[tunnelmgrd]: Warm boot support (sonic-net#2166)

* [tunnelmgrd]: Warm boot support Signed-off-by: Lawrence Lee <lawlee@microsoft.com>
liorghub · Mar 17, 2022 · 829b219 · 829b219
1 parent ad65b0a
commit 829b219
Show file tree

Hide file tree

Showing 4 changed files with 85 additions and 2 deletions.
diff --git a/cfgmgr/tunnelmgr.cpp b/cfgmgr/tunnelmgr.cpp
@@ -9,6 +9,7 @@
 #include "tokenize.h"
 #include "shellcmd.h"
 #include "exec.h"
+#include "warm_restart.h"
 
 using namespace std;
 using namespace swss;
@@ -107,7 +108,8 @@ static int cmdIpTunnelRouteDel(const std::string& pfx, std::string & res)
 TunnelMgr::TunnelMgr(DBConnector *cfgDb, DBConnector *appDb, const std::vector<std::string> &tableNames) :
         Orch(cfgDb, tableNames),
         m_appIpInIpTunnelTable(appDb, APP_TUNNEL_DECAP_TABLE_NAME),
-        m_cfgPeerTable(cfgDb, CFG_PEER_SWITCH_TABLE_NAME)
+        m_cfgPeerTable(cfgDb, CFG_PEER_SWITCH_TABLE_NAME),
+        m_cfgTunnelTable(cfgDb, CFG_TUNNEL_TABLE_NAME)
 {
     std::vector<string> peer_keys;
     m_cfgPeerTable.getKeys(peer_keys);
@@ -126,6 +128,23 @@ TunnelMgr::TunnelMgr(DBConnector *cfgDb, DBConnector *appDb, const std::vector<s
             }
         }
     }
+
+    if (WarmStart::isWarmStart())
+    {
+        std::vector<string> tunnel_keys;
+        m_cfgTunnelTable.getKeys(tunnel_keys);
+
+        for (auto tunnel: tunnel_keys)
+        {
+            m_tunnelReplay.insert(tunnel);
+        }
+        if (m_tunnelReplay.empty())
+        {
+            finalizeWarmReboot();
+        }
+
+    }
+
 
     auto consumerStateTable = new swss::ConsumerStateTable(appDb, APP_TUNNEL_ROUTE_TABLE_NAME,
                               TableConsumable::DEFAULT_POP_BATCH_SIZE, default_orch_pri);
@@ -191,6 +210,11 @@ void TunnelMgr::doTask(Consumer &consumer)
             ++it;
         }
     }
+
+    if (!replayDone && m_tunnelReplay.empty() && WarmStart::isWarmStart())
+    {
+        finalizeWarmReboot();
+    }
 }
 
 bool TunnelMgr::doTunnelTask(const KeyOpFieldsValuesTuple & t)
@@ -230,8 +254,16 @@ bool TunnelMgr::doTunnelTask(const KeyOpFieldsValuesTuple & t)
                 SWSS_LOG_NOTICE("Peer/Remote IP not configured");
             }
 
-            m_appIpInIpTunnelTable.set(tunnelName, kfvFieldsValues(t));
+            /* If the tunnel is already in hardware (i.e. present in the replay),
+             * don't try to create it again since it will cause an OA crash
+             * (warmboot case)
+             */
+            if (m_tunnelReplay.find(tunnelName) == m_tunnelReplay.end())
+            {
+                m_appIpInIpTunnelTable.set(tunnelName, kfvFieldsValues(t));
+            }
         }
+        m_tunnelReplay.erase(tunnelName);
         m_tunnelCache[tunnelName] = tunInfo;
     }
     else
@@ -356,3 +388,13 @@ bool TunnelMgr::configIpTunnel(const TunnelInfo& tunInfo)
 
     return true;
 }
+
+
+void TunnelMgr::finalizeWarmReboot()
+{
+    replayDone = true;
+    WarmStart::setWarmStartState("tunnelmgrd", WarmStart::REPLAYED);
+    SWSS_LOG_NOTICE("tunnelmgrd warmstart state set to REPLAYED");
+    WarmStart::setWarmStartState("tunnelmgrd", WarmStart::RECONCILED);
+    SWSS_LOG_NOTICE("tunnelmgrd warmstart state set to RECONCILED");
+}
diff --git a/cfgmgr/tunnelmgr.h b/cfgmgr/tunnelmgr.h
@@ -4,6 +4,8 @@
 #include "producerstatetable.h"
 #include "orch.h"
 
+#include <set>
+
 namespace swss {
 
 struct TunnelInfo
@@ -28,12 +30,18 @@ class TunnelMgr : public Orch
 
     bool configIpTunnel(const TunnelInfo& info);
 
+    void finalizeWarmReboot();
+
     ProducerStateTable m_appIpInIpTunnelTable;
     Table m_cfgPeerTable;
+    Table m_cfgTunnelTable;
 
     std::map<std::string, TunnelInfo > m_tunnelCache;
     std::map<std::string, IpPrefix> m_intfCache;
     std::string m_peerIp;
+
+    std::set<std::string> m_tunnelReplay;
+    bool replayDone = false;
 };
 
 }
diff --git a/cfgmgr/tunnelmgrd.cpp b/cfgmgr/tunnelmgrd.cpp
@@ -11,6 +11,7 @@
 #include "exec.h"
 #include "schema.h"
 #include "tunnelmgr.h"
+#include "warm_restart.h"
 
 using namespace std;
 using namespace swss;
@@ -54,6 +55,9 @@ int main(int argc, char **argv)
         DBConnector cfgDb("CONFIG_DB", 0);
         DBConnector appDb("APPL_DB", 0);
 
+        WarmStart::initialize("tunnelmgrd", "swss");
+        WarmStart::checkWarmStart("tunnelmgrd", "swss");
+
         TunnelMgr tunnelmgr(&cfgDb, &appDb, cfgTunTables);
 
         std::vector<Orch *> cfgOrchList = {&tunnelmgr};

diff --git a/tests/test_warm_reboot.py b/tests/test_warm_reboot.py
@@ -2365,6 +2365,35 @@ def test_EverflowWarmReboot(self, dvs, dvs_acl):
         dvs.start_swss()
         dvs.check_swss_ready()
 
+    def test_TunnelMgrdWarmRestart(self, dvs):
+        tunnel_name = "MuxTunnel0"
+        tunnel_table = "TUNNEL_DECAP_TABLE"
+        tunnel_params = {
+            "tunnel_type": "IPINIP",
+            "dst_ip": "10.1.0.32",
+            "dscp_mode": "uniform",
+            "ecn_mode": "standard",
+            "ttl_mode": "pipe"
+        }
+
+        pubsub = dvs.SubscribeAppDbObject(tunnel_table)
+
+        dvs.runcmd("config warm_restart enable swss")
+        config_db = dvs.get_config_db()
+        config_db.create_entry("TUNNEL", tunnel_name, tunnel_params)
+
+        app_db = dvs.get_app_db()
+        app_db.wait_for_matching_keys(tunnel_table, [tunnel_name])
+
+        nadd, ndel = dvs.CountSubscribedObjects(pubsub)
+        assert nadd == len(tunnel_params)
+        assert ndel == 1  # Expect 1 deletion as part of table creation
+
+        dvs.runcmd("supervisorctl restart tunnelmgrd")
+        dvs.check_services_ready()
+        nadd, ndel = dvs.CountSubscribedObjects(pubsub)
+        assert nadd == 0
+        assert ndel == 0
 
 # Add Dummy always-pass test at end as workaroud
 # for issue when Flaky fail on final test it invokes module tear-down before retrying