public void Open(
RecoveredOrCopiedCheckpointLsn recoveredOrCopiedCheckpointLsn,
RoleContextDrainState roleContextDrainState,
OperationProcessor recordsProcessor,
IStateReplicator fabricReplicator,
IBackupManager backupManager,
CheckpointManager checkpointManager,
TransactionManager transactionManager,
ReplicatedLogManager replicatedLogManager,
IStateManager stateManager,
TransactionalReplicatorSettings replicatorSettings,
RecoveryManager recoveryManager,
ITracer tracer)
{
this.recoveredOrCopiedCheckpointLsn = recoveredOrCopiedCheckpointLsn;
this.recoveryManager = recoveryManager;
this.replicatorSettings = replicatorSettings;
this.stateManager = stateManager;
this.tracer = tracer;
this.roleContextDrainState = roleContextDrainState;
this.recordsProcessor = recordsProcessor;
this.fabricReplicator = fabricReplicator;
this.backupManager = backupManager;
this.transactionManager = transactionManager;
this.checkpointManager = checkpointManager;
this.replicatedLogManager = replicatedLogManager;
}
public TransactionManager(
RecoveredOrCopiedCheckpointLsn recoveredOrCopiedCheckpointLsn,
TransactionMap transactionMap,
FabricPerformanceCounterSetInstance instance,
CheckpointManager checkpointManager,
ReplicatedLogManager replicatedLogManager,
OperationProcessor unlockCallbackManager,
ITracer tracer,
long flushAtBufferedBytes)
{
this.recoveredOrCopiedCheckpointLsn = recoveredOrCopiedCheckpointLsn;
this.transactionMap = transactionMap;
this.checkpointManager = checkpointManager;
this.replicatedLogManager = replicatedLogManager;
this.unlockCallbackManager = unlockCallbackManager;
this.tracer = tracer;
this.flushAtBufferedBytes = flushAtBufferedBytes;
this.beginTransactionOperationRatePerformanceCounterWriter =
new BeginTransactionOperationRatePerformanceCounterWriter(instance);
this.addOperationRatePerformanceCounterWriter =
new AddOperationRatePerformanceCounterWriter(instance);
this.transactionCommitRatePerformanceCounterWriter =
new TransactionCommitRatePerformanceCounterWriter(instance);
this.transactionAbortRatePerformanceCounterWriter =
new TransactionAbortRatePerformanceCounterWriter(instance);
this.addAtomicOperationRatePerformanceCounterWriter =
new AddAtomicOperationRatePerformanceCounterWriter(instance);
}
public async Task PerformRecoveryAsync(
RecoveredOrCopiedCheckpointLsn recoveredOrCopiedCheckpointLsn,
OperationProcessor recordsProcessor,
CheckpointManager checkpointManager,
TransactionManager transactionManager,
LogRecordsDispatcher logRecordsDispatcher,
ReplicatedLogManager replicatedLogManager)
{
var lastPhysicalRecord = PhysicalLogRecord.InvalidPhysicalLogRecord;
LogicalSequenceNumber lsn;
var recoveredRecords = new List <LogRecord>();
var recoveredLastCompletedBeginCheckpointRecord = await this.RecoveryLogsReader.GetLastCompletedBeginCheckpointRecord(this.RecoveredLastEndCheckpointRecord).ConfigureAwait(false);
var recoveredLastCompletedBeginCheckpointRecordPosition = recoveredLastCompletedBeginCheckpointRecord.RecordPosition;
var recoveryStartingPosition = recoveredLastCompletedBeginCheckpointRecordPosition - recoveredLastCompletedBeginCheckpointRecord.EarliestPendingTransactionOffset;
// Set the recovered lsn before performing recovery as it impacts unlock callback logic during recovery
recoveredOrCopiedCheckpointLsn.Update(recoveredLastCompletedBeginCheckpointRecord.Lsn);
var trace =
string.Format(
CultureInfo.InvariantCulture,
"PerformRecoveryAsync: " + Environment.NewLine
+ " Recovered last completed begin checkpoint record." + Environment.NewLine
+ " Type: {0} LSN: {1} PSN:{2} Position: {3}" + Environment.NewLine
+ " Recovery Starting Position: {4} Recovery Starting Epoch: {5},{6}" + Environment.NewLine
+ " LogCompleteCheckpointAfterRecovery: {7}" + " Recovered ProgressVector: {8}"
+ Environment.NewLine,
recoveredLastCompletedBeginCheckpointRecord.RecordType,
recoveredLastCompletedBeginCheckpointRecord.Lsn.LSN,
recoveredLastCompletedBeginCheckpointRecord.Psn.PSN,
recoveredLastCompletedBeginCheckpointRecord.RecordPosition,
recoveryStartingPosition,
recoveredLastCompletedBeginCheckpointRecord.Epoch.DataLossNumber,
recoveredLastCompletedBeginCheckpointRecord.Epoch.ConfigurationNumber,
this.logCompleteCheckpointAfterRecovery,
recoveredLastCompletedBeginCheckpointRecord.ProgressVector.ToString(Constants.ProgressVectorMaxStringSizeInKb));
FabricEvents.Events.Api(tracer.Type, trace);
LogRecord lastRecoverableRecord = null;
replicatedLogManager.LogManager.SetSequentialAccessReadSize(
this.RecoveryLogsReader.ReadStream,
LogManager.ReadAheadBytesForSequentialStream);
// Updated to the recovered value in LogRecordsMap during successful recovery
long lastPeriodicTruncationTimeTicks = 0;
using (var records = new LogRecords(this.RecoveryLogsReader.ReadStream, recoveryStartingPosition, this.tailRecordAtStart.RecordPosition))
{
var hasMoved = await records.MoveNextAsync(CancellationToken.None).ConfigureAwait(false);
Utility.Assert(hasMoved == true, "hasMoved == true");
lsn = records.Current.Lsn;
if (recoveredLastCompletedBeginCheckpointRecord.EarliestPendingTransactionOffset != 0)
{
lsn -= 1;
}
var logRecordsMap = new LogRecordsMap(
lsn,
transactionManager.TransactionsMap,
recoveredLastCompletedBeginCheckpointRecord.Epoch,
checkpointManager.LastStableLsn,
recoveredLastCompletedBeginCheckpointRecord.ProgressVector,
recoveredLastCompletedBeginCheckpointRecord,
this.tracer,
this.RecoveredLastEndCheckpointRecord);
do
{
var isRecoverableRecord = true;
var record = records.Current;
record.CompletedFlush(null);
Utility.Assert(
record.PreviousPhysicalRecord == logRecordsMap.LastPhysicalRecord,
"record.PreviousPhysicalRecord == lastPhysicalRecord");
logRecordsMap.ProcessLogRecord(record, out isRecoverableRecord);
if (isRecoverableRecord == true)
{
recoveredRecords.Add(record);
lastRecoverableRecord = record;
if (LoggingReplicator.IsBarrierRecord(record) &&
recoveredRecords.Count >= ReplicatorDynamicConstants.ParallelRecoveryBatchSize)
{
logRecordsDispatcher.DispatchLoggedRecords(new LoggedRecords(recoveredRecords, null));
recoveredRecords = new List <LogRecord>();
this.recoveryException = recordsProcessor.ServiceException;
if (this.recoveryException == null)
{
this.recoveryException = recordsProcessor.LogException;
}
if (this.recoveryException != null)
{
// If there was an apply or unlock failure, report fault does not help during recovery because the replica is not opened yet.
// The only solution here is to throw during OpenAsync
FabricEvents.Events.Api(
tracer.Type,
"PerformRecoveryAsync: RecoveryFailed");
await lastRecoverableRecord.AwaitProcessing().ConfigureAwait(false);
await recordsProcessor.WaitForAllRecordsProcessingAsync().ConfigureAwait(false);
throw this.recoveryException;
}
}
}
else
{
Utility.Assert(LoggingReplicator.IsBarrierRecord(record) == false, "IsBarrierRecord(record) == false");
record.CompletedApply(null);
record.CompletedProcessing(null);
}
hasMoved = await records.MoveNextAsync(CancellationToken.None).ConfigureAwait(false);
} while (hasMoved == true);
this.LastRecoveredAtomicRedoOperationLsn = logRecordsMap.LastRecoveredAtomicRedoOperationLsn;
checkpointManager.ResetStableLsn(logRecordsMap.LastStableLsn);
lastPhysicalRecord = logRecordsMap.LastPhysicalRecord;
lsn = logRecordsMap.LastLogicalSequenceNumber;
if (recoveredRecords.Count > 0)
{
logRecordsDispatcher.DispatchLoggedRecords(new LoggedRecords(recoveredRecords, null));
}
var tailRecord = records.Current;
FabricEvents.Events.Api(
tracer.Type,
"PerformRecoveryAsync: Current tail record Type: " + tailRecord.RecordType + " LSN: "
+ tailRecord.Lsn.LSN + " PSN: " + tailRecord.Psn.PSN + " Position: " + tailRecord.RecordPosition);
Utility.Assert(
lastPhysicalRecord == records.LastPhysicalRecord,
"lastPhysicalRecord == records.LastPhysicalRecord");
Utility.Assert(
(tailRecord == lastPhysicalRecord) || (tailRecord.PreviousPhysicalRecord == lastPhysicalRecord),
"(tailRecord == lastPhysicalRecord) || "
+ "(tailRecord.PreviousPhysicalRecord == lastPhysicalRecord), " + "LastPhysicalRecord PSN: {0}",
lastPhysicalRecord.Psn.PSN);
Utility.Assert(
logRecordsMap.LastCompletedEndCheckpointRecord != null,
"this.lastCompletedEndCheckpointRecord != null");
Utility.Assert(
logRecordsMap.LastInProgressCheckpointRecord == null,
"this.lastInProgressCheckpointRecord == null");
Utility.Assert(logRecordsMap.LastLinkedPhysicalRecord != null, "this.lastLinkedPhysicalRecord != nul");
Utility.Assert(lsn == tailRecord.Lsn, "lsn == tailRecord.LastLogicalSequenceNumber");
// Disable read ahead as indexing physical records will read the log backwards
replicatedLogManager.LogManager.SetSequentialAccessReadSize(this.RecoveryLogsReader.ReadStream, 0);
await this.RecoveryLogsReader.IndexPhysicalRecords(lastPhysicalRecord).ConfigureAwait(false);
var callbackManager = new PhysicalLogWriterCallbackManager(this.flushedRecordsCallback);
callbackManager.FlushedPsn = tailRecord.Psn + 1;
replicatedLogManager.LogManager.PrepareToLog(tailRecord, callbackManager);
replicatedLogManager.Reuse(
recoveredLastCompletedBeginCheckpointRecord.ProgressVector,
logRecordsMap.LastCompletedEndCheckpointRecord,
logRecordsMap.LastInProgressCheckpointRecord,
logRecordsMap.LastLinkedPhysicalRecord,
replicatedLogManager.LastInformationRecord,
(IndexingLogRecord)this.RecoveryLogsReader.StartingRecord,
logRecordsMap.CurrentLogTailEpoch,
lsn);
this.RecoveredLsn = lsn;
// GopalK: The order of the following statements is significant
if (this.logCompleteCheckpointAfterRecovery)
{
replicatedLogManager.CompleteCheckpoint();
}
replicatedLogManager.Information(InformationEvent.Recovered);
await replicatedLogManager.LogManager.FlushAsync("PerformRecovery").ConfigureAwait(false);
await lastRecoverableRecord.AwaitProcessing().ConfigureAwait(false);
await recordsProcessor.WaitForAllRecordsProcessingAsync().ConfigureAwait(false);
lastPeriodicTruncationTimeTicks = logRecordsMap.LastPeriodicTruncationTimeTicks;
}
this.recoveryException = recordsProcessor.ServiceException;
if (this.recoveryException == null)
{
this.recoveryException = recordsProcessor.LogException;
}
if (this.recoveryException != null)
{
FabricEvents.Events.Api(
tracer.Type,
"PerformRecoveryAsync: RecoveryFailed");
// If there was an apply or unlock failure, report fault does not help during recovery because the replica is not opened yet.
// The only solution here is to throw during OpenAsync
throw this.recoveryException;
}
checkpointManager.Recover(
recoveredLastCompletedBeginCheckpointRecord,
lastPeriodicTruncationTimeTicks);
}
