Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

HDDS-8888. Consider Datanode queue capacity when sending DeleteBlocks command #4939

Merged
merged 9 commits into from
Jan 6, 2024
Original file line number Diff line number Diff line change
Expand Up @@ -257,7 +257,17 @@ public final class OzoneConfigKeys {
public static final String OZONE_BLOCK_DELETING_SERVICE_TIMEOUT =
"ozone.block.deleting.service.timeout";
public static final String OZONE_BLOCK_DELETING_SERVICE_TIMEOUT_DEFAULT
= "300s"; // 300s for default
= "300s"; // 300s for default deleteBlocksCommandLimit

/**
* A limit to restrict the total number of delete block commands queued on a
* datanode. Note this is intended to be a temporary config until we have a
* more dynamic way of limiting load.
*/
public static final String OZONE_BLOCK_DELETING_PENDING_COMMAND_LIMIT =
"ozone.block.deleting.pending.command.limit";
public static final int OZONE_BLOCK_DELETING_PENDING_COMMAND_LIMIT_DEFAULT
= 5; // same with hdds.datanode.block.delete.queue.limit
adoroszlai marked this conversation as resolved.
Show resolved Hide resolved

public static final String OZONE_SNAPSHOT_SST_FILTERING_SERVICE_TIMEOUT =
"ozone.sst.filtering.service.timeout";
Expand Down
Original file line number Diff line number Diff line change
Expand Up @@ -41,6 +41,7 @@
import org.apache.hadoop.hdds.scm.ha.SCMServiceManager;
import org.apache.hadoop.hdds.scm.node.NodeManager;
import org.apache.hadoop.hdds.scm.node.NodeStatus;
import org.apache.hadoop.hdds.scm.node.states.NodeNotFoundException;
import org.apache.hadoop.hdds.server.events.EventPublisher;
import org.apache.hadoop.hdds.utils.BackgroundService;
import org.apache.hadoop.hdds.utils.BackgroundTask;
Expand All @@ -57,6 +58,8 @@
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;

import static org.apache.hadoop.ozone.OzoneConfigKeys.OZONE_BLOCK_DELETING_PENDING_COMMAND_LIMIT;
import static org.apache.hadoop.ozone.OzoneConfigKeys.OZONE_BLOCK_DELETING_PENDING_COMMAND_LIMIT_DEFAULT;
import static org.apache.hadoop.ozone.OzoneConfigKeys.OZONE_BLOCK_DELETING_SERVICE_TIMEOUT;
import static org.apache.hadoop.ozone.OzoneConfigKeys.OZONE_BLOCK_DELETING_SERVICE_TIMEOUT_DEFAULT;

Expand Down Expand Up @@ -90,6 +93,7 @@ public class SCMBlockDeletingService extends BackgroundService

private long safemodeExitMillis = 0;
private final long safemodeExitRunDelayMillis;
private final long deleteBlocksPendingCommandLimit;
private final Clock clock;

@SuppressWarnings("parameternumber")
Expand All @@ -110,6 +114,9 @@ public SCMBlockDeletingService(DeletedBlockLog deletedBlockLog,
HddsConfigKeys.HDDS_SCM_WAIT_TIME_AFTER_SAFE_MODE_EXIT,
HddsConfigKeys.HDDS_SCM_WAIT_TIME_AFTER_SAFE_MODE_EXIT_DEFAULT,
TimeUnit.MILLISECONDS);
this.deleteBlocksPendingCommandLimit = conf.getInt(
OZONE_BLOCK_DELETING_PENDING_COMMAND_LIMIT,
OZONE_BLOCK_DELETING_PENDING_COMMAND_LIMIT_DEFAULT);
this.clock = clock;
this.deletedBlockLog = deletedBlockLog;
this.nodeManager = nodeManager;
Expand Down Expand Up @@ -156,13 +163,12 @@ public EmptyTaskResult call() throws Exception {
List<DatanodeDetails> datanodes =
nodeManager.getNodes(NodeStatus.inServiceHealthy());
if (datanodes != null) {
// When DN node is healthy and in-service, and previous commands
// are handled for deleteBlocks Type, then it will be considered
// in this iteration
final Set<DatanodeDetails> included = datanodes.stream().filter(
dn -> nodeManager.getCommandQueueCount(dn.getUuid(),
Type.deleteBlocksCommand) == 0).collect(Collectors.toSet());
try {
// When DN node is healthy and in-service, and their number of
// 'deleteBlocks' type commands is below the limit.
// These nodes will be considered for this iteration.
final Set<DatanodeDetails> included =
getDatanodesWithinCommandLimit(datanodes);
DatanodeDeletedBlockTransactions transactions =
deletedBlockLog.getTransactions(blockDeleteLimitSize, included);

Expand Down Expand Up @@ -203,15 +209,15 @@ public EmptyTaskResult call() throws Exception {
deletedBlockLog.incrementCount(new ArrayList<>(processedTxIDs));
} catch (NotLeaderException nle) {
LOG.warn("Skip current run, since not leader any more.", nle);
return EmptyTaskResult.newResult();
} catch (NodeNotFoundException e) {
LOG.error("Datanode not found in NodeManager. Should not happen", e);
} catch (IOException e) {
// We may tolerate a number of failures for sometime
// but if it continues to fail, at some point we need to raise
// an exception and probably fail the SCM ? At present, it simply
// continues to retry the scanning.
LOG.error("Failed to get block deletion transactions from delTX log",
e);
return EmptyTaskResult.newResult();
}
}

Expand Down Expand Up @@ -273,4 +279,24 @@ public void stop() {
public ScmBlockDeletingServiceMetrics getMetrics() {
return this.metrics;
}

/**
* Filters and returns a set of healthy datanodes that have not exceeded
* the deleteBlocksPendingCommandLimit.
*
* @param datanodes a list of DatanodeDetails
* @return a set of filtered DatanodeDetails
*/
@VisibleForTesting
protected Set<DatanodeDetails> getDatanodesWithinCommandLimit(
List<DatanodeDetails> datanodes) throws NodeNotFoundException {
final Set<DatanodeDetails> included = new HashSet<>();
for (DatanodeDetails dn : datanodes) {
if (nodeManager.getTotalDatanodeCommandCount(dn,
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The basic idea looks good, but I have a concern that the queued cmd size from the last dn heartbeat has a potential delay. If the delete operation is frequent, there might be a case that the service cannot find idle data nodes to send cmd until the next heartbeat comes. The deletion process might not be smooth as you expect.
Pls correct me if i am wrong.

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

By default the SCM generates a delete command every 60s and sends it to the DN. The DN reports a heartbeat every 30s. So normally SCM can get the newer DN status.

getTotalDatanodeCommandCount returns the number of DeleteBlocksCommand, each time SCM executes DeletedBlockTransactionScanner, only one DeleteBlocksCommand is sent to a specific DN. DeletedBlockTransactionScanner execution frequency is fixed, the limit here is 5, so the SCM must execute at least 5 times before the DN's queue is full, which needs 5 min. as long as the DN can send a heartbeat of before all these commands are executed, then the SCM can continue to send delete commands to the DN.

If the queue of all DNs is full, SCM should not continue to send new commands to DN

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I see. Most of the time it should be fine.
The worst case:
A If DN reported the HB at time A with a full cmd queue.
A+ 29.9s. SCM DeletedBlockTransactionScanner executes and cannot send cmd to DN, needs to wait for next
round.
A+ 90s. SCM has updated the latest HB from DN and DeletedBlockTransactionScanner executes, and finally
send cmd to DN again.
It could lead to at most a 90-sec gap.
Right now it seems to be trivial compared to the interval of DeletedBlockTransactionScanner.
Thx for the explanation.

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yes, and when A+ 29.9s ~ A+ 90s the DN can continue to process the DeleteBlocksCommand in its command queue. And the DN's command queue is full, DN will not be idle. Because this PR is determine whether to continue sending DeleteBlocksCommand to DN according to the command queue length of DN.

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@xichen01

  1. Without processing DN HB, this will get same command again. So it is not useful to keep queue at SCM side. Else SCM will add duplicate command only, and DN will be processing duplicate commands only.
  2. DN queue is limited to "5" only as default (to limit the duplicate request queueing up when DN is slow). and handle memory in queue.

Considering parallel HB and SCM deleteBlock processing, they are synchronized using lock,

  • HB take lock, get commands, process HB response, release lock
  • SCM deleteBlock, getCommandQueueCount() take lock, get count and then release lock

So below sequence,
Scenario 1: - No issue as queue empty and next command can be added

  1. HB comes first
  2. SCM delete block processing next

Scenario 2: Here, adding same command will be duplicate and there is retry

  1. SCM delete block processing first
  2. HB comes next

Considering this, we need not have queue at SCM and above changes not required.

Type.deleteBlocksCommand) < deleteBlocksPendingCommandLimit) {
included.add(dn);
}
}
return included;
}
}
Original file line number Diff line number Diff line change
@@ -0,0 +1,159 @@
package org.apache.hadoop.hdds.scm.block;

import org.apache.hadoop.hdds.conf.OzoneConfiguration;
import org.apache.hadoop.hdds.protocol.DatanodeDetails;
import org.apache.hadoop.hdds.protocol.MockDatanodeDetails;
import org.apache.hadoop.hdds.protocol.proto.StorageContainerDatanodeProtocolProtos.SCMCommandProto.Type;
import org.apache.hadoop.hdds.protocol.proto.StorageContainerDatanodeProtocolProtos.DeletedBlocksTransaction;
import org.apache.hadoop.hdds.scm.events.SCMEvents;
import org.apache.hadoop.hdds.scm.ha.SCMContext;
import org.apache.hadoop.hdds.scm.ha.SCMServiceManager;
import org.apache.hadoop.hdds.scm.node.NodeManager;
import org.apache.hadoop.hdds.scm.node.NodeStatus;
import org.apache.hadoop.hdds.server.events.EventPublisher;
import org.apache.hadoop.ozone.protocol.commands.CommandForDatanode;
import org.junit.jupiter.api.AfterEach;
import org.junit.jupiter.api.BeforeEach;
import org.junit.jupiter.api.Test;
import org.mockito.ArgumentCaptor;

import java.time.Clock;
import java.time.ZoneOffset;
import java.util.Arrays;
import java.util.HashSet;
import java.util.List;
import java.util.Set;
import java.util.UUID;
import java.util.stream.Collectors;

import static org.apache.hadoop.ozone.OzoneConfigKeys.OZONE_BLOCK_DELETING_PENDING_COMMAND_LIMIT;
import static org.apache.hadoop.ozone.OzoneConfigKeys.OZONE_BLOCK_DELETING_PENDING_COMMAND_LIMIT_DEFAULT;
import static org.junit.jupiter.api.Assertions.assertEquals;
import static org.junit.jupiter.api.Assertions.assertFalse;
import static org.mockito.ArgumentMatchers.any;
import static org.mockito.ArgumentMatchers.anyInt;
import static org.mockito.ArgumentMatchers.anySet;
import static org.mockito.ArgumentMatchers.eq;
import static org.mockito.Mockito.mock;
import static org.mockito.Mockito.spy;
import static org.mockito.Mockito.times;
import static org.mockito.Mockito.verify;
import static org.mockito.Mockito.when;

/**
* Test SCMBlockDeletingService.
*/
public class TestSCMBlockDeletingService {
private SCMBlockDeletingService service;
private EventPublisher eventPublisher;
private List<DatanodeDetails> datanodeDetails;
private OzoneConfiguration conf;
private NodeManager nodeManager;
private ScmBlockDeletingServiceMetrics metrics;

@BeforeEach
public void setup() throws Exception {
nodeManager = mock(NodeManager.class);
eventPublisher = mock(EventPublisher.class);
conf = new OzoneConfiguration();
metrics = ScmBlockDeletingServiceMetrics.create();
when(nodeManager.getTotalDatanodeCommandCount(any(),
any())).thenReturn(0);
SCMServiceManager scmServiceManager = mock(SCMServiceManager.class);
SCMContext scmContext = mock(SCMContext.class);

DatanodeDeletedBlockTransactions ddbt =
new DatanodeDeletedBlockTransactions();
DatanodeDetails datanode1 = MockDatanodeDetails.randomDatanodeDetails();
DatanodeDetails datanode2 = MockDatanodeDetails.randomDatanodeDetails();
DatanodeDetails datanode3 = MockDatanodeDetails.randomDatanodeDetails();
datanodeDetails = Arrays.asList(datanode1, datanode2, datanode3);
when(nodeManager.getNodes(NodeStatus.inServiceHealthy())).thenReturn(
datanodeDetails);
DeletedBlocksTransaction tx1 = createTestDeleteTxn(1, Arrays.asList(1L), 1);
ddbt.addTransactionToDN(datanode1.getUuid(), tx1);
ddbt.addTransactionToDN(datanode2.getUuid(), tx1);
ddbt.addTransactionToDN(datanode3.getUuid(), tx1);
DeletedBlockLog mockDeletedBlockLog = mock(DeletedBlockLog.class);
when(mockDeletedBlockLog.getTransactions(
anyInt(), anySet())).thenReturn(ddbt);

service = spy(new SCMBlockDeletingService(
mockDeletedBlockLog, nodeManager, eventPublisher, scmContext,
scmServiceManager, conf, metrics, Clock.system(
ZoneOffset.UTC)));
when(service.shouldRun()).thenReturn(true);
}

@AfterEach
public void stop() {
service.stop();
ScmBlockDeletingServiceMetrics.unRegister();
}

@Test
public void testCall() throws Exception {
callDeletedBlockTransactionScanner();

ArgumentCaptor<CommandForDatanode> argumentCaptor =
ArgumentCaptor.forClass(CommandForDatanode.class);

// Three Datanode is healthy and in-service, and the task queue is empty,
// so the transaction will send to all three Datanode
verify(eventPublisher, times(3)).fireEvent(
eq(SCMEvents.DATANODE_COMMAND), argumentCaptor.capture());
List<CommandForDatanode> actualCommands = argumentCaptor.getAllValues();
List<UUID> actualDnIds = actualCommands.stream()
.map(CommandForDatanode::getDatanodeId)
.collect(Collectors.toList());
Set<UUID> expectedDnIdsSet = datanodeDetails.stream()
.map(DatanodeDetails::getUuid).collect(Collectors.toSet());

assertEquals(expectedDnIdsSet, new HashSet<>(actualDnIds));
assertEquals(datanodeDetails.size(),
metrics.getNumBlockDeletionCommandSent());
// Echo Command has one Transaction
assertEquals(datanodeDetails.size() * 1,
metrics.getNumBlockDeletionTransactionSent());
}

private void callDeletedBlockTransactionScanner() throws Exception {
service.getTasks().poll().call();
}

@Test
public void testLimitCommandSending() throws Exception {
int pendingCommandLimit = conf.getInt(
OZONE_BLOCK_DELETING_PENDING_COMMAND_LIMIT,
OZONE_BLOCK_DELETING_PENDING_COMMAND_LIMIT_DEFAULT);


// The number of commands pending on all Datanodes has reached the limit.
when(nodeManager.getTotalDatanodeCommandCount(any(),
any())).thenReturn(pendingCommandLimit);
assertEquals(0,
service.getDatanodesWithinCommandLimit(datanodeDetails).size());

// The number of commands pending on all Datanodes is 0
when(nodeManager.getTotalDatanodeCommandCount(any(),
any())).thenReturn(0);
assertEquals(datanodeDetails.size(),
service.getDatanodesWithinCommandLimit(datanodeDetails).size());

// The number of commands pending on first Datanodes has reached the limit.
DatanodeDetails fullDatanode = datanodeDetails.get(0);
when(nodeManager.getTotalDatanodeCommandCount(fullDatanode,
Type.deleteBlocksCommand)).thenReturn(pendingCommandLimit);
Set<DatanodeDetails> includeNodes =
service.getDatanodesWithinCommandLimit(datanodeDetails);
assertEquals(datanodeDetails.size() - 1,
includeNodes.size());
assertFalse(includeNodes.contains(fullDatanode));
}

private DeletedBlocksTransaction createTestDeleteTxn(
long txnID, List<Long> blocks, long containerID) {
return DeletedBlocksTransaction.newBuilder().setTxID(txnID)
.setContainerID(containerID).addAllLocalID(blocks).setCount(0).build();
}
}