[pulsar-broker] check replicator periodically to avoid issue due to z…

…k missing watch (apache#6674)

### Motivation
We have a regular issue when user changes replication-cluster list, many times broker misses the zk watch and broker fails to update replicator which causes either data loss or message backlog based on zk-watch missing at source or destination replication broker.

### Modification
- Broker expires replication policies at every X seconds 
- Starts a background task which checks polices (fetches new policies if it's already expired) and starts/stops replicator if needed.

### Result
Broker can start/stop replicators based on updated policies even if broker misses zk watch.

conf/broker.conf

@@ -745,6 +745,10 @@ replicationProducerQueueSize=1000
 # Replicator prefix used for replicator producer name and cursor name
 replicatorPrefix=pulsar.repl
 
+# Duration to check replication policy to avoid replicator inconsistency 
+# due to missing ZooKeeper watch (disable with value 0)
+replicatioPolicyCheckDurationSeconds=600
+
 # Default message retention time
 defaultRetentionTimeInMinutes=0
 

conf/standalone.conf

@@ -495,6 +495,10 @@ replicationConnectionsPerBroker=16
 # Replicator producer queue size
 replicationProducerQueueSize=1000
 
+# Duration to check replication policy to avoid replicator inconsistency 
+# due to missing ZooKeeper watch (disable with value 0)
+replicatioPolicyCheckDurationSeconds=600
+
 # Default message retention time
 defaultRetentionTimeInMinutes=0
 

pulsar-broker-common/src/main/java/org/apache/pulsar/broker/ServiceConfiguration.java

@@ -1270,6 +1270,12 @@ public class ServiceConfiguration implements PulsarConfiguration {
         doc = "Replicator producer queue size"
     )
     private int replicationProducerQueueSize = 1000;
+    @FieldContext(
+            category = CATEGORY_REPLICATION,
+            doc = "Duration to check replication policy to avoid replicator "
+                    + "inconsistency due to missing ZooKeeper watch (disable with value 0)"
+        )
+    private int replicatioPolicyCheckDurationSeconds = 600;
     @Deprecated
     @FieldContext(
         category = CATEGORY_REPLICATION,

pulsar-broker/src/main/java/org/apache/pulsar/broker/service/BrokerService.java

@@ -407,6 +407,7 @@ public void start() throws Exception {
         this.startConsumedLedgersMonitor();
         this.startBacklogQuotaChecker();
         this.updateBrokerPublisherThrottlingMaxRate();
+        this.startCheckReplicationPolicies();
         // register listener to capture zk-latency
         ClientCnxnAspect.addListener(zkStatsListener);
         ClientCnxnAspect.registerExecutor(pulsar.getExecutor());
@@ -449,6 +450,14 @@ protected void startMessageExpiryMonitor() {
                 TimeUnit.MINUTES);
     }
 
+    protected void startCheckReplicationPolicies() {
+        int interval = pulsar.getConfig().getReplicatioPolicyCheckDurationSeconds();
+        if (interval > 0) {
+            messageExpiryMonitor.scheduleAtFixedRate(safeRun(this::checkReplicationPolicies), interval, interval,
+                    TimeUnit.SECONDS);
+        }
+    }
+
     protected void startCompactionMonitor() {
         int interval = pulsar().getConfiguration().getBrokerServiceCompactionMonitorIntervalInSeconds();
         if (interval > 0) {
@@ -1155,6 +1164,10 @@ public void checkMessageExpiry() {
         forEachTopic(Topic::checkMessageExpiry);
     }
 
+    public void checkReplicationPolicies() {
+        forEachTopic(Topic::checkReplication);
+    }
+
     public void checkCompaction() {
         forEachTopic((t) -> {
                 if (t instanceof PersistentTopic) {