcasecmp(c->argv[j+1]->ptr,"psync2"))
                c->slave_capa |= SLAVE_CAPA_PSYNC2;
            else if (!strcasecmp(c->argv[j+1]->ptr,"rdb-channel-repl") && server.repl_rdb_channel &&
                     server.repl_diskless_sync) {
                c->slave_capa |= SLAVE_CAPA_RDB_CHANNEL_REPL;
            }
        } else if (!strcasecmp(c->argv[j]->ptr,"ack")) {
            /* REPLCONF ACK is used by slave to inform the master the amount
             * of replication stream that it processed so far. It is an
             * internal only command that normal clients should never use. */
            long long offset;

            if (!(c->flags & CLIENT_SLAVE)) return;
            if ((getLongLongFromObject(c->argv[j+1], &offset) != C_OK))
                return;
            if (offset > c->repl_ack_off)
                c->repl_ack_off = offset;
            if (c->argc > j+3 && !strcasecmp(c->argv[j+2]->ptr,"fack")) {
                if ((getLongLongFromObject(c->argv[j+3], &offset) != C_OK))
                    return;
                if (offset > c->repl_aof_off)
                    c->repl_aof_off = offset;
            }
            c->repl_ack_time = server.unixtime;
            /* If this was a diskless replication, we need to really put
             * the slave online when the first ACK is received (which
             * confirms slave is online and ready to get more data). This
             * allows for simpler and less CPU intensive EOF detection
             * when streaming RDB files.
             * There's a chance the ACK got to us before we detected that the
             * bgsave is done (since that depends on cron ticks), so run a
             * quick check first (instead of waiting for the next ACK. */
            if (server.child_type == CHILD_TYPE_RDB && c->replstate == SLAVE_STATE_WAIT_BGSAVE_END)
                checkChildrenDone();
            if (c->repl_start_cmd_stream_on_ack && c->replstate == SLAVE_STATE_ONLINE)
                replicaStartCommandStream(c);
            if (c->replstate == SLAVE_STATE_SEND_BULK_AND_STREAM)
                replicaPutOnline(c);
            /* Note: this command does not reply anything! */
            return;
        } else if (!strcasecmp(c->argv[j]->ptr,"getack")) {
            /* REPLCONF GETACK is used in order to request an ACK ASAP
             * to the slave. */
            if (server.masterhost && server.master) replicationSendAck();
            return;
        } else if (!strcasecmp(c->argv[j]->ptr,"rdb-only")) {
           /* REPLCONF RDB-ONLY is used to identify the client only wants
            * RDB snapshot without replication buffer. */
            long rdb_only = 0;
            if (getRangeLongFromObjectOrReply(c,c->argv[j+1],
                    0,1,&rdb_only,NULL) != C_OK)
                return;
            if (rdb_only == 1) c->flags |= CLIENT_REPL_RDBONLY;
            else c->flags &= ~CLIENT_REPL_RDBONLY;
        } else if (!strcasecmp(c->argv[j]->ptr,"rdb-filter-only")) {
            /* REPLCONFG RDB-FILTER-ONLY is used to define "include" filters
             * for the RDB snapshot. Currently we only support a single
             * include filter: "functions". In the future we may want to add
             * other filters like key patterns, key types, non-volatile, module
             * aux fields, ...
             * We might want to add the complementing "RDB-FILTER-EXCLUDE" to
             * filter out certain data. */
            int filter_count, i;
            sds *filters;
            if (!(filters = sdssplitargs(c->argv[j+1]->ptr, &filter_count))) {
                addReplyError(c, "Missing rdb-filter-only values");
                return;
            }
            /* By default filter out all parts of the rdb */
            c->slave_req |= SLAVE_REQ_RDB_EXCLUDE_DATA;
            c->slave_req |= SLAVE_REQ_RDB_EXCLUDE_FUNCTIONS;
            for (i = 0; i < filter_count; i++) {
                if (!strcasecmp(filters[i], "functions"))
                    c->slave_req &= ~SLAVE_REQ_RDB_EXCLUDE_FUNCTIONS;
                else {
                    addReplyErrorFormat(c, "Unsupported rdb-filter-only option: %s", (char*)filters[i]);
                    sdsfreesplitres(filters, filter_count);
                    return;
                }
            }
            sdsfreesplitres(filters, filter_count);
        } else if (!strcasecmp(c->argv[j]->ptr, "rdb-channel")) {
            long rdb_channel = 0;
            if (getRangeLongFromObjectOrReply(c, c->argv[j + 1], 0, 1, &rdb_channel, NULL) != C_OK)
                return;
            if (rdb_channel == 1) {
                c->flags |= CLIENT_REPL_RDB_CHANNEL;
                c->slave_req |= SLAVE_REQ_RDB_CHANNEL;
            } else {
                c->flags &= ~CLIENT_REPL_RDB_CHANNEL;
                c->slave_req &= ~SLAVE_REQ_RDB_CHANNEL;
            }
        } else if (!strcasecmp(c->argv[j]->ptr, "main-ch-client-id")) {
            /* REPLCONF main-ch-client-id <client-id> is used to identify
             * the current replica rdb channel with existing main channel
             * connection. */
            long long client_id = 0;
            client *main_ch;
            if (getLongLongFromObjectOrReply(c, c->argv[j + 1], &client_id, NULL) != C_OK)
                return;
            main_ch = lookupClientByID(client_id);
            if (!main_ch || main_ch->replstate != SLAVE_STATE_WAIT_RDB_CHANNEL) {
                addReplyErrorFormat(c, "Unrecognized RDB client id: %lld", client_id);
                return;
            }
            c->main_ch_client_id = (uint64_t)client_id;
        } else {
            addReplyErrorFormat(c,"Unrecognized REPLCONF option: %s",
                (char*)c->argv[j]->ptr);
            return;
        }
    }
    addReply(c,shared.ok);
}

/* This function puts a replica in the online state, and should be called just
 * after a replica received the RDB file for the initial synchronization.
 *
 * It does a few things:
 * 1) Put the slave in ONLINE state.
 * 2) Update the count of "good replicas".
 * 3) Trigger the module event.
 *
 * the return value indicates that the replica should be disconnected.
 * */
int replicaPutOnline(client *slave) {
    if (slave->flags & CLIENT_REPL_RDBONLY) {
        slave->replstate = SLAVE_STATE_RDB_TRANSMITTED;
        /* The client asked for RDB only so we should close it ASAP */
        serverLog(LL_NOTICE,
                  "RDB transfer completed, rdb only replica (%s) should be disconnected asap",
                  replicationGetSlaveName(slave));
        return 0;
    }
    slave->replstate = SLAVE_STATE_ONLINE;
    slave->repl_ack_time = server.unixtime; /* Prevent false timeout. */

    refreshGoodSlavesCount();
    /* Fire the replica change modules event. */
    moduleFireServerEvent(REDISMODULE_EVENT_REPLICA_CHANGE,
                          REDISMODULE_SUBEVENT_REPLICA_CHANGE_ONLINE,
                          NULL);
    serverLog(LL_NOTICE,"Synchronization with replica %s succeeded",
        replicationGetSlaveName(slave));
    return 1;
}

/* This function should be called just after a replica received the RDB file
 * for the initial synchronization, and we are finally ready to send the
 * incremental stream of commands.
 *
 * It does a few things:
 * 1) Close the replica's connection async if it doesn't need replication
 *    commands buffer stream, since it actually isn't a valid replica.
 * 2) Make sure the writable event is re-installed, since when calling the SYNC
 *    command we had no replies and it was disabled, and then we could
 *    accumulate output buffer data without sending it to the replica so it
 *    won't get mixed with the RDB stream. */
void replicaStartCommandStream(client *slave) {
    serverAssert(!(slave->flags & CLIENT_REPL_RDBONLY));
    slave->repl_start_cmd_stream_on_ack = 0;

    putClientInPendingWriteQueue(slave);
}

/* We call this function periodically to remove an RDB file that was
 * generated because of replication, in an instance that is otherwise
 * without any persistence. We don't want instances without persistence
 * to take RDB files around, this violates certain policies in certain
 * environments. */
void removeRDBUsedToSyncReplicas(void) {
    /* If the feature is disabled, return ASAP but also clear the
     * RDBGeneratedByReplication flag in case it was set. Otherwise if the
     * feature was enabled, but gets disabled later with CONFIG SET, the
     * flag may remain set to one: then next time the feature is re-enabled
     * via CONFIG SET we have it set even if no RDB was generated
     * because of replication recently. */
    if (!server.rdb_del_sync_files) {
        RDBGeneratedByReplication = 0;
        return;
    }

    if (allPersistenceDisabled() && RDBGeneratedByReplication) {
        client *slave;
        listNode *ln;
        listIter li;

        int delrdb = 1;
        listRewind(server.slaves,&li);
        while((ln = listNext(&li))) {
            slave = ln->value;
            if (slave->replstate == SLAVE_STATE_WAIT_BGSAVE_START ||
                slave->replstate == SLAVE_STATE_WAIT_BGSAVE_END ||
                slave->replstate == SLAVE_STATE_SEND_BULK)
            {
                delrdb = 0;
                break; /* No need to check the other replicas. */
            }
        }
        if (delrdb) {
            struct stat sb;
            if (lstat(server.rdb_filename,&sb) != -1) {
                RDBGeneratedByReplication = 0;
                serverLog(LL_NOTICE,
                    "Removing the RDB file used to feed replicas "
                    "in a persistence-less instance");
                bg_unlink(server.rdb_filename);
            }
        }
    }
}

/* Close the repldbfd and reclaim the page cache if the client hold
 * the last reference to replication DB */
void closeRepldbfd(client *myself) {
    listNode *ln;
    listIter li;
    int reclaim = 1;
    listRewind(server.slaves,&li);
    while((ln = listNext(&li))) {
        client *slave = ln->value;
        if (slave != myself && slave->replstate == SLAVE_STATE_SEND_BULK) {
            reclaim = 0;
            break;
        }
    }

    if (reclaim) {
        bioCreateCloseJob(myself->repldbfd, 0, 1);
    } else {
        close(myself->repldbfd);
    }
    myself->repldbfd = -1;
}

void sendBulkToSlave(connection *conn) {
    client *slave = connGetPrivateData(conn);
    char buf[PROTO_IOBUF_LEN];
    ssize_t nwritten, buflen;

    /* Before sending the RDB file, we send the preamble as configured by the
     * replication process. Currently the preamble is just the bulk count of
     * the file in the form "$<length>\r\n". */
    if (slave->replpreamble) {
        nwritten = connWrite(conn,slave->replpreamble,sdslen(slave->replpreamble));
        if (nwritten == -1) {
            serverLog(LL_WARNING,
                "Write error sending RDB preamble to replica: %s",
                connGetLastError(conn));
            freeClient(slave);
            return;
        }
        atomicIncr(server.stat_net_repl_output_bytes, nwritten);
        sdsrange(slave->replpreamble,nwritten,-1);
        if (sdslen(slave->replpreamble) == 0) {
            sdsfree(slave->replpreamble);
            slave->replpreamble = NULL;
            /* fall through sending data. */
        } else {
            return;
        }
    }

    /* If the preamble was already transferred, send the RDB bulk data. */
    lseek(slave->repldbfd,slave->repldboff,SEEK_SET);
    buflen = read(slave->repldbfd,buf,PROTO_IOBUF_LEN);
    if (buflen <= 0) {
        serverLog(LL_WARNING,"Read error sending DB to replica: %s",
            (buflen == 0) ? "premature EOF" : strerror(errno));
        freeClient(slave);
        return;
    }
    if ((nwritten = connWrite(conn,buf,buflen)) == -1) {
        if (connGetState(conn) != CONN_STATE_CONNECTED) {
            serverLog(LL_WARNING,"Write error sending DB to replica: %s",
                connGetLastError(conn));
            freeClient(slave);
        }
        return;
    }
    slave->repldboff += nwritten;
    atomicIncr(server.stat_net_repl_output_bytes, nwritten);
    if (slave->repldboff == slave->repldbsize) {
        closeRepldbfd(slave);
        connSetWriteHandler(slave->conn,NULL);
        if (!replicaPutOnline(slave)) {
            freeClient(slave);
            return;
        }
        replicaStartCommandStream(slave);
    }
}

/* Remove one write handler from the list of connections waiting to be writable
 * during rdb pipe transfer. */
void rdbPipeWriteHandlerConnRemoved(struct connection *conn) {
    if (!connHasWriteHandler(conn))
        return;
    connSetWriteHandler(conn, NULL);
    client *slave = connGetPrivateData(conn);
    slave->repl_last_partial_write = 0;
    server.rdb_pipe_numconns_writing--;
    /* if there are no more writes for now for this conn, or write error: */
    if (server.rdb_pipe_numconns_writing == 0) {
        if (aeCreateFileEvent(server.el, server.rdb_pipe_read, AE_READABLE, rdbPipeReadHandler,NULL) == AE_ERR) {
            serverPanic("Unrecoverable error creating server.rdb_pipe_read file event.");
        }
    }
}

/* Called in diskless master during transfer of data from the rdb pipe, when
 * the replica becomes writable again. */
void rdbPipeWriteHandler(struct connection *conn) {
    serverAssert(server.rdb_pipe_bufflen>0);
    client *slave = connGetPrivateData(conn);
    ssize_t nwritten;
    if ((nwritten = connWrite(conn, server.rdb_pipe_buff + slave->repldboff,
                              server.rdb_pipe_bufflen - slave->repldboff)) == -1)
    {
        if (connGetState(conn) == CONN_STATE_CONNECTED)
            return; /* equivalent to EAGAIN */
        serverLog(LL_WARNING,"Write error sending DB to replica: %s",
            connGetLastError(conn));
        freeClient(slave);
        return;
    } else {
        slave->repldboff += nwritten;
        atomicIncr(server.stat_net_repl_output_bytes, nwritten);
        if (slave->repldboff < server.rdb_pipe_bufflen) {
            slave->repl_last_partial_write = server.unixtime;
            return; /* more data to write.. */
        }
    }
    rdbPipeWriteHandlerConnRemoved(conn);
}

/* Called in diskless master, when there's data to read from the child's rdb pipe */
void rdbPipeReadHandler(struct aeEventLoop *eventLoop, int fd, void *clientData, int mask) {
    UNUSED(mask);
    UNUSED(clientData);
    UNUSED(eventLoop);
    int i;
    if (!server.rdb_pipe_buff)
        server.rdb_pipe_buff = zmalloc(PROTO_IOBUF_LEN);
    serverAssert(server.rdb_pipe_numconns_writing==0);

    while (1) {
        server.rdb_pipe_bufflen = read(fd, server.rdb_pipe_buff, PROTO_IOBUF_LEN);
        if (server.rdb_pipe_bufflen < 0) {
            if (errno == EAGAIN || errno == EWOULDBLOCK)
                return;
            serverLog(LL_WARNING,"Diskless rdb transfer, read error sending DB to replicas: %s", strerror(errno));
            for (i=0; i < server.rdb_pipe_numconns; i++) {
                connection *conn = server.rdb_pipe_conns[i];
                if (!conn)
                    continue;
                client *slave = connGetPrivateData(conn);
                freeClient(slave);
                server.rdb_pipe_conns[i] = NULL;
            }
            killRDBChild();
            return;
        }

        if (server.rdb_pipe_bufflen == 0) {
            /* EOF - write end was closed. */
            int stillUp = 0;
            aeDeleteFileEvent(server.el, server.rdb_pipe_read, AE_READABLE);
            for (i=0; i < server.rdb_pipe_numconns; i++)
            {
                connection *conn = server.rdb_pipe_conns[i];
                if (!conn)
                    continue;
                stillUp++;
            }
            serverLog(LL_NOTICE,"Diskless rdb transfer, done reading from pipe, %d replicas still up.", stillUp);
            /* Now that the replicas have finished reading, notify the child that it's safe to exit. 
             * When the server detects the child has exited, it can mark the replica as online, and
             * start streaming the replication buffers. */
            close(server.rdb_child_exit_pipe);
            server.rdb_child_exit_pipe = -1;
            return;
        }

        int stillAlive = 0;
        for (i=0; i < server.rdb_pipe_numconns; i++)
        {
            ssize_t nwritten;
            connection *conn = server.rdb_pipe_conns[i];
            if (!conn)
                continue;

            client *slave = connGetPrivateData(conn);
            if ((nwritten = connWrite(conn, server.rdb_pipe_buff, server.rdb_pipe_bufflen)) == -1) {
                if (connGetState(conn) != CONN_STATE_CONNECTED) {
                    serverLog(LL_WARNING,"Diskless rdb transfer, write error sending DB to replica: %s",
                        connGetLastError(conn));
                    freeClient(slave);
                    server.rdb_pipe_conns[i] = NULL;
                    continue;
                }
                /* An error and still in connected state, is equivalent to EAGAIN */
                slave->repldboff = 0;
            } else {
                /* Note: when use diskless replication, 'repldboff' is the offset
                 * of 'rdb_pipe_buff' sent rather than the offset of entire RDB. */
                slave->repldboff = nwritten;
                atomicIncr(server.stat_net_repl_output_bytes, nwritten);
            }
            /* If we were unable to write all the data to one of the replicas,
             * setup write handler (and disable pipe read handler, below) */
            if (nwritten != server.rdb_pipe_bufflen) {
                slave->repl_last_partial_write = server.unixtime;
                server.rdb_pipe_numconns_writing++;
                connSetWriteHandler(conn, rdbPipeWriteHandler);
            }
            stillAlive++;
        }

        if (stillAlive == 0) {
            serverLog(LL_WARNING,"Diskless rdb transfer, last replica dropped, killing fork child.");
            /* Avoid deleting events after killRDBChild as it may trigger new bgsaves for other replicas. */
            aeDeleteFileEvent(server.el, server.rdb_pipe_read, AE_READABLE); 
            killRDBChild();
            break;
        }
        /* Remove the pipe read handler if at least one write handler was set. */
        else if (server.rdb_pipe_numconns_writing) {
            aeDeleteFileEvent(server.el, server.rdb_pipe_read, AE_READABLE);
            break;
        }
    }
}

/* This function is called at the end of every background saving.
 *
 * The argument bgsaveerr is C_OK if the background saving succeeded
 * otherwise C_ERR is passed to the function.
 * The 'type' argument is the type of the child that terminated
 * (if it had a disk or socket target). */
void updateSlavesWaitingBgsave(int bgsaveerr, int type) {
    listNode *ln;
    listIter li;

    /* Note: there's a chance we got here from within the REPLCONF ACK command
     * so we must avoid using freeClient, otherwise we'll crash on our way up. */

    listRewind(server.slaves,&li);
    while((ln = listNext(&li))) {
        client *slave = ln->value;

        /* We can get here via freeClient()->killRDBChild()->checkChildrenDone(). skip disconnected slaves. */
        if (!slave->conn) continue;

        if (slave->replstate == SLAVE_STATE_WAIT_BGSAVE_END) {
            struct redis_stat buf;

            if (bgsaveerr != C_OK) {
                freeClientAsync(slave);
                serverLog(LL_WARNING,"SYNC failed. BGSAVE child returned an error");
                continue;
            }

            /* If this was an RDB on disk save, we have to prepare to send
             * the RDB from disk to the slave socket. Otherwise if this was
             * already an RDB -> Slaves socket transfer, used in the case of
             * diskless replication, our work is trivial, we can just put
             * the slave online. */
            if (type == RDB_CHILD_TYPE_SOCKET) {
                serverLog(LL_NOTICE,
                    "Streamed RDB transfer with replica %s succeeded (socket). Waiting for REPLCONF ACK from replica to enable streaming",
                        replicationGetSlaveName(slave));
                /* Note: we wait for a REPLCONF ACK message from the replica in
                 * order to really put it online (install the write handler
                 * so that the accumulated data can be transferred). However
                 * we change the replication state ASAP, since our slave
                 * is technically online now.
                 *
                 * So things work like that:
                 *
                 * 1. We end transferring the RDB file via socket.
                 * 2. The replica is put ONLINE but the write handler
                 *    is not installed.
                 * 3. The replica however goes really online, and pings us
                 *    back via REPLCONF ACK commands.
                 * 4. Now we finally install the write handler, and send
                 *    the buffers accumulated so far to the replica.
                 *
                 * But why we do that? Because the replica, when we stream
                 * the RDB directly via the socket, must detect the RDB
                 * EOF (end of file), that is a special random string at the
                 * end of the RDB (for streamed RDBs we don't know the length
                 * in advance). Detecting such final EOF string is much
                 * simpler and less CPU intensive if no more data is sent
                 * after such final EOF. So we don't want to glue the end of
                 * the RDB transfer with the start of the other replication
                 * data. */
                if (!replicaPutOnline(slave)) {
                    freeClientAsync(slave);
                    continue;
                }
                slave->repl_start_cmd_stream_on_ack = 1;
            } else {
                if ((slave->repldbfd = open(server.rdb_filename,O_RDONLY)) == -1 ||
                    redis_fstat(slave->repldbfd,&buf) == -1) {
                    freeClientAsync(slave);
                    serverLog(LL_WARNING,"SYNC failed. Can't open/stat DB after BGSAVE: %s", strerror(errno));
                    continue;
                }
                slave->repldboff = 0;
                slave->repldbsize = buf.st_size;
                slave->replstate = SLAVE_STATE_SEND_BULK;
                slave->replpreamble = sdscatprintf(sdsempty(),"$%lld\r\n",
                    (unsigned long long) slave->repldbsize);

                connSetWriteHandler(slave->conn,NULL);
                if (connSetWriteHandler(slave->conn,sendBulkToSlave) == C_ERR) {
                    freeClientAsync(slave);
                    continue;
                }
            }
        }
    }
}

/* Change the current instance replication ID with a new, random one.
 * This will prevent successful PSYNCs between this master and other
 * slaves, so the command should be called when something happens that
 * alters the current story of the dataset. */
void changeReplicationId(void) {
    getRandomHexChars(server.replid,CONFIG_RUN_ID_SIZE);
    server.replid[CONFIG_RUN_ID_SIZE] = '\0';
}

/* Clear (invalidate) the secondary replication ID. This happens, for
 * example, after a full resynchronization, when we start a new replication
 * history. */
void clearReplicationId2(void) {
    memset(server.replid2,'0',sizeof(server.replid));
    server.replid2[CONFIG_RUN_ID_SIZE] = '\0';
    server.second_replid_offset = -1;
}

/* Use the current replication ID / offset as secondary replication
 * ID, and change the current one in order to start a new history.
 * This should be used when an instance is switched from slave to master
 * so that it can serve PSYNC requests performed using the master
 * replication ID. */
void shiftReplicationId(void) {
    memcpy(server.replid2,server.replid,sizeof(server.replid));
    /* We set the second replid offset to the master offset + 1, since
     * the slave will ask for the first byte it has not yet received, so
     * we need to add one to the offset: for example if, as a slave, we are
     * sure we have the same history as the master for 50 bytes, after we
     * are turned into a master, we can accept a PSYNC request with offset
     * 51, since the slave asking has the same history up to the 50th
     * byte, and is asking for the new bytes starting at offset 51. */
    server.second_replid_offset = server.master_repl_offset+1;
    changeReplicationId();
    serverLog(LL_NOTICE,"Setting secondary replication ID to %s, valid up to offset: %lld. New replication ID is %s", server.replid2, server.second_replid_offset, server.replid);
}

/* ----------------------------------- SLAVE -------------------------------- */

/* Replication: Replica side. */
void slaveGetPortStr(char *buf, size_t size) {
    long long port;
    if (server.slave_announce_port) {
        port = server.slave_announce_port;
    } else if (server.tls_replication && server.tls_port) {
        port = server.tls_port;
    } else {
        port = server.port;
    }
    ll2string(buf, size, port);
}

/* Returns 1 if the given replication state is a handshake state,
 * 0 otherwise. */
int slaveIsInHandshakeState(void) {
    return server.repl_state >= REPL_STATE_RECEIVE_PING_REPLY &&
           server.repl_state <= REPL_STATE_RECEIVE_PSYNC_REPLY;
}

/* Avoid the master to detect the slave is timing out while loading the
 * RDB file in initial synchronization. We send a single newline character
 * that is valid protocol but is guaranteed to either be sent entirely or
 * not, since the byte is indivisible.
 *
 * The function is called in two contexts: while we flush the current
 * data with emptyData(), and while we load the new data received as an
 * RDB file from the master. */
void replicationSendNewlineToMaster(void) {
    static time_t newline_sent;
    if (time(NULL) != newline_sent) {
        newline_sent = time(NULL);
        /* Pinging back in this stage is best-effort. */
        if (server.repl_transfer_s) connWrite(server.repl_transfer_s, "\n", 1);
    }
}

/* Callback used by emptyData() while flushing away old data to load
 * the new dataset received by the master or to clear partial db if loading
 * fails. */
void replicationEmptyDbCallback(dict *d) {
    UNUSED(d);
    if (server.repl_state == REPL_STATE_TRANSFER)
        replicationSendNewlineToMaster();

    processEventsWhileBlocked();
}

/* Function to flush old db or the partial db on error. */
static void rdbLoadEmptyDbFunc(void) {
    serverAssert(server.loading);

    serverLog(LL_NOTICE, "MASTER <-> REPLICA sync: Flushing old data");
    int empty_db_flags = server.repl_slave_lazy_flush ? EMPTYDB_ASYNC :
                                                        EMPTYDB_NO_FLAGS;
    emptyData(-1, empty_db_flags, replicationEmptyDbCallback);
}

/* Once we have a link with the master and the synchronization was
 * performed, this function materializes the master client we store
 * at server.master, starting from the specified file descriptor. */
void replicationCreateMasterClient(connection *conn, int dbid) {
    server.master = createClient(conn);
    if (conn)
        connSetReadHandler(server.master->conn, readQueryFromClient);

    /**
     * Important note:
     * The CLIENT_DENY_BLOCKING flag is not, and should not, be set here.
     * For commands like BLPOP, it makes no sense to block the master
     * connection, and such blocking attempt will probably cause deadlock and
     * break the replication. We consider such a thing as a bug because
     * commands as BLPOP should never be sent on the replication link.
     * A possible use-case for blocking the replication link is if a module wants
     * to pass the execution to a background thread and unblock after the
     * execution is done. This is the reason why we allow blocking the replication
     * connection. */
    server.master->flags |= CLIENT_MASTER;

    /* Allocate a private query buffer for the master client instead of using the reusable query buffer.
     * This is done because the master's query buffer data needs to be preserved for my sub-replicas to use. */
    server.master->querybuf = sdsempty();
    server.master->authenticated = 1;
    server.master->reploff = server.master_initial_offset;
    server.master->read_reploff = server.master->reploff;
    server.master->user = NULL; /* This client can do everything. */
    memcpy(server.master->replid, server.master_replid,
        sizeof(server.master_replid));
    /* If master offset is set to -1, this master is old and is not
     * PSYNC capable, so we flag it accordingly. */
    if (server.master->reploff == -1)
        server.master->flags |= CLIENT_PRE_PSYNC;
    if (dbid != -1) selectDb(server.master,dbid);
}

static int useDisklessLoad(void) {
    /* compute boolean decision to use diskless load */
    int enabled = server.repl_diskless_load == REPL_DISKLESS_LOAD_SWAPDB ||
           (server.repl_diskless_load == REPL_DISKLESS_LOAD_WHEN_DB_EMPTY && dbTotalServerKeyCount()==0);

    if (enabled) {
        /* Check all modules handle read errors, otherwise it's not safe to use diskless load. */
        if (!moduleAllDatatypesHandleErrors()) {
            serverLog(LL_NOTICE,
                "Skipping diskless-load because there are modules that don't handle read errors.");
            enabled = 0;
        }
        /* Check all modules handle async replication, otherwise it's not safe to use diskless load. */
        else if (server.repl_diskless_load == REPL_DISKLESS_LOAD_SWAPDB && !moduleAllModulesHandleReplAsyncLoad()) {
            serverLog(LL_NOTICE,
                "Skipping diskless-load because there are modules that are not aware of async replication.");
            enabled = 0;
        }
    }
    return enabled;
}

/* Helper function for readSyncBulkPayload() to initialize tempDb
 * before socket-loading the new db from master. The tempDb may be populated
 * by swapMainDbWithTempDb or freed by disklessLoadDiscardTempDb later. */
redisDb *disklessLoadInitTempDb(void) {
    return initTempDb();
}

/* Helper function for readSyncBulkPayload() to discard our tempDb
 * when the loading succeeded or failed. */
void disklessLoadDiscardTempDb(redisDb *tempDb) {
    discardTempDb(tempDb);
}

/* If we know we got an entirely different data set from our master
 * we have no way to incrementally feed our replicas after that.
 * We want our replicas to resync with us as well, if we have any sub-replicas.
 * This is useful on readSyncBulkPayload in places where we just finished transferring db. */
void replicationAttachToNewMaster(void) { 
    /* Replica starts to apply data from new master, we must discard the cached
     * master structure. */
    serverAssert(server.master == NULL);
    replicationDiscardCachedMaster();

    disconnectSlaves(); /* Force our replicas to resync with us as well. */
    freeReplicationBacklog(); /* Don't allow our chained replicas to PSYNC. */
}

/* Asynchronously read the SYNC payload we receive from a master */
#define REPL_MAX_WRITTEN_BEFORE_FSYNC (1024*1024*8) /* 8 MB */
void readSyncBulkPayload(connection *conn) {
    char buf[PROTO_IOBUF_LEN];
    ssize_t nread, readlen, nwritten;
    int use_diskless_load = useDisklessLoad();
    redisDb *diskless_load_tempDb = NULL;
    functionsLibCtx* temp_functions_lib_ctx = NULL;
    int rdbchannel = (conn == server.repl_rdb_transfer_s);
    int empty_db_flags = server.repl_slave_lazy_flush ? EMPTYDB_ASYNC :
                                                        EMPTYDB_NO_FLAGS;
    off_t left;

    /* Static vars used to hold the EOF mark, and the last bytes received
     * from the server: when they match, we reached the end of the transfer. */
    static char eofmark[CONFIG_RUN_ID_SIZE];
    static char lastbytes[CONFIG_RUN_ID_SIZE];
    static int usemark = 0;

    /* If repl_transfer_size == -1 we still have to read the bulk length
     * from the master reply. */
    if (server.repl_transfer_size == -1) {
        nread = connSyncReadLine(conn,buf,1024,server.repl_syncio_timeout*1000);
        if (nread == -1) {
            serverLog(LL_WARNING,
                "I/O error reading bulk count from MASTER: %s",
                connGetLastError(conn));
            goto error;
        } else {
            /* nread here is returned by connSyncReadLine(), which calls syncReadLine() and
             * convert "\r\n" to '\0' so 1 byte is lost. */
            atomicIncr(server.stat_net_repl_input_bytes, nread+1);
        }

        if (buf[0] == '-') {
            serverLog(LL_WARNING,
                "MASTER aborted replication with an error: %s",
                buf+1);
            goto error;
        } else if (buf[0] == '\0') {
            /* At this stage just a newline works as a PING in order to take
             * the connection live. So we refresh our last interaction
             * timestamp. */
            server.repl_transfer_lastio = server.unixtime;
            return;
        } else if (buf[0] != '$') {
            serverLog(LL_WARNING,"Bad protocol from MASTER, the first byte is not '$' (we received '%s'), are you sure the host and port are right?", buf);
            goto error;
        }

        /* There are two possible forms for the bulk payload. One is the
         * usual $<count> bulk format. The other is used for diskless transfers
         * when the master does not know beforehand the size of the file to
         * transfer. In the latter case, the following format is used:
         *
         * $EOF:<40 bytes delimiter>
         *
         * At the end of the file the announced delimiter is transmitted. The
         * delimiter is long and random enough that the probability of a
         * collision with the actual file content can be ignored. */
        if (strncmp(buf+1,"EOF:",4) == 0 && strlen(buf+5) >= CONFIG_RUN_ID_SIZE) {
            usemark = 1;
            memcpy(eofmark,buf+5,CONFIG_RUN_ID_SIZE);
            memset(lastbytes,0,CONFIG_RUN_ID_SIZE);
            /* Set any repl_transfer_size to avoid entering this code path
             * at the next call. */
            server.repl_transfer_size = 0;
            serverLog(LL_NOTICE,
                "MASTER <-> REPLICA sync: receiving streamed RDB from master with EOF %s",
                use_diskless_load? "to parser":"to disk");
        } else {
            usemark = 0;
            server.repl_transfer_size = strtol(buf+1,NULL,10);
            serverLog(LL_NOTICE,
                "MASTER <-> REPLICA sync: receiving %lld bytes from master %s",
                (long long) server.repl_transfer_size,
                use_diskless_load? "to parser":"to disk");
        }
        return;
    }

    if (!use_diskless_load) {
        /* Read the data from the socket, store it to a file and search
         * for the EOF. */
        if (usemark) {
            readlen = sizeof(buf);
        } else {
            left = server.repl_transfer_size - server.repl_transfer_read;
            readlen = (left < (signed)sizeof(buf)) ? left : (signed)sizeof(buf);
        }

        nread = connRead(conn,buf,readlen);
        if (nread <= 0) {
            if (connGetState(conn) == CONN_STATE_CONNECTED) {
                /* equivalent to EAGAIN */
                return;
            }
            serverLog(LL_WARNING,"I/O error trying to sync with MASTER: %s",
                (nread == -1) ? connGetLastError(conn) : "connection lost");
            cancelReplicationHandshake(1);
            return;
        }
        atomicIncr(server.stat_net_repl_input_bytes, nread);

        /* When a mark is used, we want to detect EOF asap in order to avoid
         * writing the EOF mark into the file... */
        int eof_reached = 0;

        if (usemark) {
            /* Update the last bytes array, and check if it matches our
             * delimiter. */
            if (nread >= CONFIG_RUN_ID_SIZE) {
                memcpy(lastbytes,buf+nread-CONFIG_RUN_ID_SIZE,
                       CONFIG_RUN_ID_SIZE);
            } else {
                int rem = CONFIG_RUN_ID_SIZE-nread;
                memmove(lastbytes,lastbytes+nread,rem);
                memcpy(lastbytes+rem,buf,nread);
            }
            if (memcmp(lastbytes,eofmark,CONFIG_RUN_ID_SIZE) == 0)
                eof_reached = 1;
        }

        /* Update the last I/O time for the replication transfer (used in
         * order to detect timeouts during replication), and write what we
         * got from the socket to the dump file on disk. */
        server.repl_transfer_lastio = server.unixtime;
        if ((nwritten = write(server.repl_transfer_fd,buf,nread)) != nread) {
            serverLog(LL_WARNING,
                "Write error or short write writing to the DB dump file "
                "needed for MASTER <-> REPLICA synchronization: %s",
                (nwritten == -1) ? strerror(errno) : "short write");
            goto error;
        }
        server.repl_transfer_read += nread;

        /* Delete the last 40 bytes from the file if we reached EOF. */
        if (usemark && eof_reached) {
            if (ftruncate(server.repl_transfer_fd,
                server.repl_transfer_read - CONFIG_RUN_ID_SIZE) == -1)
            {
                serverLog(LL_WARNING,
                    "Error truncating the RDB file received from the master "
                    "for SYNC: %s", strerror(errno));
                goto error;
            }
        }

        /* Sync data on disk from time to time, otherwise at the end of the
         * transfer we may suffer a big delay as the memory buffers are copied
         * into the actual disk. */
        if (server.repl_transfer_read >=
            server.repl_transfer_last_fsync_off + REPL_MAX_WRITTEN_BEFORE_FSYNC)
        {
            off_t sync_size = server.repl_transfer_read -
                              server.repl_transfer_last_fsync_off;
            rdb_fsync_range(server.repl_transfer_fd,
                server.repl_transfer_last_fsync_off, sync_size);
            server.repl_transfer_last_fsync_off += sync_size;
        }

        /* Check if the transfer is now complete */
        if (!usemark) {
            if (server.repl_transfer_read == server.repl_transfer_size)
                eof_reached = 1;
        }

        /* If the transfer is yet not complete, we need to read more, so
         * return ASAP and wait for the handler to be called again. */
        if (!eof_reached) return;
    }

    /* We reach this point in one of the following cases:
     *
     * 1. The replica is using diskless replication, that is, it reads data
     *    directly from the socket to the Redis memory, without using
     *    a temporary RDB file on disk. In that case we just block and
     *    read everything from the socket.
     *
     * 2. Or when we are done reading from the socket to the RDB file, in
     *    such case we want just to read the RDB file in memory. */

    /* We need to stop any AOF rewriting child before flushing and parsing
     * the RDB, otherwise we'll create a copy-on-write disaster. */
    if (server.aof_state != AOF_OFF) stopAppendOnly();
    /* Also try to stop save RDB child before flushing and parsing the RDB:
     * 1. Ensure background save doesn't overwrite synced data after being loaded.
     * 2. Avoid copy-on-write disaster. */
    if (server.child_type == CHILD_TYPE_RDB) {
        if (!use_diskless_load) {
            serverLog(LL_NOTICE,
                "Replica is about to load the RDB file received from the "
                "master, but there is a pending RDB child running. "
                "Killing process %ld and removing its temp file to avoid "
                "any race",
                (long) server.child_pid);
        }
        killRDBChild();
    }

    if (use_diskless_load && server.repl_diskless_load == REPL_DISKLESS_LOAD_SWAPDB) {
        /* Initialize empty tempDb dictionaries. */
        diskless_load_tempDb = disklessLoadInitTempDb();
        temp_functions_lib_ctx = functionsLibCtxCreate();

        moduleFireServerEvent(REDISMODULE_EVENT_REPL_ASYNC_LOAD,
                              REDISMODULE_SUBEVENT_REPL_ASYNC_LOAD_STARTED,
                              NULL);
    } else {
        replicationAttachToNewMaster();
    }

    /* Before loading the DB into memory we need to delete the readable
     * handler, otherwise it will get called recursively since
     * rdbLoad() will call the event loop to process events from time to
     * time for non blocking loading. */
    connSetReadHandler(conn, NULL);
    
    serverLog(LL_NOTICE, "MASTER <-> REPLICA sync: Loading DB in memory");
    rdbSaveInfo rsi = RDB_SAVE_INFO_INIT;
    if (use_diskless_load) {
        rio rdb;
        redisDb *dbarray;
        functionsLibCtx* functions_lib_ctx;
        int asyncLoading = 0;

        if (server.repl_diskless_load == REPL_DISKLESS_LOAD_SWAPDB) {
            /* Async loading means we continue serving read commands during full resync, and
             * "swap" the new db with the old db only when loading is done.
             * It is enabled only on SWAPDB diskless replication when master replication ID hasn't changed,
             * because in that state the old content of the db represents a different point in time of the same
             * data set we're currently receiving from the master. */
            if (memcmp(server.replid, server.master_replid, CONFIG_RUN_ID_SIZE) == 0) {
                asyncLoading = 1;
            }
            dbarray = diskless_load_tempDb;
            functions_lib_ctx = temp_functions_lib_ctx;
        } else {
            dbarray = server.db;
            functions_lib_ctx = functionsLibCtxGetCurrent();
            functionsLibCtxClear(functions_lib_ctx);
        }

        disklessLoadingRio = &rdb;
        loadingSetFlags(NULL, server.repl_transfer_size, asyncLoading);
        if (server.repl_diskless_load != REPL_DISKLESS_LOAD_SWAPDB) {
            serverLog(LL_NOTICE, "MASTER <-> REPLICA sync: Flushing old data");
            /* Note that inside loadingSetFlags(), server.loading is set.
             * replicationEmptyDbCallback() may yield back to event-loop to
             * reply -LOADING. */
            emptyData(-1, empty_db_flags, replicationEmptyDbCallback);
        }
        loadingFireEvent(RDBFLAGS_REPLICATION);

        rioInitWithConn(&rdb,conn,server.repl_transfer_size);

        /* Put the socket in blocking mode to simplify RDB transfer.
         * We'll restore it when the RDB is received. */
        connBlock(conn);
        connRecvTimeout(conn, server.repl_timeout*1000);

        int loadingFailed = 0;
        rdbLoadingCtx loadingCtx = { .dbarray = dbarray, .functions_lib_ctx = functions_lib_ctx };
        if (rdbLoadRioWithLoadingCtx(&rdb,RDBFLAGS_REPLICATION,&rsi,&loadingCtx) != C_OK) {
            /* RDB loading failed. */
            serverLog(LL_WARNING,
                      "Failed trying to load the MASTER synchronization DB "
                      "from socket, check server logs.");
            loadingFailed = 1;
        } else if (usemark) {
            /* Verify the end mark is correct. */
            if (!rioRead(&rdb, buf, CONFIG_RUN_ID_SIZE) ||
                memcmp(buf, eofmark, CONFIG_RUN_ID_SIZE) != 0)
            {
                serverLog(LL_WARNING, "Replication stream EOF marker is broken");
                loadingFailed = 1;
            }
        }
        disklessLoadingRio = NULL;

        if (loadingFailed) {
            rioFreeConn(&rdb, NULL);

            if (server.repl_diskless_load == REPL_DISKLESS_LOAD_SWAPDB) {
                /* Discard potentially partially loaded tempDb. */
                moduleFireServerEvent(REDISMODULE_EVENT_REPL_ASYNC_LOAD,
                                      REDISMODULE_SUBEVENT_REPL_ASYNC_LOAD_ABORTED,
                                      NULL);

                disklessLoadDiscardTempDb(diskless_load_tempDb);
                functionsLibCtxFree(temp_functions_lib_ctx);
                serverLog(LL_NOTICE, "MASTER <-> REPLICA sync: Discarding temporary DB in background");
            } else {
                /* Remove the half-loaded data in case we started with an empty replica. */
                emptyData(-1,empty_db_flags,replicationEmptyDbCallback);
            }

            /* Note that replicationEmptyDbCallback() may yield back to event
             * loop to reply -LOADING if flushing the db takes a long time. So,
             * stopLoading() must be called after emptyData() above. */
            stopLoading(0);

            /* This must be called after stopLoading(0) as it checks loading
             * flag in case of rdbchannel replication. */
            cancelReplicationHandshake(1);

            /* Note that there's no point in restarting the AOF on SYNC
             * failure, it'll be restarted when sync succeeds or the replica
             * gets promoted. */
            return;
        }

        /* RDB loading succeeded if we reach this point. */
        if (server.repl_diskless_load == REPL_DISKLESS_LOAD_SWAPDB) {
            /* We will soon swap main db with tempDb and replicas will start
             * to apply data from new master, we must discard the cached
             * master structure and force resync of sub-replicas. */
            replicationAttachToNewMaster();

            serverLog(LL_NOTICE, "MASTER <-> REPLICA sync: Swapping active DB with loaded DB");
            swapMainDbWithTempDb(diskless_load_tempDb);

            /* swap existing functions ctx with the temporary one */
            functionsLibCtxSwapWithCurrent(temp_functions_lib_ctx);

            moduleFireServerEvent(REDISMODULE_EVENT_REPL_ASYNC_LOAD,
                        REDISMODULE_SUBEVENT_REPL_ASYNC_LOAD_COMPLETED,
                        NULL);

            /* Delete the old db as it's useless now. */
            disklessLoadDiscardTempDb(diskless_load_tempDb);
            serverLog(LL_NOTICE, "MASTER <-> REPLICA sync: Discarding old DB in background");
        }

        /* Inform about db change, as replication was diskless and didn't cause a save. */
        server.dirty++;

        stopLoading(1);

        /* Cleanup and restore the socket to the original state to continue
         * with the normal replication. */
        rioFreeConn(&rdb, NULL);
        connNonBlock(conn);
        connRecvTimeout(conn,0);
    } else {

        /* Make sure the new file (also used for persistence) is fully synced
         * (not covered by earlier calls to rdb_fsync_range). */
        if (fsync(server.repl_transfer_fd) == -1) {
            serverLog(LL_WARNING,
                "Failed trying to sync the temp DB to disk in "
                "MASTER <-> REPLICA synchronization: %s",
                strerror(errno));
            cancelReplicationHandshake(1);
            return;
        }

        /* Rename rdb like renaming rewrite aof asynchronously. */
        int old_rdb_fd = open(server.rdb_filename,O_RDONLY|O_NONBLOCK);
        if (rename(server.repl_transfer_tmpfile,server.rdb_filename) == -1) {
            serverLog(LL_WARNING,
                "Failed trying to rename the temp DB into %s in "
                "MASTER <-> REPLICA synchronization: %s",
                server.rdb_filename, strerror(errno));
            cancelReplicationHandshake(1);
            if (old_rdb_fd != -1) close(old_rdb_fd);
            return;
        }
        /* Close old rdb asynchronously. */
        if (old_rdb_fd != -1) bioCreateCloseJob(old_rdb_fd, 0, 0);

        /* Sync the directory to ensure rename is persisted */
        if (fsyncFileDir(server.rdb_filename) == -1) {
            serverLog(LL_WARNING,
                "Failed trying to sync DB directory %s in "
                "MASTER <-> REPLICA synchronization: %s",
                server.rdb_filename, strerror(errno));
            cancelReplicationHandshake(1);
            return;
        }

        if (rdbLoadWithEmptyFunc(server.rdb_filename,&rsi,RDBFLAGS_REPLICATION,rdbLoadEmptyDbFunc) != RDB_OK) {
            serverLog(LL_WARNING,
                "Failed trying to load the MASTER synchronization "
                "DB from disk, check server logs.");
            cancelReplicationHandshake(1);
            if (server.rdb_del_sync_files && allPersistenceDisabled()) {
                serverLog(LL_NOTICE,"Removing the RDB file obtained from "
                                    "the master. This replica has persistence "
                                    "disabled");
                bg_unlink(server.rdb_filename);
            }

            /* Note that there's no point in restarting the AOF on sync failure,
               it'll be restarted when sync succeeds or replica promoted. */
            return;
        }

        /* Cleanup. */
        if (server.rdb_del_sync_files && allPersistenceDisabled()) {
            serverLog(LL_NOTICE,"Removing the RDB file obtained from "
                                "the master. This replica has persistence "
                                "disabled");
            bg_unlink(server.rdb_filename);
        }

        zfree(server.repl_transfer_tmpfile);
        close(server.repl_transfer_fd);
        server.repl_transfer_fd = -1;
        server.repl_transfer_tmpfile = NULL;
    }

    /* Final setup of the connected slave <- master link */
    replicationCreateMasterClient(server.repl_transfer_s,rsi.repl_stream_db);
    server.repl_state = REPL_STATE_CONNECTED;
    server.repl_down_since = 0;

    /* Fire the master link modules event. */
    moduleFireServerEvent(REDISMODULE_EVENT_MASTER_LINK_CHANGE,
                          REDISMODULE_SUBEVENT_MASTER_LINK_UP,
                          NULL);

    /* After a full resynchronization we use the replication ID and
     * offset of the master. The secondary ID / offset are cleared since
     * we are starting a new history. */
    memcpy(server.replid,server.master->replid,sizeof(server.replid));
    server.master_repl_offset = server.master->reploff;
    clearReplicationId2();

    /* Let's create the replication backlog if needed. Slaves need to
     * accumulate the backlog regardless of the fact they have sub-slaves
     * or not, in order to behave correctly if they are promoted to
     * masters after a failover. */
    if (server.repl_backlog == NULL) createReplicationBacklog();
    serverLog(LL_NOTICE, "MASTER <-> REPLICA sync: Finished with success");

    if (server.supervised_mode == SUPERVISED_SYSTEMD) {
        redisCommunicateSystemd("STATUS=MASTER <-> REPLICA sync: Finished with success. Ready to accept connections in read-write mode.\n");
    }

    /* Send the initial ACK immediately to put this replica in online state. */
    if (usemark) replicationSendAck();

    /* Restart the AOF subsystem now that we finished the sync. This
     * will trigger an AOF rewrite, and when done will start appending
     * to the new file. */
    if (server.aof_enabled) {
        serverLog(LL_NOTICE, "MASTER <-> REPLICA sync: Starting AOF after a successful sync");
        startAppendOnlyWithRetry();
    }

    if (rdbchannel) {
        int close_asap;

        if (server.repl_rdb_transfer_s) {
            connClose(server.repl_rdb_transfer_s);
            server.repl_rdb_transfer_s = NULL;
        }
        /* At this point, RDB is loaded. If state is REPL_RDB_CH_STATE_CLOSE_ASAP,
         * it means main channel faced a problem while RDB is being loaded. It
         * stopped replication stream buffering. It's okay though. We'll stream
         * whatever we have into the db, then replica will try psync from the
         * index that it has. */
        close_asap = (server.repl_rdb_ch_state == REPL_RDB_CH_STATE_CLOSE_ASAP);
        /* Finalize fullsync */
        rdbChannelSuccess();

        /* Main channel connection was broken. Let's trigger a psync with master. */
        if (close_asap && server.master)
            freeClientAsync(server.master);
    }

    return;

error:
    cancelReplicationHandshake(1);
    return;
}

char *receiveSynchronousResponse(connection *conn) {
    char buf[256];
    /* Read the reply from the server. */
    if (connSyncReadLine(conn,buf,sizeof(buf),server.repl_syncio_timeout*1000) == -1)
    {
        serverLog(LL_WARNING, "Failed to read response from the server: %s", connGetLastError(conn));
        return NULL;
    }
    server.repl_transfer_lastio = server.unixtime;
    return sdsnew(buf);
}

/* Send a pre-formatted multi-bulk command to the connection. */
char* sendCommandRaw(connection *conn, sds cmd) {
    if (connSyncWrite(conn,cmd,sdslen(cmd),server.repl_syncio_timeout*1000) == -1) {
        return sdscatprintf(sdsempty(),"-Writing to master: %s",
                connGetLastError(conn));
    }
    return NULL;
}

/* Compose a multi-bulk command and send it to the connection.
 * Used to send AUTH and REPLCONF commands to the master before starting the
 * replication.
 *
 * Takes a list of char* arguments, terminated by a NULL argument.
 *
 * The command returns an sds string representing the result of the
 * operation. On error the first byte is a "-".
 */
char *sendCommand(connection *conn, ...) {
    va_list ap;
    sds cmd = sdsempty();
    sds cmdargs = sdsempty();
    size_t argslen = 0;
    char *arg;

    /* Create the command to send to the master, we use redis binary
     * protocol to make sure correct arguments are sent. This function
     * is not safe for all binary data. */
    va_start(ap,conn);
    while(1) {
        arg = va_arg(ap, char*);
        if (arg == NULL) break;
        cmdargs = sdscatprintf(cmdargs,"$%zu\r\n%s\r\n",strlen(arg),arg);
        argslen++;
    }

    cmd = sdscatprintf(cmd,"*%zu\r\n",argslen);
    cmd = sdscatsds(cmd,cmdargs);
    sdsfree(cmdargs);

    va_end(ap);
    char* err = sendCommandRaw(conn, cmd);
    sdsfree(cmd);
    if(err)
        return err;
    return NULL;
}

/* Compose a multi-bulk command and send it to the connection. 
 * Used to send AUTH and REPLCONF commands to the master before starting the
 * replication.
 *
 * argv_lens is optional, when NULL, strlen is used.
 *
 * The command returns an sds string representing the result of the
 * operation. On error the first byte is a "-".
 */
char *sendCommandArgv(connection *conn, int argc, char **argv, size_t *argv_lens) {
    sds cmd = sdsempty();
    char *arg;
    int i;

    /* Create the command to send to the master. */
    cmd = sdscatfmt(cmd,"*%i\r\n",argc);
    for (i=0; i<argc; i++) {
        int len;
        arg = argv[i];
        len = argv_lens ? argv_lens[i] : strlen(arg);
        cmd = sdscatfmt(cmd,"$%i\r\n",len);
        cmd = sdscatlen(cmd,arg,len);
        cmd = sdscatlen(cmd,"\r\n",2);
    }
    char* err = sendCommandRaw(conn, cmd);
    sdsfree(cmd);
    if (err)
        return err;
    return NULL;
}

/* Try a partial resynchronization with the master if we are about to reconnect.
 * If there is no cached master structure, at least try to issue a
 * "PSYNC ? -1" command in order to trigger a full resync using the PSYNC
 * command in order to obtain the master replid and the master replication
 * global offset.
 *
 * This function is designed to be called from syncWithMaster(), so the
 * following assumptions are made:
 *
 * 1) We pass the function an already connected socket "fd".
 * 2) This function does not close the file descriptor "fd". However in case
 *    of successful partial resynchronization, the function will reuse
 *    'fd' as file descriptor of the server.master client structure.
 *
 * The function is split in two halves: if read_reply is 0, the function
 * writes the PSYNC command on the socket, and a new function call is
 * needed, with read_reply set to 1, in order to read the reply of the
 * command. This is useful in order to support non blocking operations, so
 * that we write, return into the event loop, and read when there are data.
 *
 * When read_reply is 0 the function returns PSYNC_WRITE_ERR if there
 * was a write error, or PSYNC_WAIT_REPLY to signal we need another call
 * with read_reply set to 1. However even when read_reply is set to 1
 * the function may return PSYNC_WAIT_REPLY again to signal there were
 * insufficient data to read to complete its work. We should re-enter
 * into the event loop and wait in such a case.
 *
 * The function returns:
 *
 * PSYNC_CONTINUE: If the PSYNC command succeeded and we can continue.
 * PSYNC_FULLRESYNC: If PSYNC is supported but a full resync is needed.
 *                   In this case the master replid and global replication
 *                   offset is saved.
 * PSYNC_NOT_SUPPORTED: If the server does not understand PSYNC at all and
 *                      the caller should fall back to SYNC.
 * PSYNC_WRITE_ERROR: There was an error writing the command to the socket.
 * PSYNC_WAIT_REPLY: Call again the function with read_reply set to 1.
 * PSYNC_TRY_LATER: Master is currently in a transient error condition.
 *
 * Notable side effects:
 *
 * 1) As a side effect of the function call the function removes the readable
 *    event handler from "fd", unless the return value is PSYNC_WAIT_REPLY.
 * 2) server.master_initial_offset is set to the right value according
 *    to the master reply. This will be used to populate the 'server.master'
 *    structure replication offset.
 */

#define PSYNC_WRITE_ERROR 0
#define PSYNC_WAIT_REPLY 1
#define PSYNC_CONTINUE 2
#define PSYNC_FULLRESYNC 3
#define PSYNC_NOT_SUPPORTED 4
#define PSYNC_TRY_LATER 5
#define PSYNC_FULLRESYNC_RDBCHANNEL 6
int slaveTryPartialResynchronization(connection *conn, int read_reply) {
    char *psync_replid;
    char psync_offset[32];
    sds reply;

    /* Writing half */
    if (!read_reply) {
        /* Initially set master_initial_offset to -1 to mark the current
         * master replid and offset as not valid. Later if we'll be able to do
         * a FULL resync using the PSYNC command we'll set the offset at the
         * right value, so that this information will be propagated to the
         * client structure representing the master into server.master. */
        server.master_initial_offset = -1;

        if (server.cached_master) {
            psync_replid = server.cached_master->replid;
            snprintf(psync_offset,sizeof(psync_offset),"%lld", server.cached_master->reploff+1);
            serverLog(LL_NOTICE,"Trying a partial resynchronization (request %s:%s).", psync_replid, psync_offset);
        } else {
            serverLog(LL_NOTICE,"Partial resynchronization not possible (no cached master)");
            psync_replid = "?";
            memcpy(psync_offset,"-1",3);
        }

        /* Issue the PSYNC command, if this is a master with a failover in
         * progress then send the failover argument to the replica to cause it
         * to become a master */
        if (server.failover_state == FAILOVER_IN_PROGRESS) {
            reply = sendCommand(conn,"PSYNC",psync_replid,psync_offset,"FAILOVER",NULL);
        } else {
            reply = sendCommand(conn,"PSYNC",psync_replid,psync_offset,NULL);
        }

        if (reply != NULL) {
            serverLog(LL_WARNING,"Unable to send PSYNC to master: %s",reply);
            sdsfree(reply);
            connSetReadHandler(conn, NULL);
            return PSYNC_WRITE_ERROR;
        }
        return PSYNC_WAIT_REPLY;
    }

    /* Reading half */
    reply = receiveSynchronousResponse(conn);
    /* Master did not reply to PSYNC */
    if (reply == NULL) {
        connSetReadHandler(conn, NULL);
        serverLog(LL_WARNING, "Master did not reply to PSYNC, will try later");
        return PSYNC_TRY_LATER;
    }

    if (sdslen(reply) == 0) {
        /* The master may send empty newlines after it receives PSYNC
         * and before to reply, just to keep the connection alive. */
        sdsfree(reply);
        return PSYNC_WAIT_REPLY;
    }

    connSetReadHandler(conn, NULL);

    if (!strncmp(reply,"+FULLRESYNC",11)) {
        char *replid = NULL, *offset = NULL;

        /* FULL RESYNC, parse the reply in order to extract the replid
         * and the replication offset. */
        replid = strchr(reply,' ');
        if (replid) {
            replid++;
            offset = strchr(replid,' ');
            if (offset) offset++;
        }
        if (!replid || !offset || (offset-replid-1) != CONFIG_RUN_ID_SIZE) {
            serverLog(LL_WARNING,
                "Master replied with wrong +FULLRESYNC syntax.");
            /* This is an unexpected condition, actually the +FULLRESYNC
             * reply means that the master supports PSYNC, but the reply
             * format seems wrong. To stay safe we blank the master
             * replid to make sure next PSYNCs will fail. */
            memset(server.master_replid,0,CONFIG_RUN_ID_SIZE+1);
        } else {
            memcpy(server.master_replid, replid, offset-replid-1);
            server.master_replid[CONFIG_RUN_ID_SIZE] = '\0';
            server.master_initial_offset = strtoll(offset,NULL,10);
            serverLog(LL_NOTICE,"Full resync from master: %s:%lld",
                server.master_replid,
                server.master_initial_offset);
        }
        sdsfree(reply);
        return PSYNC_FULLRESYNC;
    }

    if (!strncmp(reply, "+RDBCHANNELSYNC", strlen("+RDBCHANNELSYNC"))) {
        char *client_id = strchr(reply,' ');
        if (client_id)
            client_id++;

        if (!client_id) {
            serverLog(LL_WARNING,
                      "Master replied with wrong +RDBCHANNELSYNC syntax: %s", reply);
            return PSYNC_NOT_SUPPORTED;
        }
        server.repl_main_ch_client_id = strtoll(client_id, NULL, 10);;
        /* A response of +RDBCHANNELSYNC from the master implies that partial
         * synchronization is not possible and that the master supports full
         * sync using dedicated RDB channel. Full sync will continue that way.*/
        serverLog(LL_NOTICE, "PSYNC is not possible, initialize RDB channel.");
        sdsfree(reply);
        return PSYNC_FULLRESYNC_RDBCHANNEL;
    }

    if (!strncmp(reply,"+CONTINUE",9)) {
        /* Partial resync was accepted. */
        serverLog(LL_NOTICE,
            "Successful partial resynchronization with master.");

        /* Check the new replication ID advertised by the master. If it
         * changed, we need to set the new ID as primary ID, and set
         * secondary ID as the old master ID up to the current offset, so
         * that our sub-slaves will be able to PSYNC with us after a
         * disconnection. */
        char *start = reply+10;
        char *end = reply+9;
        while(end[0] != '\r' && end[0] != '\n' && end[0] != '\0') end++;
        if (end-start == CONFIG_RUN_ID_SIZE) {
            char new[CONFIG_RUN_ID_SIZE+1];
            memcpy(new,start,CONFIG_RUN_ID_SIZE);
            new[CONFIG_RUN_ID_SIZE] = '\0';

            if (strcmp(new,server.cached_master->replid)) {
                /* Master ID changed. */
                serverLog(LL_NOTICE,"Master replication ID changed to %s",new);

                /* Set the old ID as our ID2, up to the current offset+1. */
                memcpy(server.replid2,server.cached_master->replid,
                    sizeof(server.replid2));
                server.second_replid_offset = server.master_repl_offset+1;

                /* Update the cached master ID and our own primary ID to the
                 * new one. */
                memcpy(server.replid,new,sizeof(server.replid));
                memcpy(server.cached_master->replid,new,sizeof(server.replid));

                /* Disconnect all the sub-slaves: they need to be notified. */
                disconnectSlaves();
            }
        }

        /* Setup the replication to continue. */
        sdsfree(reply);
        replicationResurrectCachedMaster(conn);

        /* If this instance was restarted and we read the metadata to
         * PSYNC from the persistence file, our replication backlog could
         * be still not initialized. Create it. */
        if (server.repl_backlog == NULL) createReplicationBacklog();
        return PSYNC_CONTINUE;
    }

    /* If we reach this point we received either an error (since the master does
     * not understand PSYNC or because it is in a special state and cannot
     * serve our request), or an unexpected reply from the master.
     *
     * Return PSYNC_NOT_SUPPORTED on errors we don't understand, otherwise
     * return PSYNC_TRY_LATER if we believe this is a transient error. */

    if (!strncmp(reply,"-NOMASTERLINK",13) ||
        !strncmp(reply,"-LOADING",8))
    {
        serverLog(LL_NOTICE,
            "Master is currently unable to PSYNC "
            "but should be in the future: %s", reply);
        sdsfree(reply);
        return PSYNC_TRY_LATER;
    }

    if (strncmp(reply,"-ERR",4)) {
        /* If it's not an error, log the unexpected event. */
        serverLog(LL_WARNING,
            "Unexpected reply to PSYNC from master: %s", reply);
    } else {
        serverLog(LL_NOTICE,
            "Master does not support PSYNC or is in "
            "error state (reply: %s)", reply);
    }
    sdsfree(reply);
    return PSYNC_NOT_SUPPORTED;
}

/* This handler fires when the non blocking connect was able to
 * establish a connection with the master. */
void syncWithMaster(connection *conn) {
    char tmpfile[256], *err = NULL;
    int dfd = -1, maxtries = 5;
    int psync_result;

    /* If this event fired after the user turned the instance into a master
     * with SLAVEOF NO ONE we must just return ASAP. */
    if (server.repl_state == REPL_STATE_NONE) {
        connClose(conn);
        return;
    }

    /* Check for errors in the socket: after a non blocking connect() we
     * may find that the socket is in error state. */
    if (connGetState(conn) != CONN_STATE_CONNECTED) {
        serverLog(LL_WARNING,"Error condition on socket for SYNC: %s",
                connGetLastError(conn));
        goto error;
    }

    /* Send a PING to check the master is able to reply without errors. */
    if (server.repl_state == REPL_STATE_CONNECTING) {
        serverLog(LL_NOTICE,"Non blocking connect for SYNC fired the event.");
        /* Delete the writable event so that the readable event remains
         * registered and we can wait for the PONG reply. */
        connSetReadHandler(conn, syncWithMaster);
        connSetWriteHandler(conn, NULL);
        server.repl_state = REPL_STATE_RECEIVE_PING_REPLY;
        /* Send the PING, don't check for errors at all, we have the timeout
         * that will take care about this. */
        err = sendCommand(conn,"PING",NULL);
        if (err) goto write_error;
        return;
    }

    /* Receive the PONG command. */
    if (server.repl_state == REPL_STATE_RECEIVE_PING_REPLY) {
        err = receiveSynchronousResponse(conn);

        /* The master did not reply */
        if (err == NULL) goto no_response_error;

        /* We accept only two replies as valid, a positive +PONG reply
         * (we just check for "+") or an authentication error.
         * Note that older versions of Redis replied with "operation not
         * permitted" instead of using a proper error code, so we test
         * both. */
        if (err[0] != '+' &&
            strncmp(err,"-NOAUTH",7) != 0 &&
            strncmp(err,"-NOPERM",7) != 0 &&
            strncmp(err,"-ERR operation not permitted",28) != 0)
        {
            serverLog(LL_WARNING,"Error reply to PING from master: '%s'",err);
            sdsfree(err);
            goto error;
        } else {
            serverLog(LL_NOTICE,
                "Master replied to PING, replication can continue...");
        }
        sdsfree(err);
        err = NULL;
        server.repl_state = REPL_STATE_SEND_HANDSHAKE;
    }

    if (server.repl_state == REPL_STATE_SEND_HANDSHAKE) {
        /* AUTH with the master if required. */
        if (server.masterauth) {
            char *args[3] = {"AUTH",NULL,NULL};
            size_t lens[3] = {4,0,0};
            int argc = 1;
            if (server.masteruser) {
                args[argc] = server.masteruser;
                lens[argc] = strlen(server.masteruser);
                argc++;
            }
            args[argc] = server.masterauth;
            lens[argc] = sdslen(server.masterauth);
            argc++;
            err = sendCommandArgv(conn, argc, args, lens);
            if (err) goto write_error;
        }

        /* Set the slave port, so that Master's INFO command can list the
         * slave listening port correctly. */
        {
            char buf[LONG_STR_SIZE];

            slaveGetPortStr(buf, sizeof(buf));
            err = sendCommand(conn,"REPLCONF",
                    "listening-port",buf, NULL);
            if (err) goto write_error;
        }

        /* Set the slave ip, so that Master's INFO command can list the
         * slave IP address port correctly in case of port forwarding or NAT.
         * Skip REPLCONF ip-address if there is no slave-announce-ip option set. */
        if (server.slave_announce_ip) {
            err = sendCommand(conn,"REPLCONF",
                    "ip-address",server.slave_announce_ip, NULL);
            if (err) goto write_error;
        }

        /* Inform the master of our (slave) capabilities.
         *
         * EOF: supports EOF-style RDB transfer for diskless replication.
         * PSYNC2: supports PSYNC v2, so understands +CONTINUE <new repl ID>.
         *
         * The master will ignore capabilities it does not understand. */
        err = sendCommand(conn,"REPLCONF",
                          "capa","eof","capa","psync2",
                          server.repl_rdb_channel ? "capa" : NULL, "rdb-channel-repl", NULL);

        if (err) goto write_error;

        server.repl_state = REPL_STATE_RECEIVE_AUTH_REPLY;
        return;
    }

    if (server.repl_state == REPL_STATE_RECEIVE_AUTH_REPLY && !server.masterauth)
        server.repl_state = REPL_STATE_RECEIVE_PORT_REPLY;

    /* Receive AUTH reply. */
    if (server.repl_state == REPL_STATE_RECEIVE_AUTH_REPLY) {
        err = receiveSynchronousResponse(conn);
        if (err == NULL) goto no_response_error;
        if (err[0] == '-') {
            serverLog(LL_WARNING,"Unable to AUTH to MASTER: %s",err);
            sdsfree(err);
            goto error;
        }
        sdsfree(err);
        err = NULL;
        server.repl_state = REPL_STATE_RECEIVE_PORT_REPLY;
        return;
    }

    /* Receive REPLCONF listening-port reply. */
    if (server.repl_state == REPL_STATE_RECEIVE_PORT_REPLY) {
        err = receiveSynchronousResponse(conn);
        if (err == NULL) goto no_response_error;
        /* Ignore the error if any, not all the Redis versions support
         * REPLCONF listening-port. */
        if (err[0] == '-') {
            serverLog(LL_NOTICE,"(Non critical) Master does not understand "
                                "REPLCONF listening-port: %s", err);
        }
        sdsfree(err);
        server.repl_state = REPL_STATE_RECEIVE_IP_REPLY;
        return;
    }

    if (server.repl_state == REPL_STATE_RECEIVE_IP_REPLY && !server.slave_announce_ip)
        server.repl_state = REPL_STATE_RECEIVE_CAPA_REPLY;

    /* Receive REPLCONF ip-address reply. */
    if (server.repl_state == REPL_STATE_RECEIVE_IP_REPLY) {
        err = receiveSynchronousResponse(conn);
        if (err == NULL) goto no_response_error;
        /* Ignore the error if any, not all the Redis versions support
         * REPLCONF ip-address. */
        if (err[0] == '-') {
            serverLog(LL_NOTICE,"(Non critical) Master does not understand "
                                "REPLCONF ip-address: %s", err);
        }
        sdsfree(err);
        server.repl_state = REPL_STATE_RECEIVE_CAPA_REPLY;
        return;
    }

    /* Receive CAPA reply. */
    if (server.repl_state == REPL_STATE_RECEIVE_CAPA_REPLY) {
        err = receiveSynchronousResponse(conn);
        if (err == NULL) goto no_response_error;
        /* Ignore the error if any, not all the Redis versions support
         * REPLCONF capa. */
        if (err[0] == '-') {
            serverLog(LL_NOTICE,"(Non critical) Master does not understand "
                                  "REPLCONF capa: %s", err);
        }
        sdsfree(err);
        err = NULL;
        server.repl_state = REPL_STATE_SEND_PSYNC;
    }

    /* Try a partial resynchronization. If we don't have a cached master
     * slaveTryPartialResynchronization() will at least try to use PSYNC
     * to start a full resynchronization so that we get the master replid
     * and the global offset, to try a partial resync at the next
     * reconnection attempt. */
    if (server.repl_state == REPL_STATE_SEND_PSYNC) {
        if (slaveTryPartialResynchronization(conn,0) == PSYNC_WRITE_ERROR) {
            err = sdsnew("Write error sending the PSYNC command.");
            abortFailover("Write error to failover target");
            goto write_error;
        }
        server.repl_state = REPL_STATE_RECEIVE_PSYNC_REPLY;
        return;
    }

    /* If reached this point, we should be in REPL_STATE_RECEIVE_PSYNC_REPLY. */
    if (server.repl_state != REPL_STATE_RECEIVE_PSYNC_REPLY) {
        serverLog(LL_WARNING,"syncWithMaster(): state machine error, "
                             "state should be RECEIVE_PSYNC but is %d",
                             server.repl_state);
        goto error;
    }

    psync_result = slaveTryPartialResynchronization(conn,1);
    if (psync_result == PSYNC_WAIT_REPLY) return; /* Try again later... */

    /* Check the status of the planned failover. We expect PSYNC_CONTINUE,
     * but there is nothing technically wrong with a full resync which
     * could happen in edge cases. */
    if (server.failover_state == FAILOVER_IN_PROGRESS) {
        if (psync_result == PSYNC_CONTINUE ||
            psync_result == PSYNC_FULLRESYNC ||
            psync_result == PSYNC_FULLRESYNC_RDBCHANNEL)
        {
            clearFailoverState();
        } else {
            abortFailover("Failover target rejected psync request");
            return;
        }
    }

    /* If the master is in an transient error, we should try to PSYNC
     * from scratch later, so go to the error path. This happens when
     * the server is loading the dataset or is not connected with its
     * master and so forth. */
    if (psync_result == PSYNC_TRY_LATER) goto error;

    /* Note: if PSYNC does not return WAIT_REPLY, it will take care of
     * uninstalling the read handler from the file descriptor. */

    if (psync_result == PSYNC_CONTINUE) {
        serverLog(LL_NOTICE, "MASTER <-> REPLICA sync: Master accepted a Partial Resynchronization.");
        if (server.supervised_mode == SUPERVISED_SYSTEMD) {
            redisCommunicateSystemd("STATUS=MASTER <-> REPLICA sync: Partial Resynchronization accepted. Ready to accept connections in read-write mode.\n");
        }
        return;
    }

    /* Fall back to SYNC if needed. Otherwise psync_result == PSYNC_FULLRESYNC
     * and the server.master_replid and master_initial_offset are
     * already populated. */
    if (psync_result == PSYNC_NOT_SUPPORTED) {
        serverLog(LL_NOTICE,"Retrying with SYNC...");
        if (connSyncWrite(conn,"SYNC\r\n",6,server.repl_syncio_timeout*1000) == -1) {
            serverLog(LL_WARNING,"I/O error writing to MASTER: %s",
                connGetLastError(conn));
            goto error;
        }
    }

    /* Prepare a suitable temp file for bulk transfer */
    if (!useDisklessLoad()) {
        while(maxtries--) {
            snprintf(tmpfile,256,
                "temp-%d.%ld.rdb",(int)server.unixtime,(long int)getpid());
            dfd = open(tmpfile,O_CREAT|O_WRONLY|O_EXCL,0644);
            if (dfd != -1) break;
            sleep(1);
        }
        if (dfd == -1) {
            serverLog(LL_WARNING,"Opening the temp file needed for MASTER <-> REPLICA synchronization: %s",strerror(errno));
            goto error;
        }
        server.repl_transfer_tmpfile = zstrdup(tmpfile);
        server.repl_transfer_fd = dfd;
    }

    server.repl_transfer_size = -1;
    server.repl_transfer_read = 0;
    server.repl_transfer_last_fsync_off = 0;
    server.repl_transfer_lastio = server.unixtime;

    /* Using rdb channel replication, the master responded +RDBCHANNELSYNC.
     * We need to initialize the RDB channel. */
    if (psync_result == PSYNC_FULLRESYNC_RDBCHANNEL) {
        /* Create RDB connection */
        server.repl_rdb_transfer_s = connCreate(server.el, connTypeOfReplication());
        if (connConnect(server.repl_rdb_transfer_s, server.masterhost,
                        server.masterport, server.bind_source_addr,
                        rdbChannelFullSyncWithMaster) == C_ERR) {
            serverLog(LL_WARNING, "Unable to connect to master: %s", connGetLastError(server.repl_rdb_transfer_s));
            goto error;
        }
        server.repl_rdb_ch_state = REPL_RDB_CH_SEND_HANDSHAKE;
        connSetReadHandler(server.repl_transfer_s, NULL);
        return;
    }

    /* Setup the non blocking download of the bulk file. */
    if (connSetReadHandler(conn, readSyncBulkPayload)
            == C_ERR)
    {
        char conninfo[CONN_INFO_LEN];
        serverLog(LL_WARNING,
            "Can't create readable event for SYNC: %s (%s)",
            strerror(errno), connGetInfo(conn, conninfo, sizeof(conninfo)));
        goto error;
    }

    server.repl_state = REPL_STATE_TRANSFER;
    return;

no_response_error: /* Handle receiveSynchronousResponse() error when master has no reply */
    serverLog(LL_WARNING, "Master did not respond to command during SYNC handshake");
    /* Fall through to regular error handling */

error:
    if (dfd != -1) close(dfd);
    connClose(conn);
    if (server.repl_rdb_transfer_s)
        connClose(server.repl_rdb_transfer_s);
    server.repl_rdb_transfer_s = NULL;
    server.repl_transfer_s = NULL;
    if (server.repl_transfer_fd != -1)
        close(server.repl_transfer_fd);
    if (server.repl_transfer_tmpfile)
        zfree(server.repl_transfer_tmpfile);
    server.repl_transfer_tmpfile = NULL;
    server.repl_transfer_fd = -1;
    server.repl_state = REPL_STATE_CONNECT;
    return;

write_error: /* Handle sendCommand() errors. */
    serverLog(LL_WARNING,"Sending command to master in replication handshake: %s", err);
    sdsfree(err);
    goto error;
}

int connectWithMaster(void) {
    server.repl_transfer_s = connCreate(server.el, connTypeOfReplication());
    if (connConnect(server.repl_transfer_s, server.masterhost, server.masterport,
                server.bind_source_addr, syncWithMaster) == C_ERR) {
        serverLog(LL_WARNING,"Unable to connect to MASTER: %s",
                connGetLastError(server.repl_transfer_s));
        connClose(server.repl_transfer_s);
        server.repl_transfer_s = NULL;
        return C_ERR;
    }


    server.repl_transfer_lastio = server.unixtime;
    server.repl_state = REPL_STATE_CONNECTING;
    serverLog(LL_NOTICE,"MASTER <-> REPLICA sync started");
    return C_OK;
}

/* This function can be called when a non blocking connection is currently
 * in progress to undo it.
 * Never call this function directly, use cancelReplicationHandshake() instead.
 */
void undoConnectWithMaster(void) {
    connClose(server.repl_transfer_s);
    server.repl_transfer_s = NULL;
}

/* Abort the async download of the bulk dataset while SYNC-ing with master.
 * Never call this function directly, use cancelReplicationHandshake() instead.
 */
void replicationAbortSyncTransfer(void) {
    serverAssert(server.repl_state == REPL_STATE_TRANSFER);
    undoConnectWithMaster();
    if (server.repl_transfer_fd!=-1) {
        close(server.repl_transfer_fd);
        bg_unlink(server.repl_transfer_tmpfile);
        zfree(server.repl_transfer_tmpfile);
        server.repl_transfer_tmpfile = NULL;
        server.repl_transfer_fd = -1;
    }
}

/* This function aborts a non blocking replication attempt if there is one
 * in progress, by canceling the non-blocking connect attempt or
 * the initial bulk transfer.
 *
 * If there was a replication handshake in progress 1 is returned and
 * the replication state (server.repl_state) set to REPL_STATE_CONNECT.
 *
 * Otherwise zero is returned and no operation is performed at all. */
int cancelReplicationHandshake(int reconnect) {
    if (rdbChannelAbortRdbTransfer() != C_OK)
        return 1;

    if (server.repl_state == REPL_STATE_TRANSFER) {
        replicationAbortSyncTransfer();
        server.repl_state = REPL_STATE_CONNECT;
    } else if (server.repl_state == REPL_STATE_CONNECTING ||
               slaveIsInHandshakeState())
    {
        undoConnectWithMaster();
        server.repl_state = REPL_STATE_CONNECT;
    } else {
        return 0;
    }

    if (!reconnect)
        return 1;

    /* try to re-connect without waiting for replicationCron, this is needed
     * for the "diskless loading short read" test. */
    serverLog(LL_NOTICE,"Reconnecting to MASTER %s:%d after failure",
        server.masterhost, server.masterport);
    connectWithMaster();

    return 1;
}

/* Set replication to the specified master address and port. */
void replicationSetMaster(char *ip, int port) {
    int was_master = server.masterhost == NULL;

    sdsfree(server.masterhost);
    server.masterhost = NULL;
    if (server.master) {
        freeClient(server.master);
    }
    disconnectAllBlockedClients(); /* Clients blocked in master, now slave. */

    /* Setting masterhost only after the call to freeClient since it calls
     * replicationHandleMasterDisconnection which can trigger a re-connect
     * directly from within that call. */
    server.masterhost = sdsnew(ip);
    server.masterport = port;

    /* Update oom_score_adj */
    setOOMScoreAdj(-1);

    /* Here we don't disconnect with replicas, since they may hopefully be able
     * to partially resync with us. We will disconnect with replicas and force
     * them to resync with us when changing replid on partially resync with new
     * master, or finishing transferring RDB and preparing loading DB on full
     * sync with new master. */

    cancelReplicationHandshake(0);
    /* Before destroying our master state, create a cached master using
     * our own parameters, to later PSYNC with the new master. */
    if (was_master) {
        replicationDiscardCachedMaster();
        replicationCacheMasterUsingMyself();
    }

    /* Fire the role change modules event. */
    moduleFireServerEvent(REDISMODULE_EVENT_REPLICATION_ROLE_CHANGED,
                          REDISMODULE_EVENT_REPLROLECHANGED_NOW_REPLICA,
                          NULL);

    /* Fire the master link modules event. */
    if (server.repl_state == REPL_STATE_CONNECTED)
        moduleFireServerEvent(REDISMODULE_EVENT_MASTER_LINK_CHANGE,
                              REDISMODULE_SUBEVENT_MASTER_LINK_DOWN,
                              NULL);

    server.repl_state = REPL_STATE_CONNECT;
    serverLog(LL_NOTICE,"Connecting to MASTER %s:%d",
        server.masterhost, server.masterport);
    connectWithMaster();
}

/* Cancel replication, setting the instance as a master itself. */
void replicationUnsetMaster(void) {
    if (server.masterhost == NULL) return; /* Nothing to do. */

    /* Fire the master link modules event. */
    if (server.repl_state == REPL_STATE_CONNECTED)
        moduleFireServerEvent(REDISMODULE_EVENT_MASTER_LINK_CHANGE,
                              REDISMODULE_SUBEVENT_MASTER_LINK_DOWN,
                              NULL);

    /* Clear masterhost first, since the freeClient calls
     * replicationHandleMasterDisconnection which can attempt to re-connect. */
    sdsfree(server.masterhost);
    server.masterhost = NULL;
    if (server.master) freeClient(server.master);
    replicationDiscardCachedMaster();
    cancelReplicationHandshake(0);
    /* When a slave is turned into a master, the current replication ID
     * (that was inherited from the master at synchronization time) is
     * used as secondary ID up to the current offset, and a new replication
     * ID is created to continue with a new replication history. */
    shiftReplicationId();
    /* Disconnecting all the slaves is required: we need to inform slaves
     * of the replication ID change (see shiftReplicationId() call). However
     * the slaves will be able to partially resync with us, so it will be
     * a very fast reconnection. */
    disconnectSlaves();
    server.repl_state = REPL_STATE_NONE;

    /* We need to make sure the new master will start the replication stream
     * with a SELECT statement. This is forced after a full resync, but
     * with PSYNC version 2, there is no need for full resync after a
     * master switch. */
    server.slaveseldb = -1;

    /* Update oom_score_adj */
    setOOMScoreAdj(-1);

    /* Once we turn from slave to master, we consider the starting time without
     * slaves (that is used to count the replication backlog time to live) as
     * starting from now. Otherwise the backlog will be freed after a
     * failover if slaves do not connect immediately. */
    server.repl_no_slaves_since = server.unixtime;
    
    /* Reset down time so it'll be ready for when we turn into replica again. */
    server.repl_down_since = 0;

    /* Fire the role change modules event. */
    moduleFireServerEvent(REDISMODULE_EVENT_REPLICATION_ROLE_CHANGED,
                          REDISMODULE_EVENT_REPLROLECHANGED_NOW_MASTER,
                          NULL);

    /* Restart the AOF subsystem in case we shut it down during a sync when
     * we were still a slave. */
    if (server.aof_enabled && server.aof_state == AOF_OFF) {
        serverLog(LL_NOTICE, "Restarting AOF after becoming master");
        startAppendOnlyWithRetry();
    }
}

/* This function is called when the slave lose the connection with the
 * master into an unexpected way. */
void replicationHandleMasterDisconnection(void) {
    /* Fire the master link modules event. */
    if (server.repl_state == REPL_STATE_CONNECTED)
        moduleFireServerEvent(REDISMODULE_EVENT_MASTER_LINK_CHANGE,
                              REDISMODULE_SUBEVENT_MASTER_LINK_DOWN,
                              NULL);

    server.master = NULL;
    server.repl_state = REPL_STATE_CONNECT;
    server.repl_down_since = server.unixtime;
    /* We lost connection with our master, don't disconnect slaves yet,
     * maybe we'll be able to PSYNC with our master later. We'll disconnect
     * the slaves only if we'll have to do a full resync with our master. */

    /* Try to re-connect immediately rather than wait for replicationCron
     * waiting 1 second may risk backlog being recycled. */
    if (server.masterhost) {
        serverLog(LL_NOTICE,"Reconnecting to MASTER %s:%d",
            server.masterhost, server.masterport);
        connectWithMaster();
    }
}

/* Rdb channel for full sync
 *
 * - During a full sync, when master is delivering RDB to the replica, incoming
 *   write commands are kept in a replication buffer in order to be sent to the
 *   replica once RDB delivery is completed. If RDB delivery takes a long time,
 *   it might create memory pressure on master. Also, once a replica connection
 *   accumulates replication data which is larger than output buffer limits,
 *   master will kill replica connection. This may cause a replication failure.
 *
 *   The main benefit of the rdb channel replication is streaming incoming
 *   commands in parallel to the RDB delivery. This approach shifts replication
 *   stream buffering to the replica and reduces load on master. We do this by
 *   opening another connection for RDB delivery. The main channel on replica
 *   will be receiving replication stream while rdb channel is receiving the RDB.
 *
 *   This feature also helps to reduce master's main process CPU load. By
 *   opening a dedicated connection for the RDB transfer, the bgsave process has
 *   direct access to the new connection and it will stream RDB directly to the
 *   replicas. Before this change, due to TLS connection restriction, the bgsave
 *   process was writing RDB bytes to a pipe and the main process was forwarding
 *   it to the replica. This is no longer necessary, the main process can avoid
 *   these expensive socket read/write syscalls.
 *
 *  Implementation
 *  - When replica connects to the master, it sends 'rdb-channel-repl' as part
 *    of capability exchange to let master to know replica supports rdb channel.
 *  - When replica lacks sufficient data for PSYNC, master sends +RDBCHANNELSYNC
 *    reply with replica's client id. As the next step, the replica opens a new
 *    connection (rdb-channel) and configures it against the master with the
 *    appropriate capabilities and requirements. It also sends given client id
 *    back to master over rdbchannel so that master can associate these
 *    channels (initial replica connection will be referred as main-channel)
 *    Then, replica requests fullsync using the RDB channel.
 *  - Prior to forking, master attaches the replica's main channel to the
 *    replication backlog to deliver replication stream starting at the snapshot
 *    end offset.
 *  - The master main process sends replication stream via the main channel,
 *    while the bgsave process sends the RDB directly to the replica via the
 *    rdb-channel. Replica accumulates replication stream in a local buffer,
 *    while the RDB is being loaded into the memory.
 *  - Once the replica completes loading the rdb, it drops the rdb channel and
 *    streams the accumulated replication stream into the db. Sync is completed.
 *
 * * Replica state machine *
 *
 * Main channel state
 * ┌───────────────────┐
 * │RECEIVE_PING_REPLY │
 * └────────┬──────────┘
 *          │ +PONG
 * ┌────────▼──────────┐
 * │SEND_HANDSHAKE     │                     RDB channel state
 * └────────┬──────────┘            ┌───────────────────────────────┐
 *          │+OK                ┌───► RDB_CH_SEND_HANDSHAKE         │
 * ┌────────▼──────────┐        │   └──────────────┬────────────────┘
 * │RECEIVE_AUTH_REPLY │        │    REPLCONF main-ch-client-id <clientid>
 * └────────┬──────────┘        │   ┌──────────────▼────────────────┐
 *          │+OK                │   │ RDB_CH_RECEIVE_AUTH_REPLY     │
 * ┌────────▼──────────┐        │   └──────────────┬────────────────┘
 * │RECEIVE_PORT_REPLY │        │                  │ +OK
 * └────────┬──────────┘        │   ┌──────────────▼────────────────┐
 *          │+OK                │   │  RDB_CH_RECEIVE_REPLCONF_REPLY│
 * ┌────────▼──────────┐        │   └──────────────┬────────────────┘
 * │RECEIVE_IP_REPLY   │        │                  │ +OK
 * └────────┬──────────┘        │   ┌──────────────▼────────────────┐
 *          │+OK                │   │ RDB_CH_RECEIVE_FULLRESYNC     │
 * ┌────────▼──────────┐        │   └──────────────┬────────────────┘
 * │RECEIVE_CAPA_REPLY │        │                  │+FULLRESYNC
 * └────────┬──────────┘        │                  │Rdb delivery
 *          │                   │   ┌──────────────▼────────────────┐
 * ┌────────▼──────────┐        │   │ RDB_CH_RDB_LOADING            │
 * │SEND_PSYNC         │        │   └──────────────┬────────────────┘
 * └─┬─────────────────┘        │                  │ Done loading
 *   │PSYNC (use cached-master) │                  │
 * ┌─▼─────────────────┐        │                  │
 * │RECEIVE_PSYNC_REPLY│        │    ┌────────────►│ Replica streams replication
 * └─┬─────────────────┘        │    │             │ buffer into memory
 *   │                          │    │             │
 *   │+RDBCHANNELSYNC client-id │    │             │
 *   ├──────┬───────────────────┘    │             │
 *   │      │ Main channel           │             │
 *   │      │ accumulates repl data  │             │
 *   │   ┌──▼────────────────┐       │     ┌───────▼───────────┐
 *   │   │ REPL_TRANSFER     ├───────┘     │    CONNECTED      │
 *   │   └───────────────────┘             └────▲───▲──────────┘
 *   │                                          │   │
 *   │                                          │   │
 *   │  +FULLRESYNC    ┌───────────────────┐    │   │
 *   ├────────────────► REPL_TRANSFER      ├────┘   │
 *   │                 └───────────────────┘        │
 *   │  +CONTINUE                                   │
 *   └──────────────────────────────────────────────┘
 */

/* Replication: Replica side. */
static int rdbChannelSendHandshake(connection *conn, sds *err) {
    /* AUTH with the master if required. */
    if (server.masterauth) {
        char *args[] = {"AUTH", NULL, NULL};
        size_t lens[] = {4, 0, 0};
        int argc = 1;
        if (server.masteruser) {
            args[argc] = server.masteruser;
            lens[argc] = strlen(server.masteruser);
            argc++;
        }
        args[argc] = server.masterauth;
        lens[argc] = sdslen(server.masterauth);
        argc++;
        *err = sendCommandArgv(conn, argc, args, lens);
        if (*err) {
            serverLog(LL_WARNING, "Error sending AUTH to master in rdb channel replication handshake: %s", *err);
            return C_ERR;
        }
    }

    char buf[LONG_STR_SIZE];
    slaveGetPortStr(buf, sizeof(buf));

    char cid[LONG_STR_SIZE];
    ull2string(cid, sizeof(cid), server.repl_main_ch_client_id);

    *err = sendCommand(conn, "REPLCONF", "capa", "eof", "rdb-only", "1",
                       "rdb-channel", "1", "main-ch-client-id", cid,
                       "listening-port", buf, NULL);
    if (*err) {
        serverLog(LL_WARNING, "Error sending REPLCONF command to master in rdb channel handshake: %s", *err);
        return C_ERR;
    }

    if (connSetReadHandler(conn, rdbChannelFullSyncWithMaster) == C_ERR) {
        char conninfo[CONN_INFO_LEN];
        serverLog(LL_WARNING, "Can't create readable event for SYNC: %s (%s)",
                  strerror(errno), connGetInfo(conn, conninfo, sizeof(conninfo)));
        return C_ERR;
    }
    return C_OK;
}

/* Replication: Replica side. */
static int rdbChannelHandleAuthReply(connection *conn, sds *err) {
    *err = receiveSynchronousResponse(conn);
    if (*err == NULL) {
        serverLog(LL_WARNING, "Master did not respond to auth command during rdb channel handshake");
        return C_ERR;
    }
    if ((*err)[0] == '-') {
        serverLog(LL_WARNING, "Unable to AUTH to master: %s", *err);
        return C_ERR;
    }
    server.repl_rdb_ch_state = REPL_RDB_CH_RECEIVE_REPLCONF_REPLY;
    return C_OK;
}

/* Replication: Replica side. */
static int rdbChannelHandleReplconfReply(connection *conn, sds *err) {
    *err = receiveSynchronousResponse(conn);
    if (*err == NULL) {
        serverLog(LL_WARNING, "Master did not respond to replconf command during rdb channel handshake");
        return C_ERR;
    }
    if (*err[0] == '-') {
        serverLog(LL_WARNING, "Master replied error to replconf: %s", *err);
        return C_ERR;
    }
    sdsfree(*err);

    if (server.repl_debug_pause & REPL_DEBUG_BEFORE_RDB_CHANNEL)
        debugPauseProcess();

    /* Request rdb from master */
    *err = sendCommand(conn, "PSYNC", "?", "-1", NULL);
    if (*err) {
        serverLog(LL_WARNING, "I/O error writing to Master: %s", *err);
        return C_ERR;
    }

    return C_OK;
}

/* Replication: Replica side. */
static int rdbChannelHandleFullresyncReply(connection *conn, sds *err) {
    char *replid = NULL, *offset = NULL;

    *err = receiveSynchronousResponse(conn);
    if (*err == NULL)
        return C_ERR;

    if (*err[0] == '\0') {
        /* Retry again later */
        serverLog(LL_DEBUG, "Received empty psync reply");
        return C_RETRY;
    }

    /* FULL RESYNC, parse the reply in order to extract the replid
     * and the replication offset. */
    replid = strchr(*err,' ');
    if (replid) {
        replid++;
        offset = strchr(replid, ' ');
        if (offset) offset++;
    }
    if (!replid || !offset || (offset-replid-1) != CONFIG_RUN_ID_SIZE) {
        serverLog(LL_WARNING, "Received unexpected psync reply: %s", *err);
        return C_ERR;
    }
    memcpy(server.master_replid, replid, offset-replid-1);
    server.master_replid[CONFIG_RUN_ID_SIZE] = '\0';
    server.master_initial_offset = strtoll(offset,NULL,10);

    /* Prepare the main and rdb channels for rdb and repl stream delivery.*/
    server.repl_state = REPL_STATE_TRANSFER;
    rdbChannelReplDataBufInit();

    serverLog(LL_NOTICE, "Starting to receive RDB and replication stream in parallel.");

    /* RDB is still loading. Setup connection to accumulate repl data.  */
    if (connSetReadHandler(server.repl_transfer_s,
                           rdbChannelBufferReplData) != C_OK)
    {
        serverLog(LL_WARNING, "Can't set read handler for main channel: %s",
                  strerror(errno));
        return C_ERR;
    }

    /* Prepare RDB channel connection for RDB download. */
    if (connSetReadHandler(server.repl_rdb_transfer_s,
                           readSyncBulkPayload) != C_OK)
    {
        char inf[CONN_INFO_LEN];
        serverLog(LL_WARNING,
                  "Can't create readable event for rdb channel connection: %s (%s)",
                  strerror(errno),
                  connGetInfo(server.repl_rdb_transfer_s, inf, sizeof(inf)));
        return C_ERR;
    }

    return C_OK;
}

/* Replication: Replica side.
 * This connection handler is used to initialize the RDB channel connection.*/
static void rdbChannelFullSyncWithMaster(connection *conn) {
    int ret = 0;
    char *err = NULL;
    serverAssert(conn == server.repl_rdb_transfer_s);

    /* Check for errors in the socket: after a non blocking connect() we
     * may find that the socket is in error state. */
    if (connGetState(conn) != CONN_STATE_CONNECTED) {
        serverLog(LL_WARNING, "Error condition on socket for rdb channel replication: %s",
                  connGetLastError(conn));
        goto error;
    }
    switch (server.repl_rdb_ch_state) {
        case REPL_RDB_CH_SEND_HANDSHAKE:
            ret = rdbChannelSendHandshake(conn, &err);
            if (ret == C_OK)
                server.repl_rdb_ch_state = REPL_RDB_CH_RECEIVE_AUTH_REPLY;
            break;
        case REPL_RDB_CH_RECEIVE_AUTH_REPLY:
            if (server.masterauth) {
                ret = rdbChannelHandleAuthReply(conn, &err);
                if (ret == C_OK)
                    server.repl_rdb_ch_state = REPL_RDB_CH_RECEIVE_REPLCONF_REPLY;
                /* Wait for next bulk before trying to read replconf reply. */
                break;
            }
            server.repl_rdb_ch_state = REPL_RDB_CH_RECEIVE_REPLCONF_REPLY;
            /* fall through */
        case REPL_RDB_CH_RECEIVE_REPLCONF_REPLY:
            ret = rdbChannelHandleReplconfReply(conn, &err);
            if (ret == C_OK)
                server.repl_rdb_ch_state = REPL_RDB_CH_RECEIVE_FULLRESYNC;
            break;
        case REPL_RDB_CH_RECEIVE_FULLRESYNC:
            ret = rdbChannelHandleFullresyncReply(conn, &err);
            if (ret == C_OK)
                server.repl_rdb_ch_state = REPL_RDB_CH_RDB_LOADING;
            break;
        default:
            serverPanic("Unknown rdb channel state: %d", server.repl_rdb_ch_state);
    }

    if (ret == C_ERR)
        goto error;

    sdsfree(err);
    return;

error:
    if (err) {
        serverLog(LL_WARNING, "rdb channel sync failed with error: %s", err);
        sdsfree(err);
    }
    if (server.repl_transfer_s) {
        connClose(server.repl_transfer_s);
        server.repl_transfer_s = NULL;
    }
    server.repl_state = REPL_STATE_CONNECT;
    rdbChannelAbortRdbTransfer();
}

/* Replication: Replica side.
 * Initialize replica's local replication buffer to accumulate repl stream
 * during rdb channel sync. */
static void rdbChannelReplDataBufInit(void) {
    serverAssert(server.repl_full_sync_buffer.blocks == NULL);
    server.repl_full_sync_buffer.size = 0;
    server.repl_full_sync_buffer.used = 0;
    server.repl_full_sync_buffer.mem_used = 0;
    server.repl_full_sync_buffer.blocks = listCreate();
    server.repl_full_sync_buffer.blocks->free = zfree;
}

/* Replication: Replica side.
 * Free replica's local replication buffer */
static void rdbChannelReplDataBufFree(void) {
    listRelease(server.repl_full_sync_buffer.blocks);
    server.repl_full_sync_buffer.blocks = NULL;
    server.repl_full_sync_buffer.size = 0;
    server.repl_full_sync_buffer.used = 0;
    server.repl_full_sync_buffer.mem_used = 0;
}

/* Replication: Replica side.
 * Reads replication data from master connection into the repl buffer block */
int rdbChannelReadIntoBuf(connection *conn, replDataBufBlock *b) {
    atomicIncr(server.stat_io_reads_processed[IOTHREAD_MAIN_THREAD_ID], 1);

    int nread = connRead(conn, b->buf + b->used, b->size - b->used);
    if (nread <= 0) {
        if (nread == 0 || connGetState(conn) != CONN_STATE_CONNECTED) {
            serverLog(LL_WARNING, "Main channel error while reading from master: %s",
                      connGetLastError(conn));
            cancelReplicationHandshake(1);
        }
        return -1;
    }

    b->used += nread;
    server.repl_full_sync_buffer.used += nread;
    atomicIncr(server.stat_net_repl_input_bytes, nread);

    return nread;
}

/* Replication: Replica side.
 * Read handler for buffering incoming repl data during RDB download/loading. */
void rdbChannelBufferReplData(connection *conn) {
    const int buflen = 1024 * 1024;
    const int minread = 16 * 1024;
    int nread = 0;
    int needs_read = 1;

    listNode *ln = listLast(server.repl_full_sync_buffer.blocks);
    replDataBufBlock *tail = ln ? listNodeValue(ln) : NULL;

    /* Try to append last node. */
    if (tail && tail->size > tail->used) {
        nread = rdbChannelReadIntoBuf(conn, tail);
        if (nread <= 0)
            return;

        /* If buffer is filled fully, there might be more data in socket buffer.
         * Only read again if we've read small amount (less than minread). */
        needs_read = (tail->size == tail->used) && nread < minread;
    }

    if (needs_read) {
        unsigned long long limit;
        size_t usable_size;

        /* For accumulation limit, if 'replica-full-sync-buffer-limit' is set,
         * we'll use it. Otherwise, 'client-output-buffer-limit <replica>' is
         * the limit.*/
        limit = server.repl_full_sync_buffer_limit;
        if (limit == 0)
             limit = server.client_obuf_limits[CLIENT_TYPE_SLAVE].hard_limit_bytes;

        if (limit != 0 && server.repl_full_sync_buffer.size > limit) {
            serverLog(LL_NOTICE, "Replication buffer limit has been reached (%llu bytes), "
                                 "stopped buffering replication stream. Further accumulation may occur on master side. ", limit);
            connSetReadHandler(conn, NULL);
            return;
        }

        tail = zmalloc_usable(buflen, &usable_size);
        tail->size = usable_size - sizeof(replDataBufBlock);
        tail->used = 0;

        listAddNodeTail(server.repl_full_sync_buffer.blocks, tail);
        server.repl_full_sync_buffer.size += tail->size;
        server.repl_full_sync_buffer.mem_used += usable_size + sizeof(listNode);

        /* Update buffer's peak */
        if (server.repl_full_sync_buffer.peak < server.repl_full_sync_buffer.size)
            server.repl_full_sync_buffer.peak = server.repl_full_sync_buffer.size;

        rdbChannelReadIntoBuf(conn, tail);
    }
}

/* Replication: Replica side.
 * Streams accumulated replication data into the database. */
int rdbChannelStreamReplDataToDb(client *c) {
    int ret = C_OK;
    size_t size, used, offset = 0;
    listNode *n = NULL;
    replDataBufBlock *o;

    serverAssert(c->flags & CLIENT_MASTER);

    if (!server.repl_full_sync_buffer.blocks)
        return C_OK;

    blockingOperationStarts();
    protectClient(c);
    while ((n = listFirst(server.repl_full_sync_buffer.blocks))) {
        o = listNodeValue(n);
        size = o->size;
        used = o->used;
        c->querybuf = sdscatlen(c->querybuf, o->buf, used);
        c->read_reploff += (long long int) used;
        listDelNode(server.repl_full_sync_buffer.blocks, n);

        /* We don't expect error return value but just in case. */
        ret = processInputBuffer(c);
        if (ret != C_OK)
            break;

        server.repl_full_sync_buffer.used -= used;
        server.repl_full_sync_buffer.size -= size;
        server.repl_full_sync_buffer.mem_used -= (size + sizeof(listNode) +
                                                    sizeof(replDataBufBlock));
        if (server.repl_debug_pause & REPL_DEBUG_ON_STREAMING_REPL_BUF)
            debugPauseProcess();

        /* Check if we should yield back to the event loop */
        if (server.loading_process_events_interval_bytes &&
            ((offset + used) / server.loading_process_events_interval_bytes >
             offset / server.loading_process_events_interval_bytes))
        {
            replicationSendNewlineToMaster();
            processEventsWhileBlocked();
        }

        offset += used;
        /* Check if master client was freed in processEventsWhileBlocked().
         * It can happen if we receive 'replicaof' command or 'client kill'
         * command for the master. */
        if (c->flags & CLIENT_CLOSE_ASAP || !server.repl_full_sync_buffer.blocks) {
            ret = C_ERR;
            break;
        }
    }
    unprotectClient(c);
    blockingOperationEnds();

    if (ret != C_OK) {
        serverLog(LL_WARNING, "Master client was freed while streaming accumulated replication data to db.");
        return C_ERR;
    }

    return C_OK;
}

/* Replication: Replica side.
 * On rdb channel failure, close rdb-connection and reset state.
 * Return C_OK if cleanup is done. Otherwise, returns C_ERR which means cleanup
 * will be done asynchronously. */
static int rdbChannelAbortRdbTransfer(void) {
    if (server.repl_rdb_ch_state == REPL_RDB_CH_STATE_NONE)
        return C_OK;

    if (server.repl_rdb_transfer_s) {
        if (server.loading) {
            /* If loading flag is set, we want to handle cleanup asynchronously.
             * We set REPL_RDB_CH_STATE_CLOSE_ASAP so it will be handled just
             * outside loading loop.*/
            serverLog(LL_NOTICE, "Aborting rdb channel sync while loading the RDB.");

            if (disklessLoadingRio)
                /* Mark rio with abort flag, next rioRead() will return error.*/
                rioAbort(disklessLoadingRio);
            else {
                /* For disk based loading, we can wait until loading is done.
                 * This way, replica will have a chance for a successful psync
                 * later.*/
                serverLog(LL_NOTICE, "After loading RDB, replica will try psync with master.");
            }

            if (server.repl_transfer_s)
                connSetReadHandler(server.repl_transfer_s, NULL);

            server.repl_rdb_ch_state = REPL_RDB_CH_STATE_CLOSE_ASAP;
            return C_ERR;
        }
        connClose(server.repl_rdb_transfer_s);
        server.repl_rdb_transfer_s = NULL;
    }

    serverLog(LL_NOTICE, "Aborting rdb channel sync");

    if (server.repl_transfer_fd != -1) {
        close(server.repl_transfer_fd);
        server.repl_transfer_fd = -1;
    }
    if (server.repl_transfer_tmpfile) {
        bg_unlink(server.repl_transfer_tmpfile);
        zfree(server.repl_transfer_tmpfile);
        server.repl_transfer_tmpfile = NULL;
    }
    rdbChannelReplDataBufFree();
    server.repl_rdb_ch_state = REPL_RDB_CH_STATE_NONE;
    return C_OK;
}

/* Replica side. After loading the rdb, stream replication buffer to the db. */
static void rdbChannelSuccess(void) {
    serverLog(LL_NOTICE, "MASTER <-> REPLICA sync: Starting to stream replication buffer into the db"
                         " (%zu bytes).", server.repl_full_sync_buffer.used);

    if (rdbChannelStreamReplDataToDb(server.master) == C_ERR) {
        serverLog(LL_WARNING, "Failed to stream local replication buffer into the db");

        rdbChannelAbortRdbTransfer();
        if (server.master)
            freeClientAsync(server.master);
        return;
    }

    serverLog(LL_NOTICE, "MASTER <-> REPLICA sync: Successfully streamed replication buffer into the db");
    rdbChannelReplDataBufFree();
    server.repl_rdb_ch_state = REPL_RDB_CH_STATE_NONE;
}

void replicaofCommand(client *c) {
    /* SLAVEOF is not allowed in cluster mode as replication is automatically
     * configured using the current address of the master node. */
    if (server.cluster_enabled) {
        addReplyError(c,"REPLICAOF not allowed in cluster mode.");
        return;
    }

    if (server.failover_state != NO_FAILOVER) {
        addReplyError(c,"REPLICAOF not allowed while failing over.");
        return;
    }

    /* The special host/port combination "NO" "ONE" turns the instance
     * into a master. Otherwise the new master address is set. */
    if (!strcasecmp(c->argv[1]->ptr,"no") &&
        !strcasecmp(c->argv[2]->ptr,"one")) {
        if (server.masterhost) {
            replicationUnsetMaster();
            sds client = catClientInfoString(sdsempty(),c);
            serverLog(LL_NOTICE,"MASTER MODE enabled (user request from '%s')",
                client);
            sdsfree(client);
        }
    } else {
        long port;

        if (c->flags & CLIENT_SLAVE)
        {
            /* If a client is already a replica they cannot run this command,
             * because it involves flushing all replicas (including this
             * client) */
            addReplyError(c, "Command is not valid when client is a replica.");
            return;
        }

        if (getRangeLongFromObjectOrReply(c, c->argv[2], 0, 65535, &port,
                                          "Invalid master port") != C_OK)
            return;

        /* Check if we are already attached to the specified master */
        if (server.masterhost && !strcasecmp(server.masterhost,c->argv[1]->ptr)
            && server.masterport == port) {
            serverLog(LL_NOTICE,"REPLICAOF would result into synchronization "
                                "with the master we are already connected "
                                "with. No operation performed.");
            addReplySds(c,sdsnew("+OK Already connected to specified "
                                 "master\r\n"));
            return;
        }
        /* There was no previous master or the user specified a different one,
         * we can continue. */
        replicationSetMaster(c->argv[1]->ptr, port);
        sds client = catClientInfoString(sdsempty(),c);
        serverLog(LL_NOTICE,"REPLICAOF %s:%d enabled (user request from '%s')",
            server.masterhost, server.masterport, client);
        sdsfree(client);
    }
    addReply(c,shared.ok);
}

/* ROLE command: provide information about the role of the instance
 * (master or slave) and additional information related to replication
 * in an easy to process format. */
void roleCommand(client *c) {
    if (server.sentinel_mode) {
        sentinelRoleCommand(c);
        return;
    }

    if (server.masterhost == NULL) {
        listIter li;
        listNode *ln;
        void *mbcount;
        int slaves = 0;

        addReplyArrayLen(c,3);
        addReplyBulkCBuffer(c,"master",6);
        addReplyLongLong(c,server.master_repl_offset);
        mbcount = addReplyDeferredLen(c);
        listRewind(server.slaves,&li);
        while((ln = listNext(&li))) {
            client *slave = ln->value;
            char ip[NET_IP_STR_LEN], *slaveaddr = slave->slave_addr;

            if (!slaveaddr) {
                if (connAddrPeerName(slave->conn,ip,sizeof(ip),NULL) == -1)
                    continue;
                slaveaddr = ip;
            }
            if (slave->replstate != SLAVE_STATE_ONLINE) continue;
            addReplyArrayLen(c,3);
            addReplyBulkCString(c,slaveaddr);
            addReplyBulkLongLong(c,slave->slave_listening_port);
            addReplyBulkLongLong(c,slave->repl_ack_off);
            slaves++;
        }
        setDeferredArrayLen(c,mbcount,slaves);
    } else {
        char *slavestate = NULL;

        addReplyArrayLen(c,5);
        addReplyBulkCBuffer(c,"slave",5);
        addReplyBulkCString(c,server.masterhost);
        addReplyLongLong(c,server.masterport);
        if (slaveIsInHandshakeState()) {
            slavestate = "handshake";
        } else {
            switch(server.repl_state) {
            case REPL_STATE_NONE: slavestate = "none"; break;
            case REPL_STATE_CONNECT: slavestate = "connect"; break;
            case REPL_STATE_CONNECTING: slavestate = "connecting"; break;
            case REPL_STATE_TRANSFER: slavestate = "sync"; break;
            case REPL_STATE_CONNECTED: slavestate = "connected"; break;
            default: slavestate = "unknown"; break;
            }
        }
        addReplyBulkCString(c,slavestate);
        addReplyLongLong(c,server.master ? server.master->reploff : -1);
    }
}

/* Send a REPLCONF ACK command to the master to inform it about the current
 * processed offset. If we are not connected with a master, the command has
 * no effects. */
void replicationSendAck(void) {
    client *c = server.master;

    if (c != NULL) {
        int send_fack = server.fsynced_reploff != -1;
        c->flags |= CLIENT_MASTER_FORCE_REPLY;
        addReplyArrayLen(c,send_fack ? 5 : 3);
        addReplyBulkCString(c,"REPLCONF");
        addReplyBulkCString(c,"ACK");
        addReplyBulkLongLong(c,c->reploff);
        if (send_fack) {
            addReplyBulkCString(c,"FACK");
            addReplyBulkLongLong(c,server.fsynced_reploff);
        }
        c->flags &= ~CLIENT_MASTER_FORCE_REPLY;
    }
}

/* ---------------------- MASTER CACHING FOR PSYNC -------------------------- */

/* In order to implement partial synchronization we need to be able to cache
 * our master's client structure after a transient disconnection.
 * It is cached into server.cached_master and flushed away using the following
 * functions. */

/* This function is called by freeClient() in order to cache the master
 * client structure instead of destroying it. freeClient() will return
 * ASAP after this function returns, so every action needed to avoid problems
 * with a client that is really "suspended" has to be done by this function.
 *
 * The other functions that will deal with the cached master are:
 *
 * replicationDiscardCachedMaster() that will make sure to kill the client
 * as for some reason we don't want to use it in the future.
 *
 * replicationResurrectCachedMaster() that is used after a successful PSYNC
 * handshake in order to reactivate the cached master.
 */
void replicationCacheMaster(client *c) {
    serverAssert(server.master != NULL && server.cached_master == NULL);
    serverLog(LL_NOTICE,"Caching the disconnected master state.");

    /* Unlink the client from the server structures. */
    unlinkClient(c);

    /* Reset the master client so that's ready to accept new commands:
     * we want to discard the non processed query buffers and non processed
     * offsets, including pending transactions, already populated arguments,
     * pending outputs to the master. */
    sdsclear(server.master->querybuf);
    server.master->qb_pos = 0;
    server.master->repl_applied = 0;
    server.master->read_reploff = server.master->reploff;
    if (c->flags & CLIENT_MULTI) discardTransaction(c);
    listEmpty(c->reply);
    c->sentlen = 0;
    c->reply_bytes = 0;
    c->bufpos = 0;
    resetClient(c);

    /* Save the master. Server.master will be set to null later by
     * replicationHandleMasterDisconnection(). */
    server.cached_master = server.master;

    /* Invalidate the Peer ID cache. */
    if (c->peerid) {
        sdsfree(c->peerid);
        c->peerid = NULL;
    }
    /* Invalidate the Sock Name cache. */
    if (c->sockname) {
        sdsfree(c->sockname);
        c->sockname = NULL;
    }

    /* Caching the master happens instead of the actual freeClient() call,
     * so make sure to adjust the replication state. This function will
     * also set server.master to NULL. */
    replicationHandleMasterDisconnection();
}

/* This function is called when a master is turned into a slave, in order to
 * create from scratch a cached master for the new client, that will allow
 * to PSYNC with the slave that was promoted as the new master after a
 * failover.
 *
 * Assuming this instance was previously the master instance of the new master,
 * the new master will accept its replication ID, and potential also the
 * current offset if no data was lost during the failover. So we use our
 * current replication ID and offset in order to synthesize a cached master. */
void replicationCacheMasterUsingMyself(void) {
    serverLog(LL_NOTICE,
        "Before turning into a replica, using my own master parameters "
        "to synthesize a cached master: I may be able to synchronize with "
        "the new master with just a partial transfer.");

    /* This will be used to populate the field server.master->reploff
     * by replicationCreateMasterClient(). We'll later set the created
     * master as server.cached_master, so the replica will use such
     * offset for PSYNC. */
    server.master_initial_offset = server.master_repl_offset;

    /* The master client we create can be set to any DBID, because
     * the new master will start its replication stream with SELECT. */
    replicationCreateMasterClient(NULL,-1);

    /* Use our own ID / offset. */
    memcpy(server.master->replid, server.replid, sizeof(server.replid));

    /* Set as cached master. */
    unlinkClient(server.master);
    server.cached_master = server.master;
    server.master = NULL;
}

/* Free a cached master, called when there are no longer the conditions for
 * a partial resync on reconnection. */
void replicationDiscardCachedMaster(void) {
    if (server.cached_master == NULL) return;

    serverLog(LL_NOTICE,"Discarding previously cached master state.");
    server.cached_master->flags &= ~CLIENT_MASTER;
    freeClient(server.cached_master);
    server.cached_master = NULL;
}

/* Turn the cached master into the current master, using the file descriptor
 * passed as argument as the socket for the new master.
 *
 * This function is called when successfully setup a partial resynchronization
 * so the stream of data that we'll receive will start from where this
 * master left. */
void replicationResurrectCachedMaster(connection *conn) {
    server.master = server.cached_master;
    server.cached_master = NULL;
    server.master->conn = conn;
    connSetPrivateData(server.master->conn, server.master);
    server.master->flags &= ~(CLIENT_CLOSE_AFTER_REPLY|CLIENT_CLOSE_ASAP);
    server.master->authenticated = 1;
    server.master->lastinteraction = server.unixtime;
    server.repl_state = REPL_STATE_CONNECTED;
    server.repl_down_since = 0;

    /* Fire the master link modules event. */
    moduleFireServerEvent(REDISMODULE_EVENT_MASTER_LINK_CHANGE,
                          REDISMODULE_SUBEVENT_MASTER_LINK_UP,
                          NULL);

    /* Re-add to the list of clients. */
    linkClient(server.master);
    if (connSetReadHandler(server.master->conn, readQueryFromClient)) {
        serverLog(LL_WARNING,"Error resurrecting the cached master, impossible to add the readable handler: %s", strerror(errno));
        freeClientAsync(server.master); /* Close ASAP. */
    }

    /* We may also need to install the write handler as well if there is
     * pending data in the write buffers. */
    if (clientHasPendingReplies(server.master)) {
        if (connSetWriteHandler(server.master->conn, sendReplyToClient)) {
            serverLog(LL_WARNING,"Error resurrecting the cached master, impossible to add the writable handler: %s", strerror(errno));
            freeClientAsync(server.master); /* Close ASAP. */
        }
    }
}

/* ------------------------- MIN-SLAVES-TO-WRITE  --------------------------- */

/* This function counts the number of slaves with lag <= min-slaves-max-lag.
 * If the option is active, the server will prevent writes if there are not
 * enough connected slaves with the specified lag (or less). */
void refreshGoodSlavesCount(void) {
    listIter li;
    listNode *ln;
    int good = 0;

    if (!server.repl_min_slaves_to_write ||
        !server.repl_min_slaves_max_lag) return;

    listRewind(server.slaves,&li);
    while((ln = listNext(&li))) {
        client *slave = ln->value;
        time_t lag = server.unixtime - slave->repl_ack_time;

        if (slave->replstate == SLAVE_STATE_ONLINE &&
            lag <= server.repl_min_slaves_max_lag) good++;
    }
    server.repl_good_slaves_count = good;
}

/* return true if status of good replicas is OK. otherwise false */
int checkGoodReplicasStatus(void) {
    return server.masterhost || /* not a primary status should be OK */
           !server.repl_min_slaves_max_lag || /* Min slave max lag not configured */
           !server.repl_min_slaves_to_write || /* Min slave to write not configured */
           server.repl_good_slaves_count >= server.repl_min_slaves_to_write; /* check if we have enough slaves */
}

/* ----------------------- SYNCHRONOUS REPLICATION --------------------------
 * Redis synchronous replication design can be summarized in points:
 *
 * - Redis masters have a global replication offset, used by PSYNC.
 * - Master increment the offset every time new commands are sent to slaves.
 * - Slaves ping back masters with the offset processed so far.
 *
 * So synchronous replication adds a new WAIT command in the form:
 *
 *   WAIT <num_replicas> <milliseconds_timeout>
 *
 * That returns the number of replicas that processed the query when
 * we finally have at least num_replicas, or when the timeout was
 * reached.
 *
 * The command is implemented in this way:
 *
 * - Every time a client processes a command, we remember the replication
 *   offset after sending that command to the slaves.
 * - When WAIT is called, we ask slaves to send an acknowledgement ASAP.
 *   The client is blocked at the same time (see blocked.c).
 * - Once we receive enough ACKs for a given offset or when the timeout
 *   is reached, the WAIT command is unblocked and the reply sent to the
 *   client.
 */

/* This just set a flag so that we broadcast a REPLCONF GETACK command
 * to all the slaves in the beforeSleep() function. Note that this way
 * we "group" all the clients that want to wait for synchronous replication
 * in a given event loop iteration, and send a single GETACK for them all. */
void replicationRequestAckFromSlaves(void) {
    server.get_ack_from_slaves = 1;
}

/* Return the number of slaves that already acknowledged the specified
 * replication offset. */
int replicationCountAcksByOffset(long long offset) {
    listIter li;
    listNode *ln;
    int count = 0;

    listRewind(server.slaves,&li);
    while((ln = listNext(&li))) {
        client *slave = ln->value;

        if (slave->replstate != SLAVE_STATE_ONLINE) continue;
        if (slave->repl_ack_off >= offset) count++;
    }
    return count;
}

/* Return the number of replicas that already acknowledged the specified
 * replication offset being AOF fsynced. */
int replicationCountAOFAcksByOffset(long long offset) {
    listIter li;
    listNode *ln;
    int count = 0;

    listRewind(server.slaves,&li);
    while((ln = listNext(&li))) {
        client *slave = ln->value;

        if (slave->replstate != SLAVE_STATE_ONLINE) continue;
        if (slave->repl_aof_off >= offset) count++;
    }
    return count;
}

/* WAIT for N replicas to acknowledge the processing of our latest
 * write command (and all the previous commands). */
void waitCommand(client *c) {
    mstime_t timeout;
    long numreplicas, ackreplicas;
    long long offset = c->woff;

    if (server.masterhost) {
        addReplyError(c,"WAIT cannot be used with replica instances. Please also note that since Redis 4.0 if a replica is configured to be writable (which is not the default) writes to replicas are just local and are not propagated.");
        return;
    }

    /* Argument parsing. */
    if (getLongFromObjectOrReply(c,c->argv[1],&numreplicas,NULL) != C_OK)
        return;
    if (getTimeoutFromObjectOrReply(c,c->argv[2],&timeout,UNIT_MILLISECONDS)
        != C_OK) return;

    /* First try without blocking at all. */
    ackreplicas = replicationCountAcksByOffset(c->woff);
    if (ackreplicas >= numreplicas || c->flags & CLIENT_DENY_BLOCKING) {
        addReplyLongLong(c,ackreplicas);
        return;
    }

    /* Otherwise block the client and put it into our list of clients
     * waiting for ack from slaves. */
    blockForReplication(c,timeout,offset,numreplicas);

    /* Make sure that the server will send an ACK request to all the slaves
     * before returning to the event loop. */
    replicationRequestAckFromSlaves();
}

/* WAIT for N replicas and / or local master to acknowledge our latest
 * write command got synced to the disk. */
void waitaofCommand(client *c) {
    mstime_t timeout;
    long numreplicas, numlocal, ackreplicas, acklocal;

    /* Argument parsing. */
    if (getRangeLongFromObjectOrReply(c,c->argv[1],0,1,&numlocal,NULL) != C_OK)
        return;
    if (getPositiveLongFromObjectOrReply(c,c->argv[2],&numreplicas,NULL) != C_OK)
        return;
    if (getTimeoutFromObjectOrReply(c,c->argv[3],&timeout,UNIT_MILLISECONDS) != C_OK)
        return;

    if (server.masterhost) {
        addReplyError(c,"WAITAOF cannot be used with replica instances. Please also note that writes to replicas are just local and are not propagated.");
        return;
    }
    if (numlocal && !server.aof_enabled) {
        addReplyError(c, "WAITAOF cannot be used when numlocal is set but appendonly is disabled.");
        return;
    }

    /* First try without blocking at all. */
    ackreplicas = replicationCountAOFAcksByOffset(c->woff);
    acklocal = server.fsynced_reploff >= c->woff;
    if ((ackreplicas >= numreplicas && acklocal >= numlocal) || c->flags & CLIENT_DENY_BLOCKING) {
        addReplyArrayLen(c,2);
        addReplyLongLong(c,acklocal);
        addReplyLongLong(c,ackreplicas);
        return;
    }

    /* Otherwise block the client and put it into our list of clients
     * waiting for ack from slaves. */
    blockForAofFsync(c,timeout,c->woff,numlocal,numreplicas);

    /* Make sure that the server will send an ACK request to all the slaves
     * before returning to the event loop. */
    replicationRequestAckFromSlaves();
}

/* This is called by unblockClient() to perform the blocking op type
 * specific cleanup. We just remove the client from the list of clients
 * waiting for replica acks. Never call it directly, call unblockClient()
 * instead. */
void unblockClientWaitingReplicas(client *c) {
    listNode *ln = listSearchKey(server.clients_waiting_acks,c);
    serverAssert(ln != NULL);
    listDelNode(server.clients_waiting_acks,ln);
    updateStatsOnUnblock(c, 0, 0, 0);
}

/* Check if there are clients blocked in WAIT or WAITAOF that can be unblocked
 * since we received enough ACKs from slaves. */
void processClientsWaitingReplicas(void) {
    long long last_offset = 0;
    long long last_aof_offset = 0;
    int last_numreplicas = 0;
    int last_aof_numreplicas = 0;

    listIter li;
    listNode *ln;

    listRewind(server.clients_waiting_acks,&li);
    while((ln = listNext(&li))) {
        int numlocal = 0;
        int numreplicas = 0;

        client *c = ln->value;
        int is_wait_aof = c->bstate.btype == BLOCKED_WAITAOF;

        if (is_wait_aof && c->bstate.numlocal && !server.aof_enabled) {
            addReplyError(c, "WAITAOF cannot be used when numlocal is set but appendonly is disabled.");
            unblockClient(c, 1);
            continue;
        }

        /* Every time we find a client that is satisfied for a given
         * offset and number of replicas, we remember it so the next client
         * may be unblocked without calling replicationCountAcksByOffset()
         * or calling replicationCountAOFAcksByOffset()
         * if the requested offset / replicas were equal or less. */
        if (!is_wait_aof && last_offset && last_offset >= c->bstate.reploffset &&
                           last_numreplicas >= c->bstate.numreplicas)
        {
            numreplicas = last_numreplicas;
        } else if (is_wait_aof && last_aof_offset && last_aof_offset >= c->bstate.reploffset &&
                    last_aof_numreplicas >= c->bstate.numreplicas)
        {
            numreplicas = last_aof_numreplicas;
        } else {
            numreplicas = is_wait_aof ?
                replicationCountAOFAcksByOffset(c->bstate.reploffset) :
                replicationCountAcksByOffset(c->bstate.reploffset);

            /* Check if the number of replicas is satisfied. */
            if (numreplicas < c->bstate.numreplicas) continue;

            if (is_wait_aof) {
                last_aof_offset = c->bstate.reploffset;
                last_aof_numreplicas = numreplicas;
            } else {
                last_offset = c->bstate.reploffset;
                last_numreplicas = numreplicas;
            }
        }

        /* Check if the local constraint of WAITAOF is served */
        if (is_wait_aof) {
            numlocal = server.fsynced_reploff >= c->bstate.reploffset;
            if (numlocal < c->bstate.numlocal) continue;
        }

        /* Reply before unblocking, because unblock client calls reqresAppendResponse */
        if (is_wait_aof) {
            /* WAITAOF has an array reply */
            addReplyArrayLen(c, 2);
            addReplyLongLong(c, numlocal);
            addReplyLongLong(c, numreplicas);
        } else {
            addReplyLongLong(c, numreplicas);
        }

        unblockClient(c, 1);
    }
}

/* Return the slave replication offset for this instance, that is
 * the offset for which we already processed the master replication stream. */
long long replicationGetSlaveOffset(void) {
    long long offset = 0;

    if (server.masterhost != NULL) {
        if (server.master) {
            offset = server.master->reploff;
        } else if (server.cached_master) {
            offset = server.cached_master->reploff;
        }
    }
    /* offset may be -1 when the master does not support it at all, however
     * this function is designed to return an offset that can express the
     * amount of data processed by the master, so we return a positive
     * integer. */
    if (offset < 0) offset = 0;
    return offset;
}

/* --------------------------- REPLICATION CRON  ---------------------------- */

/* Replication cron function, called 1 time per second. */
void replicationCron(void) {
    /* Check failover status first, to see if we need to start
     * handling the failover. */
    updateFailoverStatus();

    /* Non blocking connection timeout? */
    if (server.masterhost &&
        (server.repl_state == REPL_STATE_CONNECTING ||
         slaveIsInHandshakeState()) &&
         (time(NULL)-server.repl_transfer_lastio) > server.repl_timeout)
    {
        serverLog(LL_WARNING,"Timeout connecting to the MASTER...");
        cancelReplicationHandshake(1);
    }

    /* Bulk transfer I/O timeout? */
    if (server.masterhost && server.repl_state == REPL_STATE_TRANSFER &&
        (time(NULL)-server.repl_transfer_lastio) > server.repl_timeout)
    {
        serverLog(LL_WARNING,"Timeout receiving bulk data from MASTER... If the problem persists try to set the 'repl-timeout' parameter in redis.conf to a larger value.");
        cancelReplicationHandshake(1);
    }

    /* Timed out master when we are an already connected slave? */
    if (server.masterhost && server.repl_state == REPL_STATE_CONNECTED &&
        (time(NULL)-server.master->lastinteraction) > server.repl_timeout)
    {
        serverLog(LL_WARNING,"MASTER timeout: no data nor PING received...");
        freeClient(server.master);
    }

    /* Check if we should connect to a MASTER */
    if (server.repl_state == REPL_STATE_CONNECT) {
        serverLog(LL_NOTICE,"Connecting to MASTER %s:%d",
            server.masterhost, server.masterport);
        connectWithMaster();
    }

    /* Send ACK to master from time to time.
     * Note that we do not send periodic acks to masters that don't
     * support PSYNC and replication offsets. */
    if (server.masterhost && server.master &&
        !(server.master->flags & CLIENT_PRE_PSYNC))
        replicationSendAck();

    /* If we have attached slaves, PING them from time to time.
     * So slaves can implement an explicit timeout to masters, and will
     * be able to detect a link disconnection even if the TCP connection
     * will not actually go down. */
    listIter li;
    listNode *ln;
    robj *ping_argv[1];

    /* First, send PING according to ping_slave_period. The reason why master
     * sends PING is to keep the connection with replica active, so master need
     * not send PING to replicas if already sent replication stream in the past
     * repl_ping_slave_period time. */
    if (server.masterhost == NULL && listLength(server.slaves) &&
        server.unixtime >= server.repl_stream_lastio + server.repl_ping_slave_period)
    {
        /* Note that we don't send the PING if the clients are paused during
         * a Redis Cluster manual failover: the PING we send will otherwise
         * alter the replication offsets of master and slave, and will no longer
         * match the one stored into 'mf_master_offset' state. */
        int manual_failover_in_progress =
            ((server.cluster_enabled &&
              clusterManualFailoverTimeLimit()) ||
            server.failover_end_time) &&
            isPausedActionsWithUpdate(PAUSE_ACTION_REPLICA);

        if (!manual_failover_in_progress) {
            ping_argv[0] = shared.ping;
            replicationFeedSlaves(server.slaves, -1,
                ping_argv, 1);
        }
    }

    /* Second, send a newline to all the slaves in pre-synchronization
     * stage, that is, slaves waiting for the master to create the RDB file.
     *
     * Also send the a newline to all the chained slaves we have, if we lost
     * connection from our master, to keep the slaves aware that their
     * master is online. This is needed since sub-slaves only receive proxied
     * data from top-level masters, so there is no explicit pinging in order
     * to avoid altering the replication offsets. This special out of band
     * pings (newlines) can be sent, they will have no effect in the offset.
     *
     * The newline will be ignored by the slave but will refresh the
     * last interaction timer preventing a timeout. In this case we ignore the
     * ping period and refresh the connection once per second since certain
     * timeouts are set at a few seconds (example: PSYNC response). */
    listRewind(server.slaves,&li);
    while((ln = listNext(&li))) {
        client *slave = ln->value;

        int is_presync =
            (slave->replstate == SLAVE_STATE_WAIT_BGSAVE_START ||
            (slave->replstate == SLAVE_STATE_WAIT_BGSAVE_END &&
             server.rdb_child_type != RDB_CHILD_TYPE_SOCKET));

        if (is_presync && !(slave->flags & CLIENT_CLOSE_ASAP)) {
            connWrite(slave->conn, "\n", 1);
        }
    }

    /* Disconnect timedout slaves. */
    if (listLength(server.slaves)) {
        listIter li;
        listNode *ln;

        listRewind(server.slaves,&li);
        while((ln = listNext(&li))) {
            client *slave = ln->value;

            if (slave->replstate == SLAVE_STATE_ONLINE) {
                if (slave->flags & CLIENT_PRE_PSYNC)
                    continue;
                if ((server.unixtime - slave->repl_ack_time) > server.repl_timeout) {
                    serverLog(LL_WARNING, "Disconnecting timedout replica (streaming sync): %s",
                          replicationGetSlaveName(slave));
                    freeClient(slave);
                    continue;
                }
            }
            /* We consider disconnecting only diskless replicas because disk-based replicas aren't fed
             * by the fork child so if a disk-based replica is stuck it doesn't prevent the fork child
             * from terminating. */
            if (slave->replstate == SLAVE_STATE_WAIT_BGSAVE_END && server.rdb_child_type == RDB_CHILD_TYPE_SOCKET) {
                if (slave->repl_last_partial_write != 0 &&
                    (server.unixtime - slave->repl_last_partial_write) > server.repl_timeout)
                {
                    serverLog(LL_WARNING, "Disconnecting timedout replica (full sync): %s",
                          replicationGetSlaveName(slave));
                    freeClient(slave);
                    continue;
                }
            }
        }
    }

    /* If this is a master without attached slaves and there is a replication
     * backlog active, in order to reclaim memory we can free it after some
     * (configured) time. Note that this cannot be done for slaves: slaves
     * without sub-slaves attached should still accumulate data into the
     * backlog, in order to reply to PSYNC queries if they are turned into
     * masters after a failover. */
    if (listLength(server.slaves) == 0 && server.repl_backlog_time_limit &&
        server.repl_backlog && server.masterhost == NULL)
    {
        time_t idle = server.unixtime - server.repl_no_slaves_since;

        if (idle > server.repl_backlog_time_limit) {
            /* When we free the backlog, we always use a new
             * replication ID and clear the ID2. This is needed
             * because when there is no backlog, the master_repl_offset
             * is not updated, but we would still retain our replication
             * ID, leading to the following problem:
             *
             * 1. We are a master instance.
             * 2. Our slave is promoted to master. It's repl-id-2 will
             *    be the same as our repl-id.
             * 3. We, yet as master, receive some updates, that will not
             *    increment the master_repl_offset.
             * 4. Later we are turned into a slave, connect to the new
             *    master that will accept our PSYNC request by second
             *    replication ID, but there will be data inconsistency
             *    because we received writes. */
            changeReplicationId();
            clearReplicationId2();
            freeReplicationBacklog();
            serverLog(LL_NOTICE,
                "Replication backlog freed after %d seconds "
                "without connected replicas.",
                (int) server.repl_backlog_time_limit);
        }
    }

    replicationStartPendingFork();

    /* Remove the RDB file used for replication if Redis is not running
     * with any persistence. */
    removeRDBUsedToSyncReplicas();

    /* Sanity check replication buffer, the first block of replication buffer blocks
     * must be referenced by someone, since it will be freed when not referenced,
     * otherwise, server will OOM. also, its refcount must not be more than
     * replicas number + 1(replication backlog). */
    if (listLength(server.repl_buffer_blocks) > 0) {
        replBufBlock *o = listNodeValue(listFirst(server.repl_buffer_blocks));
        serverAssert(o->refcount > 0 &&
            o->refcount <= (int)listLength(server.slaves)+1);
    }

    /* Refresh the number of slaves with lag <= min-slaves-max-lag. */
    refreshGoodSlavesCount();
}

int shouldStartChildReplication(int *mincapa_out, int *req_out) {
    /* We should start a BGSAVE good for replication if we have slaves in
     * WAIT_BGSAVE_START state.
     *
     * In case of diskless replication, we make sure to wait the specified
     * number of seconds (according to configuration) so that other slaves
     * have the time to arrive before we start streaming. */
    if (!hasActiveChildProcess()) {
        time_t idle, max_idle = 0;
        int slaves_waiting = 0;
        int mincapa;
        int req;
        int first = 1;
        listNode *ln;
        listIter li;

        listRewind(server.slaves,&li);
        while((ln = listNext(&li))) {
            client *slave = ln->value;
            if (slave->replstate == SLAVE_STATE_WAIT_BGSAVE_START) {
                if (first) {
                    /* Get first slave's requirements */
                    req = slave->slave_req;
                } else if (req != slave->slave_req) {
                    /* Skip slaves that don't match */
                    continue;
                }
                idle = server.unixtime - slave->lastinteraction;
                if (idle > max_idle) max_idle = idle;
                slaves_waiting++;
                mincapa = first ? slave->slave_capa : (mincapa & slave->slave_capa);
                first = 0;
            }
        }

        if (slaves_waiting &&
            (!server.repl_diskless_sync ||
             (server.repl_diskless_sync_max_replicas > 0 &&
              slaves_waiting >= server.repl_diskless_sync_max_replicas) ||
             max_idle >= server.repl_diskless_sync_delay))
        {
            if (mincapa_out)
                *mincapa_out = mincapa;
            if (req_out)
                *req_out = req;
            return 1;
        }
    }

    return 0;
}

void replicationStartPendingFork(void) {
    int mincapa = -1;
    int req = -1;

    if (shouldStartChildReplication(&mincapa, &req)) {
        /* Start the BGSAVE. The called function may start a
         * BGSAVE with socket target or disk target depending on the
         * configuration and slaves capabilities and requirements. */
        startBgsaveForReplication(mincapa, req);
    }
}

/* Find replica at IP:PORT from replica list */
static client *findReplica(char *host, int port) {
    listIter li;
    listNode *ln;
    client *replica;

    listRewind(server.slaves,&li);
    while((ln = listNext(&li))) {
        replica = ln->value;
        char ip[NET_IP_STR_LEN], *replicaip = replica->slave_addr;

        if (!replicaip) {
            if (connAddrPeerName(replica->conn, ip, sizeof(ip), NULL) == -1)
                continue;
            replicaip = ip;
        }

        if (!strcasecmp(host, replicaip) &&
                (port == replica->slave_listening_port))
            return replica;
    }

    return NULL;
}

const char *getFailoverStateString(void) {
    switch(server.failover_state) {
        case NO_FAILOVER: return "no-failover";
        case FAILOVER_IN_PROGRESS: return "failover-in-progress";
        case FAILOVER_WAIT_FOR_SYNC: return "waiting-for-sync";
        default: return "unknown";
    }
}

/* Resets the internal failover configuration, this needs
 * to be called after a failover either succeeds or fails
 * as it includes the client unpause. */
void clearFailoverState(void) {
    server.failover_end_time = 0;
    server.force_failover = 0;
    zfree(server.target_replica_host);
    server.target_replica_host = NULL;
    server.target_replica_port = 0;
    server.failover_state = NO_FAILOVER;
    unpauseActions(PAUSE_DURING_FAILOVER);
}

/* Abort an ongoing failover if one is going on. */
void abortFailover(const char *err) {
    if (server.failover_state == NO_FAILOVER) return;

    if (server.target_replica_host) {
        serverLog(LL_NOTICE,"FAILOVER to %s:%d aborted: %s",
            server.target_replica_host,server.target_replica_port,err);  
    } else {
        serverLog(LL_NOTICE,"FAILOVER to any replica aborted: %s",err);  
    }
    if (server.failover_state == FAILOVER_IN_PROGRESS) {
        replicationUnsetMaster();
    }
    clearFailoverState();
}

/* 
 * FAILOVER [TO <HOST> <PORT> [FORCE]] [ABORT] [TIMEOUT <timeout>]
 * 
 * This command will coordinate a failover between the master and one
 * of its replicas. The happy path contains the following steps:
 * 1) The master will initiate a client pause write, to stop replication
 * traffic.
 * 2) The master will periodically check if any of its replicas has
 * consumed the entire replication stream through acks. 
 * 3) Once any replica has caught up, the master will itself become a replica.
 * 4) The master will send a PSYNC FAILOVER request to the target replica, which
 * if accepted will cause the replica to become the new master and start a sync.
 * 
 * FAILOVER ABORT is the only way to abort a failover command, as replicaof
 * will be disabled. This may be needed if the failover is unable to progress. 
 * 
 * The optional arguments [TO <HOST> <IP>] allows designating a specific replica
 * to be failed over to.
 * 
 * FORCE flag indicates that even if the target replica is not caught up,
 * failover to it anyway. This must be specified with a timeout and a target
 * HOST and IP.
 * 
 * TIMEOUT <timeout> indicates how long should the primary wait for 
 * a replica to sync up before aborting. If not specified, the failover
 * will attempt forever and must be manually aborted.
 */
void failoverCommand(client *c) {
    if (!clusterAllowFailoverCmd(c)) {
        return;
    }

    /* Handle special case for abort */
    if ((c->argc == 2) && !strcasecmp(c->argv[1]->ptr,"abort")) {
        if (server.failover_state == NO_FAILOVER) {
            addReplyError(c, "No failover in progress.");
            return;
        }

        abortFailover("Failover manually aborted");
        addReply(c,shared.ok);
        return;
    }

    long timeout_in_ms = 0;
    int force_flag = 0;
    long port = 0;
    char *host = NULL;

    /* Parse the command for syntax and arguments. */
    for (int j = 1; j < c->argc; j++) {
        if (!strcasecmp(c->argv[j]->ptr,"timeout") && (j + 1 < c->argc) &&
            timeout_in_ms == 0)
        {
            if (getLongFromObjectOrReply(c,c->argv[j + 1],
                        &timeout_in_ms,NULL) != C_OK) return;
            if (timeout_in_ms <= 0) {
                addReplyError(c,"FAILOVER timeout must be greater than 0");
                return;
            }
            j++;
        } else if (!strcasecmp(c->argv[j]->ptr,"to") && (j + 2 < c->argc) &&
            !host) 
        {
            if (getLongFromObjectOrReply(c,c->argv[j + 2],&port,NULL) != C_OK)
                return;
            host = c->argv[j + 1]->ptr;
            j += 2;
        } else if (!strcasecmp(c->argv[j]->ptr,"force") && !force_flag) {
            force_flag = 1;
        } else {
            addReplyErrorObject(c,shared.syntaxerr);
            return;
        }
    }

    if (server.failover_state != NO_FAILOVER) {
        addReplyError(c,"FAILOVER already in progress.");
        return;
    }

    if (server.masterhost) {
        addReplyError(c,"FAILOVER is not valid when server is a replica.");
        return;
    }

    if (listLength(server.slaves) == 0) {
        addReplyError(c,"FAILOVER requires connected replicas.");
        return; 
    }

    if (force_flag && (!timeout_in_ms || !host)) {
        addReplyError(c,"FAILOVER with force option requires both a timeout "
            "and target HOST and IP.");
        return;     
    }

    /* If a replica address was provided, validate that it is connected. */
    if (host) {
        client *replica = findReplica(host, port);

        if (replica == NULL) {
            addReplyError(c,"FAILOVER target HOST and PORT is not "
                            "a replica.");
            return;
        }

        /* Check if requested replica is online */
        if (replica->replstate != SLAVE_STATE_ONLINE) {
            addReplyError(c,"FAILOVER target replica is not online.");
            return;
        }

        server.target_replica_host = zstrdup(host);
        server.target_replica_port = port;
        serverLog(LL_NOTICE,"FAILOVER requested to %s:%ld.",host,port);
    } else {
        serverLog(LL_NOTICE,"FAILOVER requested to any replica.");
    }

    mstime_t now = commandTimeSnapshot();
    if (timeout_in_ms) {
        server.failover_end_time = now + timeout_in_ms;
    }
    
    server.force_failover = force_flag;
    server.failover_state = FAILOVER_WAIT_FOR_SYNC;
    /* Cluster failover will unpause eventually */
    pauseActions(PAUSE_DURING_FAILOVER,
                 LLONG_MAX,
                 PAUSE_ACTIONS_CLIENT_WRITE_SET);
    addReply(c,shared.ok);
}

/* Failover cron function, checks coordinated failover state. 
 *
 * Implementation note: The current implementation calls replicationSetMaster()
 * to start the failover request, this has some unintended side effects if the
 * failover doesn't work like blocked clients will be unblocked and replicas will
 * be disconnected. This could be optimized further.
 */
void updateFailoverStatus(void) {
    if (server.failover_state != FAILOVER_WAIT_FOR_SYNC) return;
    mstime_t now = server.mstime;

    /* Check if failover operation has timed out */
    if (server.failover_end_time && server.failover_end_time <= now) {
        if (server.force_failover) {
            serverLog(LL_NOTICE,
                "FAILOVER to %s:%d time out exceeded, failing over.",
                server.target_replica_host, server.target_replica_port);
            server.failover_state = FAILOVER_IN_PROGRESS;
            /* If timeout has expired force a failover if requested. */
            replicationSetMaster(server.target_replica_host,
                server.target_replica_port);
            return;
        } else {
            /* Force was not requested, so timeout. */
            abortFailover("Replica never caught up before timeout");
            return;
        }
    }

    /* Check to see if the replica has caught up so failover can start */
    client *replica = NULL;
    if (server.target_replica_host) {
        replica = findReplica(server.target_replica_host, 
            server.target_replica_port);
    } else {
        listIter li;
        listNode *ln;

        listRewind(server.slaves,&li);
        /* Find any replica that has matched our repl_offset */
        while((ln = listNext(&li))) {
            replica = ln->value;
            if (replica->repl_ack_off == server.master_repl_offset) {
                char ip[NET_IP_STR_LEN], *replicaaddr = replica->slave_addr;

                if (!replicaaddr) {
                    if (connAddrPeerName(replica->conn,ip,sizeof(ip),NULL) == -1)
                        continue;
                    replicaaddr = ip;
                }

                /* We are now failing over to this specific node */
                server.target_replica_host = zstrdup(replicaaddr);
                server.target_replica_port = replica->slave_listening_port;
                break;
            }
        }
    }

    /* We've found a replica that is caught up */
    if (replica && (replica->repl_ack_off == server.master_repl_offset)) {
        server.failover_state = FAILOVER_IN_PROGRESS;
        serverLog(LL_NOTICE,
                "Failover target %s:%d is synced, failing over.",
                server.target_replica_host, server.target_replica_port);
        /* Designated replica is caught up, failover to it. */
        replicationSetMaster(server.target_replica_host,
            server.target_replica_port);
    }
}