import akka.persistence.RecoveryCompleted;
import akka.persistence.SnapshotOffer;
-import akka.persistence.SnapshotSelectionCriteria;
import com.google.common.base.Stopwatch;
-import java.io.ByteArrayInputStream;
-import java.io.ObjectInputStream;
-import org.opendaylight.controller.cluster.DataPersistenceProvider;
+import java.util.Collections;
import org.opendaylight.controller.cluster.PersistentDataProvider;
-import org.opendaylight.controller.cluster.raft.base.messages.ApplyJournalEntries;
import org.opendaylight.controller.cluster.raft.base.messages.ApplySnapshot;
-import org.opendaylight.controller.cluster.raft.base.messages.DeleteEntries;
-import org.opendaylight.controller.cluster.raft.base.messages.UpdateElectionTerm;
+import org.opendaylight.controller.cluster.raft.persisted.ApplyJournalEntries;
+import org.opendaylight.controller.cluster.raft.persisted.DeleteEntries;
+import org.opendaylight.controller.cluster.raft.persisted.EmptyState;
+import org.opendaylight.controller.cluster.raft.persisted.MigratedSerializable;
+import org.opendaylight.controller.cluster.raft.persisted.ServerConfigurationPayload;
+import org.opendaylight.controller.cluster.raft.persisted.Snapshot;
+import org.opendaylight.controller.cluster.raft.persisted.Snapshot.State;
+import org.opendaylight.controller.cluster.raft.persisted.UpdateElectionTerm;
+import org.opendaylight.controller.cluster.raft.protobuff.client.messages.PersistentPayload;
import org.slf4j.Logger;
/**
private int currentRecoveryBatchCount;
private boolean dataRecoveredWithPersistenceDisabled;
private boolean anyDataRecovered;
+ private boolean hasMigratedDataRecovered;
private Stopwatch recoveryTimer;
private final Logger log;
this.log = context.getLogger();
}
- boolean handleRecoveryMessage(Object message, PersistentDataProvider persistentProvider) {
+ boolean handleRecoveryMessage(final Object message, final PersistentDataProvider persistentProvider) {
log.trace("{}: handleRecoveryMessage: {}", context.getId(), message);
anyDataRecovered = anyDataRecovered || !(message instanceof RecoveryCompleted);
+ if (isMigratedSerializable(message)) {
+ hasMigratedDataRecovered = true;
+ }
+
boolean recoveryComplete = false;
- DataPersistenceProvider persistence = context.getPersistenceProvider();
- if (message instanceof org.opendaylight.controller.cluster.raft.RaftActor.UpdateElectionTerm) {
- // Handle this message for backwards compatibility with pre-Lithium versions.
- org.opendaylight.controller.cluster.raft.RaftActor.UpdateElectionTerm update =
- (org.opendaylight.controller.cluster.raft.RaftActor.UpdateElectionTerm)message;
- context.getTermInformation().update(update.getCurrentTerm(), update.getVotedFor());
- } else if (message instanceof UpdateElectionTerm) {
+ if (message instanceof UpdateElectionTerm) {
context.getTermInformation().update(((UpdateElectionTerm) message).getCurrentTerm(),
((UpdateElectionTerm) message).getVotedFor());
- } else if(persistence.isRecoveryApplicable()) {
- if (message instanceof SnapshotOffer) {
- onRecoveredSnapshot((SnapshotOffer) message);
- } else if (message instanceof ReplicatedLogEntry) {
- onRecoveredJournalLogEntry((ReplicatedLogEntry) message);
- } else if (message instanceof ApplyJournalEntries) {
- onRecoveredApplyLogEntries(((ApplyJournalEntries) message).getToIndex());
- } else if (message instanceof DeleteEntries) {
- replicatedLog().removeFrom(((DeleteEntries) message).getFromIndex());
- } else if (message instanceof org.opendaylight.controller.cluster.raft.RaftActor.DeleteEntries) {
- // Handle this message for backwards compatibility with pre-Lithium versions.
- replicatedLog().removeFrom(((org.opendaylight.controller.cluster.raft.RaftActor.DeleteEntries) message).getFromIndex());
- } else if (message instanceof RecoveryCompleted) {
- onRecoveryCompletedMessage();
- possiblyRestoreFromSnapshot();
- recoveryComplete = true;
- }
+ } else if (message instanceof SnapshotOffer) {
+ onRecoveredSnapshot((SnapshotOffer) message);
+ } else if (message instanceof ReplicatedLogEntry) {
+ onRecoveredJournalLogEntry((ReplicatedLogEntry) message);
+ } else if (message instanceof ApplyJournalEntries) {
+ onRecoveredApplyLogEntries(((ApplyJournalEntries) message).getToIndex());
+ } else if (message instanceof DeleteEntries) {
+ onDeleteEntries((DeleteEntries) message);
+ } else if (message instanceof ServerConfigurationPayload) {
+ context.updatePeerIds((ServerConfigurationPayload)message);
} else if (message instanceof RecoveryCompleted) {
recoveryComplete = true;
-
- if(dataRecoveredWithPersistenceDisabled) {
- // Data persistence is disabled but we recovered some data entries so we must have just
- // transitioned to disabled or a persistence backup was restored. Either way, delete all the
- // messages from the akka journal for efficiency and so that we do not end up with consistency
- // issues in case persistence is -re-enabled.
- persistentProvider.deleteMessages(persistentProvider.getLastSequenceNumber());
-
- // Delete all the akka snapshots as they will not be needed
- persistentProvider.deleteSnapshots(new SnapshotSelectionCriteria(scala.Long.MaxValue(),
- scala.Long.MaxValue(), 0L, 0L));
-
- // Since we cleaned out the journal, we need to re-write the current election info.
- context.getTermInformation().updateAndPersist(context.getTermInformation().getCurrentTerm(),
- context.getTermInformation().getVotedFor());
- }
-
- possiblyRestoreFromSnapshot();
- } else {
- boolean isServerConfigPayload = false;
- if(message instanceof ReplicatedLogEntry){
- ReplicatedLogEntry repLogEntry = (ReplicatedLogEntry)message;
- if(isServerConfigurationPayload(repLogEntry)){
- isServerConfigPayload = true;
- context.updatePeerIds((ServerConfigurationPayload)repLogEntry.getData());
- }
- }
-
- if(!isServerConfigPayload){
- dataRecoveredWithPersistenceDisabled = true;
- }
+ onRecoveryCompletedMessage(persistentProvider);
}
return recoveryComplete;
}
+ @SuppressWarnings("checkstyle:IllegalCatch")
private void possiblyRestoreFromSnapshot() {
- byte[] restoreFromSnapshot = cohort.getRestoreFromSnapshot();
- if(restoreFromSnapshot == null) {
+ Snapshot restoreFromSnapshot = cohort.getRestoreFromSnapshot();
+ if (restoreFromSnapshot == null) {
return;
}
- if(anyDataRecovered) {
+ if (anyDataRecovered) {
log.warn("{}: The provided restore snapshot was not applied because the persistence store is not empty",
context.getId());
return;
}
- try(ObjectInputStream ois = new ObjectInputStream(new ByteArrayInputStream(restoreFromSnapshot))) {
- Snapshot snapshot = (Snapshot) ois.readObject();
-
- log.debug("{}: Deserialized restore snapshot: {}", context.getId(), snapshot);
+ log.debug("{}: Restore snapshot: {}", context.getId(), restoreFromSnapshot);
- context.getSnapshotManager().apply(new ApplySnapshot(snapshot));
- } catch(Exception e) {
- log.error("{}: Error deserializing snapshot restore", context.getId(), e);
- }
+ context.getSnapshotManager().apply(new ApplySnapshot(restoreFromSnapshot));
}
private ReplicatedLog replicatedLog() {
}
private void initRecoveryTimer() {
- if(recoveryTimer == null) {
+ if (recoveryTimer == null) {
recoveryTimer = Stopwatch.createStarted();
}
}
- private void onRecoveredSnapshot(SnapshotOffer offer) {
- if(log.isDebugEnabled()) {
- log.debug("{}: SnapshotOffer called..", context.getId());
- }
+ private void onRecoveredSnapshot(final SnapshotOffer offer) {
+ log.debug("{}: SnapshotOffer called.", context.getId());
initRecoveryTimer();
Snapshot snapshot = (Snapshot) offer.snapshot();
+ for (ReplicatedLogEntry entry: snapshot.getUnAppliedEntries()) {
+ if (isMigratedPayload(entry)) {
+ hasMigratedDataRecovered = true;
+ }
+ }
+
+ if (!context.getPersistenceProvider().isRecoveryApplicable()) {
+ // We may have just transitioned to disabled and have a snapshot containing state data and/or log
+ // entries - we don't want to preserve these, only the server config and election term info.
+
+ snapshot = Snapshot.create(
+ EmptyState.INSTANCE, Collections.emptyList(), -1, -1, -1, -1,
+ snapshot.getElectionTerm(), snapshot.getElectionVotedFor(), snapshot.getServerConfiguration());
+ }
+
// Create a replicated log with the snapshot information
// The replicated log can be used later on to retrieve this snapshot
// when we need to install it on a peer
context.setCommitIndex(snapshot.getLastAppliedIndex());
context.getTermInformation().update(snapshot.getElectionTerm(), snapshot.getElectionVotedFor());
- Stopwatch timer = Stopwatch.createStarted();
+ final Stopwatch timer = Stopwatch.createStarted();
// Apply the snapshot to the actors state
- cohort.applyRecoverySnapshot(snapshot.getState());
+ final State snapshotState = snapshot.getState();
+ if (snapshotState.needsMigration()) {
+ hasMigratedDataRecovered = true;
+ }
+ if (!(snapshotState instanceof EmptyState)) {
+ cohort.applyRecoverySnapshot(snapshotState);
+ }
if (snapshot.getServerConfiguration() != null) {
context.updatePeerIds(snapshot.getServerConfiguration());
timer.stop();
log.info("Recovery snapshot applied for {} in {}: snapshotIndex={}, snapshotTerm={}, journal-size={}",
- context.getId(), timer.toString(), replicatedLog().getSnapshotIndex(),
- replicatedLog().getSnapshotTerm(), replicatedLog().size());
+ context.getId(), timer, replicatedLog().getSnapshotIndex(), replicatedLog().getSnapshotTerm(),
+ replicatedLog().size());
}
- private void onRecoveredJournalLogEntry(ReplicatedLogEntry logEntry) {
- if(log.isDebugEnabled()) {
+ private void onRecoveredJournalLogEntry(final ReplicatedLogEntry logEntry) {
+ if (log.isDebugEnabled()) {
log.debug("{}: Received ReplicatedLogEntry for recovery: index: {}, size: {}", context.getId(),
logEntry.getIndex(), logEntry.size());
}
- if(isServerConfigurationPayload(logEntry)){
+ if (isServerConfigurationPayload(logEntry)) {
context.updatePeerIds((ServerConfigurationPayload)logEntry.getData());
}
- replicatedLog().append(logEntry);
+
+ if (isMigratedPayload(logEntry)) {
+ hasMigratedDataRecovered = true;
+ }
+
+ if (context.getPersistenceProvider().isRecoveryApplicable()) {
+ replicatedLog().append(logEntry);
+ } else if (!isPersistentPayload(logEntry)) {
+ dataRecoveredWithPersistenceDisabled = true;
+ }
}
- private void onRecoveredApplyLogEntries(long toIndex) {
+ private void onRecoveredApplyLogEntries(final long toIndex) {
+ if (!context.getPersistenceProvider().isRecoveryApplicable()) {
+ dataRecoveredWithPersistenceDisabled = true;
+ return;
+ }
+
long lastUnappliedIndex = context.getLastApplied() + 1;
- if(log.isDebugEnabled()) {
+ if (log.isDebugEnabled()) {
// it can happen that lastUnappliedIndex > toIndex, if the AJE is in the persistent journal
// but the entry itself has made it to that state and recovered via the snapshot
log.debug("{}: Received apply journal entries for recovery, applying to state: {} to {}",
long lastApplied = lastUnappliedIndex - 1;
for (long i = lastUnappliedIndex; i <= toIndex; i++) {
ReplicatedLogEntry logEntry = replicatedLog().get(i);
- if(logEntry != null) {
+ if (logEntry != null) {
lastApplied++;
batchRecoveredLogEntry(logEntry);
} else {
context.setCommitIndex(lastApplied);
}
- private void batchRecoveredLogEntry(ReplicatedLogEntry logEntry) {
+ private void onDeleteEntries(final DeleteEntries deleteEntries) {
+ if (context.getPersistenceProvider().isRecoveryApplicable()) {
+ replicatedLog().removeFrom(deleteEntries.getFromIndex());
+ } else {
+ dataRecoveredWithPersistenceDisabled = true;
+ }
+ }
+
+ private void batchRecoveredLogEntry(final ReplicatedLogEntry logEntry) {
initRecoveryTimer();
int batchSize = context.getConfigParams().getJournalRecoveryLogBatchSize();
- if(!isServerConfigurationPayload(logEntry)){
- if(currentRecoveryBatchCount == 0) {
+ if (!isServerConfigurationPayload(logEntry)) {
+ if (currentRecoveryBatchCount == 0) {
cohort.startLogRecoveryBatch(batchSize);
}
cohort.appendRecoveredLogEntry(logEntry.getData());
- if(++currentRecoveryBatchCount >= batchSize) {
+ if (++currentRecoveryBatchCount >= batchSize) {
endCurrentLogRecoveryBatch();
}
}
currentRecoveryBatchCount = 0;
}
- private void onRecoveryCompletedMessage() {
- if(currentRecoveryBatchCount > 0) {
+ private void onRecoveryCompletedMessage(final PersistentDataProvider persistentProvider) {
+ if (currentRecoveryBatchCount > 0) {
endCurrentLogRecoveryBatch();
}
String recoveryTime = "";
- if(recoveryTimer != null) {
+ if (recoveryTimer != null) {
recoveryTimer.stop();
recoveryTime = " in " + recoveryTimer.toString();
recoveryTimer = null;
}
- log.info("Recovery completed" + recoveryTime + " - Switching actor to Follower - " +
- "Persistence Id = " + context.getId() +
- " Last index in log = {}, snapshotIndex = {}, snapshotTerm = {}, " +
- "journal-size = {}", replicatedLog().lastIndex(), replicatedLog().getSnapshotIndex(),
- replicatedLog().getSnapshotTerm(), replicatedLog().size());
+ log.info("{}: Recovery completed {} - Switching actor to Follower - last log index = {}, last log term = {}, "
+ + "snapshot index = {}, snapshot term = {}, journal size = {}", context.getId(), recoveryTime,
+ replicatedLog().lastIndex(), replicatedLog().lastTerm(), replicatedLog().getSnapshotIndex(),
+ replicatedLog().getSnapshotTerm(), replicatedLog().size());
+
+ if (dataRecoveredWithPersistenceDisabled
+ || hasMigratedDataRecovered && !context.getPersistenceProvider().isRecoveryApplicable()) {
+ if (hasMigratedDataRecovered) {
+ log.info("{}: Saving snapshot after recovery due to migrated messages", context.getId());
+ } else {
+ log.info("{}: Saving snapshot after recovery due to data persistence disabled", context.getId());
+ }
+
+ // Either data persistence is disabled and we recovered some data entries (ie we must have just
+ // transitioned to disabled or a persistence backup was restored) or we recovered migrated
+ // messages. Either way, we persist a snapshot and delete all the messages from the akka journal
+ // to clean out unwanted messages.
+
+ Snapshot snapshot = Snapshot.create(
+ EmptyState.INSTANCE, Collections.<ReplicatedLogEntry>emptyList(),
+ -1, -1, -1, -1,
+ context.getTermInformation().getCurrentTerm(), context.getTermInformation().getVotedFor(),
+ context.getPeerServerInfo(true));
+
+ persistentProvider.saveSnapshot(snapshot);
+
+ persistentProvider.deleteMessages(persistentProvider.getLastSequenceNumber());
+ } else if (hasMigratedDataRecovered) {
+ log.info("{}: Snapshot capture initiated after recovery due to migrated messages", context.getId());
+
+ context.getSnapshotManager().capture(replicatedLog().last(), -1);
+ } else {
+ possiblyRestoreFromSnapshot();
+ }
+ }
+
+ private static boolean isServerConfigurationPayload(final ReplicatedLogEntry repLogEntry) {
+ return repLogEntry.getData() instanceof ServerConfigurationPayload;
+ }
+
+ private static boolean isPersistentPayload(final ReplicatedLogEntry repLogEntry) {
+ return repLogEntry.getData() instanceof PersistentPayload;
+ }
+
+ private static boolean isMigratedPayload(final ReplicatedLogEntry repLogEntry) {
+ return isMigratedSerializable(repLogEntry.getData());
}
- private static boolean isServerConfigurationPayload(ReplicatedLogEntry repLogEntry){
- return (repLogEntry.getData() instanceof ServerConfigurationPayload);
+ private static boolean isMigratedSerializable(final Object message) {
+ return message instanceof MigratedSerializable && ((MigratedSerializable)message).isMigrated();
}
}
Code Review / controller.git / blobdiff