BUG 2817 - Basic implementation of RemoveServer in the Raft code
[controller.git] / opendaylight / md-sal / sal-akka-raft / src / main / java / org / opendaylight / controller / cluster / raft / SnapshotManager.java
index 77875a567ed45e2552f56b23e5bac2a70a35d47d..10574e3a02cf881c689a88e6f81de0f021ffe70b 100644 (file)
@@ -12,8 +12,10 @@ import akka.japi.Procedure;
 import akka.persistence.SnapshotSelectionCriteria;
 import com.google.common.annotations.VisibleForTesting;
 import java.util.List;
+import org.opendaylight.controller.cluster.raft.base.messages.ApplySnapshot;
 import org.opendaylight.controller.cluster.raft.base.messages.CaptureSnapshot;
 import org.opendaylight.controller.cluster.raft.base.messages.SendInstallSnapshot;
+import org.opendaylight.controller.cluster.raft.base.messages.SnapshotComplete;
 import org.opendaylight.controller.cluster.raft.behaviors.RaftActorBehavior;
 import org.slf4j.Logger;
 
@@ -37,7 +39,7 @@ public class SnapshotManager implements SnapshotState {
 
     private Procedure<Void> createSnapshotProcedure;
 
-    private Snapshot applySnapshot;
+    private ApplySnapshot applySnapshot;
     private Procedure<byte[]> applySnapshotProcedure;
 
     public SnapshotManager(RaftActorContext context, Logger logger) {
@@ -45,6 +47,10 @@ public class SnapshotManager implements SnapshotState {
         this.LOG = logger;
     }
 
+    public boolean isApplying() {
+        return applySnapshot != null;
+    }
+
     @Override
     public boolean isCapturing() {
         return currentState.isCapturing();
@@ -61,7 +67,7 @@ public class SnapshotManager implements SnapshotState {
     }
 
     @Override
-    public void apply(Snapshot snapshot) {
+    public void apply(ApplySnapshot snapshot) {
         currentState.apply(snapshot);
     }
 
@@ -103,18 +109,49 @@ public class SnapshotManager implements SnapshotState {
     }
 
     private boolean hasFollowers(){
-        return context.getPeerAddresses().keySet().size() > 0;
+        return context.hasFollowers();
     }
 
     private String persistenceId(){
         return context.getId();
     }
 
+    public CaptureSnapshot newCaptureSnapshot(ReplicatedLogEntry lastLogEntry, long replicatedToAllIndex,
+            boolean installSnapshotInitiated) {
+        TermInformationReader lastAppliedTermInfoReader =
+                lastAppliedTermInformationReader.init(context.getReplicatedLog(), context.getLastApplied(),
+                        lastLogEntry, hasFollowers());
+
+        long lastAppliedIndex = lastAppliedTermInfoReader.getIndex();
+        long lastAppliedTerm = lastAppliedTermInfoReader.getTerm();
+
+        TermInformationReader replicatedToAllTermInfoReader =
+                replicatedToAllTermInformationReader.init(context.getReplicatedLog(), replicatedToAllIndex);
+
+        long newReplicatedToAllIndex = replicatedToAllTermInfoReader.getIndex();
+        long newReplicatedToAllTerm = replicatedToAllTermInfoReader.getTerm();
+
+        List<ReplicatedLogEntry> unAppliedEntries = context.getReplicatedLog().getFrom(lastAppliedIndex + 1);
+
+        long lastLogEntryIndex = lastAppliedIndex;
+        long lastLogEntryTerm = lastAppliedTerm;
+        if(lastLogEntry != null) {
+            lastLogEntryIndex = lastLogEntry.getIndex();
+            lastLogEntryTerm = lastLogEntry.getTerm();
+        } else {
+            LOG.debug("Capturing Snapshot : lastLogEntry is null. Using lastAppliedIndex {} and lastAppliedTerm {} instead.",
+                lastAppliedIndex, lastAppliedTerm);
+        }
+
+        return new CaptureSnapshot(lastLogEntryIndex, lastLogEntryTerm, lastAppliedIndex, lastAppliedTerm,
+                newReplicatedToAllIndex, newReplicatedToAllTerm, unAppliedEntries, installSnapshotInitiated);
+    }
+
     private class AbstractSnapshotState implements SnapshotState {
 
         @Override
         public boolean isCapturing() {
-            return false;
+            return true;
         }
 
         @Override
@@ -130,7 +167,7 @@ public class SnapshotManager implements SnapshotState {
         }
 
         @Override
-        public void apply(Snapshot snapshot) {
+        public void apply(ApplySnapshot snapshot) {
             LOG.debug("apply should not be called in state {}", this);
         }
 
@@ -188,27 +225,13 @@ public class SnapshotManager implements SnapshotState {
 
     private class Idle extends AbstractSnapshotState {
 
-        private boolean capture(ReplicatedLogEntry lastLogEntry, long replicatedToAllIndex, String targetFollower) {
-            TermInformationReader lastAppliedTermInfoReader =
-                    lastAppliedTermInformationReader.init(context.getReplicatedLog(), context.getLastApplied(),
-                            lastLogEntry, hasFollowers());
-
-            long lastAppliedIndex = lastAppliedTermInfoReader.getIndex();
-            long lastAppliedTerm = lastAppliedTermInfoReader.getTerm();
-
-            TermInformationReader replicatedToAllTermInfoReader =
-                    replicatedToAllTermInformationReader.init(context.getReplicatedLog(), replicatedToAllIndex);
-
-            long newReplicatedToAllIndex = replicatedToAllTermInfoReader.getIndex();
-            long newReplicatedToAllTerm = replicatedToAllTermInfoReader.getTerm();
-
-            // send a CaptureSnapshot to self to make the expensive operation async.
-
-            List<ReplicatedLogEntry> unAppliedEntries = context.getReplicatedLog().getFrom(lastAppliedIndex + 1);
+        @Override
+        public boolean isCapturing() {
+            return false;
+        }
 
-            captureSnapshot = new CaptureSnapshot(lastLogEntry.getIndex(),
-                    lastLogEntry.getTerm(), lastAppliedIndex, lastAppliedTerm,
-                    newReplicatedToAllIndex, newReplicatedToAllTerm, unAppliedEntries, targetFollower != null);
+        private boolean capture(ReplicatedLogEntry lastLogEntry, long replicatedToAllIndex, String targetFollower) {
+            captureSnapshot = newCaptureSnapshot(lastLogEntry, replicatedToAllIndex, targetFollower != null);
 
             if(captureSnapshot.isInstallSnapshotInitiated()) {
                 LOG.info("{}: Initiating snapshot capture {} to install on {}",
@@ -245,14 +268,14 @@ public class SnapshotManager implements SnapshotState {
         }
 
         @Override
-        public void apply(Snapshot snapshot) {
-            applySnapshot = snapshot;
+        public void apply(ApplySnapshot applySnapshot) {
+            SnapshotManager.this.applySnapshot = applySnapshot;
 
             lastSequenceNumber = context.getPersistenceProvider().getLastSequenceNumber();
 
             LOG.debug("lastSequenceNumber prior to persisting applied snapshot: {}", lastSequenceNumber);
 
-            context.getPersistenceProvider().saveSnapshot(snapshot);
+            context.getPersistenceProvider().saveSnapshot(applySnapshot.getSnapshot());
 
             SnapshotManager.this.currentState = PERSISTING;
         }
@@ -270,11 +293,6 @@ public class SnapshotManager implements SnapshotState {
 
     private class Creating extends AbstractSnapshotState {
 
-        @Override
-        public boolean isCapturing() {
-            return true;
-        }
-
         @Override
         public void persist(byte[] snapshotBytes, RaftActorBehavior currentBehavior, long totalMemory) {
             // create a snapshot object from the state provided and save it
@@ -283,25 +301,38 @@ public class SnapshotManager implements SnapshotState {
             Snapshot snapshot = Snapshot.create(snapshotBytes,
                     captureSnapshot.getUnAppliedEntries(),
                     captureSnapshot.getLastIndex(), captureSnapshot.getLastTerm(),
-                    captureSnapshot.getLastAppliedIndex(), captureSnapshot.getLastAppliedTerm());
+                    captureSnapshot.getLastAppliedIndex(), captureSnapshot.getLastAppliedTerm(),
+                    context.getTermInformation().getCurrentTerm(),
+                    context.getTermInformation().getVotedFor(), context.getPeerServerInfo());
 
             context.getPersistenceProvider().saveSnapshot(snapshot);
 
-            LOG.info("{}: Persisting of snapshot done:{}", persistenceId(), snapshot.getLogMessage());
+            LOG.info("{}: Persisting of snapshot done: {}", persistenceId(), snapshot);
 
             long dataThreshold = totalMemory *
                     context.getConfigParams().getSnapshotDataThresholdPercentage() / 100;
-            if (context.getReplicatedLog().dataSize() > dataThreshold) {
+            boolean dataSizeThresholdExceeded = context.getReplicatedLog().dataSize() > dataThreshold;
 
+            boolean logSizeExceededSnapshotBatchCount =
+                    context.getReplicatedLog().size() >= context.getConfigParams().getSnapshotBatchCount();
+
+            if (dataSizeThresholdExceeded || logSizeExceededSnapshotBatchCount) {
                 if(LOG.isDebugEnabled()) {
-                    LOG.debug("{}: dataSize {} exceeds dataThreshold {} - doing snapshotPreCommit with index {}",
-                            persistenceId(), context.getReplicatedLog().dataSize(), dataThreshold,
-                            captureSnapshot.getLastAppliedIndex());
+                    if(dataSizeThresholdExceeded) {
+                        LOG.debug("{}: log data size {} exceeds the memory threshold {} - doing snapshotPreCommit with index {}",
+                                context.getId(), context.getReplicatedLog().dataSize(), dataThreshold,
+                                captureSnapshot.getLastAppliedIndex());
+                    } else {
+                        LOG.debug("{}: log size {} exceeds the snapshot batch count {} - doing snapshotPreCommit with index {}",
+                                context.getId(), context.getReplicatedLog().size(),
+                                context.getConfigParams().getSnapshotBatchCount(), captureSnapshot.getLastAppliedIndex());
+                    }
                 }
 
-                // if memory is less, clear the log based on lastApplied.
-                // this could/should only happen if one of the followers is down
-                // as normally we keep removing from the log when its replicated to all.
+                // We either exceeded the memory threshold or the log size exceeded the snapshot batch
+                // count so, to keep the log memory footprint in check, clear the log based on lastApplied.
+                // This could/should only happen if one of the followers is down as normally we keep
+                // removing from the log as entries are replicated to all.
                 context.getReplicatedLog().snapshotPreCommit(captureSnapshot.getLastAppliedIndex(),
                         captureSnapshot.getLastAppliedTerm());
 
@@ -326,9 +357,9 @@ public class SnapshotManager implements SnapshotState {
                         context.getReplicatedLog().getSnapshotTerm());
             }
 
-            LOG.info("{}: Removed in-memory snapshotted entries, adjusted snaphsotIndex:{} " +
-                            "and term:{}", persistenceId(), captureSnapshot.getLastAppliedIndex(),
-                    captureSnapshot.getLastAppliedTerm());
+            LOG.info("{}: Removed in-memory snapshotted entries, adjusted snaphsotIndex: {} " +
+                    "and term: {}", context.getId(), context.getReplicatedLog().getSnapshotIndex(),
+                    context.getReplicatedLog().getSnapshotTerm());
 
             if (context.getId().equals(currentBehavior.getLeaderId())
                     && captureSnapshot.isInstallSnapshotInitiated()) {
@@ -351,16 +382,23 @@ public class SnapshotManager implements SnapshotState {
 
         @Override
         public void commit(long sequenceNumber, RaftActorBehavior currentBehavior) {
-            LOG.debug("Snapshot success sequence number:", sequenceNumber);
+            LOG.debug("Snapshot success sequence number: {}", sequenceNumber);
 
             if(applySnapshot != null) {
                 try {
-                    applySnapshotProcedure.apply(applySnapshot.getState());
+                    Snapshot snapshot = applySnapshot.getSnapshot();
 
                     //clears the followers log, sets the snapshot index to ensure adjusted-index works
-                    context.setReplicatedLog(ReplicatedLogImpl.newInstance(applySnapshot, context, currentBehavior));
-                    context.setLastApplied(applySnapshot.getLastAppliedIndex());
-                    context.setCommitIndex(applySnapshot.getLastAppliedIndex());
+                    context.setReplicatedLog(ReplicatedLogImpl.newInstance(snapshot, context, currentBehavior));
+                    context.setLastApplied(snapshot.getLastAppliedIndex());
+                    context.setCommitIndex(snapshot.getLastAppliedIndex());
+                    context.getTermInformation().update(snapshot.getElectionTerm(), snapshot.getElectionVotedFor());
+
+                    if(snapshot.getState().length > 0 ) {
+                        applySnapshotProcedure.apply(snapshot.getState());
+                    }
+
+                    applySnapshot.getCallback().onSuccess();
                 } catch (Exception e) {
                     LOG.error("Error applying snapshot", e);
                 }
@@ -373,9 +411,7 @@ public class SnapshotManager implements SnapshotState {
 
             context.getPersistenceProvider().deleteMessages(lastSequenceNumber);
 
-            lastSequenceNumber = -1;
-            applySnapshot = null;
-            SnapshotManager.this.currentState = IDLE;
+            snapshotComplete();
         }
 
         @Override
@@ -389,11 +425,19 @@ public class SnapshotManager implements SnapshotState {
                         context.getReplicatedLog().getSnapshotIndex(),
                         context.getReplicatedLog().getSnapshotTerm(),
                         context.getReplicatedLog().size());
+            } else {
+                applySnapshot.getCallback().onFailure();
             }
 
+            snapshotComplete();
+        }
+
+        private void snapshotComplete() {
             lastSequenceNumber = -1;
             applySnapshot = null;
             SnapshotManager.this.currentState = IDLE;
+
+            context.getActor().tell(SnapshotComplete.INSTANCE, context.getActor());
         }
 
         @Override