public WorkerData(Task task, IChangeFeedObserver observer, ChangeFeedObserverContext context, CancellationTokenSource cancellation)
{
this.Task = task;
this.Observer = observer;
this.Context = context;
this.Cancellation = cancellation;
}
public WorkerData(Task task, IChangeFeedObserver observer, ChangeFeedObserverContext context, CancellationTokenSource cancellation, DocumentServiceLease lease)
{
Debug.Assert(task != null);
Debug.Assert(observer != null);
Debug.Assert(context != null);
Debug.Assert(cancellation != null);
this.Task = task;
this.Observer = observer;
this.Context = context;
this.Cancellation = cancellation;
this.Lease = lease;
}
internal async Task CheckpointAsync(string continuation, ChangeFeedObserverContext context)
{
if (string.IsNullOrEmpty(continuation))
{
throw new ArgumentException("continuation");
}
if (context == null)
{
throw new ArgumentNullException("context");
}
if (string.IsNullOrEmpty(context.PartitionKeyRangeId))
{
throw new ArgumentException("context.PartitionKeyRangeId");
}
WorkerData workerData;
this.partitionKeyRangeIdToWorkerMap.TryGetValue(context.PartitionKeyRangeId, out workerData);
if (workerData == null)
{
TraceLog.Warning(string.Format("CheckpointAsync: called at wrong time, failed to get worker data for partition {0}. Most likely the partition is not longer owned by this host.", context.PartitionKeyRangeId));
throw new LeaseLostException(string.Format("Failed to find lease for partition {0} in the set of owned leases.", context.PartitionKeyRangeId));
}
if (workerData.Lease == null)
{
TraceLog.Error(string.Format("CheckpointAsync: found the worker data but lease is null, for partition {0}. This should never happen.", context.PartitionKeyRangeId));
throw new LeaseLostException(string.Format("Failed to find lease for partition {0}.", context.PartitionKeyRangeId));
}
await workerData.CheckpointInProgress.WaitAsync();
try
{
if (workerData.Cancellation.IsCancellationRequested)
{
TraceLog.Warning(string.Format("CheckpointAsync: called at wrong time, partition {0} is shutting down. The ownership of the partition by this host is about to end.", context.PartitionKeyRangeId));
throw new LeaseLostException(string.Format("CheckpointAsync: partition {0} is shutting down.", context.PartitionKeyRangeId));
}
workerData.Lease = await this.CheckpointAsync(workerData.Lease, continuation, context);
}
finally
{
workerData.CheckpointInProgress.Release();
}
}
async Task <DocumentServiceLease> CheckpointAsync(DocumentServiceLease lease, string continuation, ChangeFeedObserverContext context)
{
Debug.Assert(lease != null);
Debug.Assert(!string.IsNullOrEmpty(continuation));
DocumentServiceLease result = null;
try
{
result = (DocumentServiceLease)await this.checkpointManager.CheckpointAsync(lease, continuation, lease.SequenceNumber + 1);
Debug.Assert(result.ContinuationToken == continuation, "ContinuationToken was not updated!");
TraceLog.Informational(string.Format("Checkpoint: partition {0}, new continuation '{1}'", lease.PartitionId, continuation));
}
catch (LeaseLostException)
{
TraceLog.Warning(string.Format("Partition {0}: failed to checkpoint due to lost lease", context.PartitionKeyRangeId));
throw;
}
catch (Exception ex)
{
TraceLog.Error(string.Format("Partition {0}: failed to checkpoint due to unexpected error: {1}", context.PartitionKeyRangeId, ex.Message));
throw;
}
Debug.Assert(result != null);
return(await Task.FromResult <DocumentServiceLease>(result));
}
async Task IPartitionObserver <DocumentServiceLease> .OnPartitionAcquiredAsync(DocumentServiceLease lease)
{
Debug.Assert(lease != null && !string.IsNullOrEmpty(lease.Owner), "lease");
TraceLog.Informational(string.Format("Host '{0}' partition {1}: acquired!", this.HostName, lease.PartitionId));
#if DEBUG
Interlocked.Increment(ref this.partitionCount);
#endif
IChangeFeedObserver observer = this.observerFactory.CreateObserver();
ChangeFeedObserverContext context = new ChangeFeedObserverContext {
PartitionKeyRangeId = lease.PartitionId
};
CancellationTokenSource cancellation = new CancellationTokenSource();
// Create ChangeFeedOptions to use for this worker.
ChangeFeedOptions options = new ChangeFeedOptions
{
MaxItemCount = this.changeFeedOptions.MaxItemCount,
PartitionKeyRangeId = this.changeFeedOptions.PartitionKeyRangeId,
SessionToken = this.changeFeedOptions.SessionToken,
StartFromBeginning = this.changeFeedOptions.StartFromBeginning,
RequestContinuation = this.changeFeedOptions.RequestContinuation
};
var workerTask = await Task.Factory.StartNew(async() =>
{
ChangeFeedObserverCloseReason?closeReason = null;
try
{
try
{
await observer.OpenAsync(context);
}
catch (Exception ex)
{
TraceLog.Error(string.Format("IChangeFeedObserver.OpenAsync exception: {0}", ex));
closeReason = ChangeFeedObserverCloseReason.ObserverError;
throw;
}
options.PartitionKeyRangeId = lease.PartitionId;
if (!string.IsNullOrEmpty(lease.ContinuationToken))
{
options.RequestContinuation = lease.ContinuationToken;
}
CheckpointStats checkpointStats = null;
if (!this.statsSinceLastCheckpoint.TryGetValue(lease.PartitionId, out checkpointStats) || checkpointStats == null)
{
// It could be that the lease was created by different host and we picked it up.
checkpointStats = this.statsSinceLastCheckpoint.AddOrUpdate(
lease.PartitionId,
new CheckpointStats(),
(partitionId, existingStats) => existingStats);
Trace.TraceWarning(string.Format("Added stats for partition '{0}' for which the lease was picked up after the host was started.", lease.PartitionId));
}
IDocumentQuery <Document> query = this.documentClient.CreateDocumentChangeFeedQuery(this.collectionSelfLink, options);
TraceLog.Verbose(string.Format("Worker start: partition '{0}', continuation '{1}'", lease.PartitionId, lease.ContinuationToken));
string lastContinuation = options.RequestContinuation;
try
{
while (this.isShutdown == 0)
{
do
{
ExceptionDispatchInfo exceptionDispatchInfo = null;
FeedResponse <Document> response = null;
try
{
response = await query.ExecuteNextAsync <Document>();
lastContinuation = response.ResponseContinuation;
}
catch (DocumentClientException ex)
{
exceptionDispatchInfo = ExceptionDispatchInfo.Capture(ex);
}
if (exceptionDispatchInfo != null)
{
DocumentClientException dcex = (DocumentClientException)exceptionDispatchInfo.SourceException;
if (StatusCode.NotFound == (StatusCode)dcex.StatusCode && SubStatusCode.ReadSessionNotAvailable != (SubStatusCode)GetSubStatusCode(dcex))
{
// Most likely, the database or collection was removed while we were enumerating.
// Shut down. The user will need to start over.
// Note: this has to be a new task, can't await for shutdown here, as shudown awaits for all worker tasks.
TraceLog.Error(string.Format("Partition {0}: resource gone (subStatus={1}). Aborting.", context.PartitionKeyRangeId, GetSubStatusCode(dcex)));
await Task.Factory.StartNew(() => this.StopAsync(ChangeFeedObserverCloseReason.ResourceGone));
break;
}
else if (StatusCode.Gone == (StatusCode)dcex.StatusCode)
{
SubStatusCode subStatusCode = (SubStatusCode)GetSubStatusCode(dcex);
if (SubStatusCode.PartitionKeyRangeGone == subStatusCode)
{
bool isSuccess = await HandleSplitAsync(context.PartitionKeyRangeId, lastContinuation, lease.Id);
if (!isSuccess)
{
TraceLog.Error(string.Format("Partition {0}: HandleSplit failed! Aborting.", context.PartitionKeyRangeId));
await Task.Factory.StartNew(() => this.StopAsync(ChangeFeedObserverCloseReason.ResourceGone));
break;
}
// Throw LeaseLostException so that we take the lease down.
throw new LeaseLostException(lease, exceptionDispatchInfo.SourceException, true);
}
else if (SubStatusCode.Splitting == subStatusCode)
{
TraceLog.Warning(string.Format("Partition {0} is splitting. Will retry to read changes until split finishes. {1}", context.PartitionKeyRangeId, dcex.Message));
}
else
{
exceptionDispatchInfo.Throw();
}
}
else if (StatusCode.TooManyRequests == (StatusCode)dcex.StatusCode ||
StatusCode.ServiceUnavailable == (StatusCode)dcex.StatusCode)
{
TraceLog.Warning(string.Format("Partition {0}: retriable exception : {1}", context.PartitionKeyRangeId, dcex.Message));
}
else
{
exceptionDispatchInfo.Throw();
}
await Task.Delay(dcex.RetryAfter != TimeSpan.Zero ? dcex.RetryAfter : this.options.FeedPollDelay, cancellation.Token);
}
if (response != null)
{
if (response.Count > 0)
{
List <Document> docs = new List <Document>();
docs.AddRange(response);
try
{
context.FeedResponse = response;
await observer.ProcessChangesAsync(context, docs);
}
catch (Exception ex)
{
TraceLog.Error(string.Format("IChangeFeedObserver.ProcessChangesAsync exception: {0}", ex));
closeReason = ChangeFeedObserverCloseReason.ObserverError;
throw;
}
finally
{
context.FeedResponse = null;
}
}
checkpointStats.ProcessedDocCount += (uint)response.Count;
if (IsCheckpointNeeded(lease, checkpointStats))
{
lease = await CheckpointAsync(lease, response.ResponseContinuation, context);
checkpointStats.Reset();
}
else if (response.Count > 0)
{
TraceLog.Informational(string.Format("Checkpoint: not checkpointing for partition {0}, {1} docs, new continuation '{2}' as frequency condition is not met", lease.PartitionId, response.Count, response.ResponseContinuation));
}
}
}while (query.HasMoreResults && this.isShutdown == 0);
if (this.isShutdown == 0)
{
await Task.Delay(this.options.FeedPollDelay, cancellation.Token);
}
} // Outer while (this.isShutdown == 0) loop.
closeReason = ChangeFeedObserverCloseReason.Shutdown;
}
catch (TaskCanceledException)
{
Debug.Assert(cancellation.IsCancellationRequested, "cancellation.IsCancellationRequested");
TraceLog.Informational(string.Format("Cancel signal received for partition {0} worker!", context.PartitionKeyRangeId));
}
}
catch (LeaseLostException ex)
{
closeReason = ex.IsGone ? ChangeFeedObserverCloseReason.LeaseGone : ChangeFeedObserverCloseReason.LeaseLost;
}
catch (Exception ex)
{
TraceLog.Error(string.Format("Partition {0} exception: {1}", context.PartitionKeyRangeId, ex));
if (!closeReason.HasValue)
{
closeReason = ChangeFeedObserverCloseReason.Unknown;
}
}
if (closeReason.HasValue)
{
TraceLog.Informational(string.Format("Releasing lease for partition {0} due to an error, reason: {1}!", context.PartitionKeyRangeId, closeReason.Value));
// Note: this has to be a new task, because OnPartitionReleasedAsync awaits for worker task.
await Task.Factory.StartNew(async() => await this.partitionManager.TryReleasePartitionAsync(context.PartitionKeyRangeId, true, closeReason.Value));
}
TraceLog.Informational(string.Format("Partition {0}: worker finished!", context.PartitionKeyRangeId));
});
var newWorkerData = new WorkerData(workerTask, observer, context, cancellation);
this.partitionKeyRangeIdToWorkerMap.AddOrUpdate(context.PartitionKeyRangeId, newWorkerData, (string id, WorkerData d) => { return(newWorkerData); });
}
async Task IPartitionObserver <DocumentServiceLease> .OnPartitionAcquiredAsync(DocumentServiceLease lease)
{
Debug.Assert(lease != null && !string.IsNullOrEmpty(lease.Owner), "lease");
TraceLog.Informational(string.Format("Host '{0}' partition {1}: acquired!", this.HostName, lease.PartitionId));
#if DEBUG
Interlocked.Increment(ref this.partitionCount);
#endif
IChangeFeedObserver observer = this.observerFactory.CreateObserver();
ChangeFeedObserverContext context = new ChangeFeedObserverContext {
PartitionKeyRangeId = lease.PartitionId
};
CancellationTokenSource cancellation = new CancellationTokenSource();
// Create ChangeFeedOptions to use for this worker.
ChangeFeedOptions options = new ChangeFeedOptions
{
MaxItemCount = this.changeFeedOptions.MaxItemCount,
PartitionKeyRangeId = this.changeFeedOptions.PartitionKeyRangeId,
SessionToken = this.changeFeedOptions.SessionToken,
StartFromBeginning = this.changeFeedOptions.StartFromBeginning,
RequestContinuation = this.changeFeedOptions.RequestContinuation
};
var workerTask = await Task.Factory.StartNew(async() =>
{
ChangeFeedObserverCloseReason?closeReason = null;
try
{
try
{
await observer.OpenAsync(context);
}
catch (Exception ex)
{
TraceLog.Error(string.Format("IChangeFeedObserver.OpenAsync exception: {0}", ex));
closeReason = ChangeFeedObserverCloseReason.ObserverError;
throw;
}
options.PartitionKeyRangeId = lease.PartitionId;
if (!string.IsNullOrEmpty(lease.ContinuationToken))
{
options.RequestContinuation = lease.ContinuationToken;
}
IDocumentQuery <Document> query = this.documentClient.CreateDocumentChangeFeedQuery(this.collectionSelfLink, options);
TraceLog.Verbose(string.Format("Worker start: partition '{0}', continuation '{1}'", lease.PartitionId, lease.ContinuationToken));
try
{
while (this.isShutdown == 0)
{
do
{
DocumentClientException dcex = null;
FeedResponse <Document> response = null;
try
{
response = await query.ExecuteNextAsync <Document>();
}
catch (DocumentClientException ex)
{
if (StatusCode.NotFound != (StatusCode)ex.StatusCode &&
StatusCode.TooManyRequests != (StatusCode)ex.StatusCode &&
StatusCode.ServiceUnavailable != (StatusCode)ex.StatusCode)
{
throw;
}
dcex = ex;
}
if (dcex != null)
{
const int ReadSessionNotAvailable = 1002;
if (StatusCode.NotFound == (StatusCode)dcex.StatusCode && GetSubStatusCode(dcex) != ReadSessionNotAvailable)
{
// Most likely, the database or collection was removed while we were enumerating.
// Shut down. The user will need to start over.
// Note: this has to be a new task, can't await for shutdown here, as shudown awaits for all worker tasks.
await Task.Factory.StartNew(() => this.StopAsync(ChangeFeedObserverCloseReason.ResourceGone));
break;
}
else
{
Debug.Assert(StatusCode.TooManyRequests == (StatusCode)dcex.StatusCode || StatusCode.ServiceUnavailable == (StatusCode)dcex.StatusCode);
TraceLog.Warning(string.Format("Partition {0}: retriable exception : {1}", context.PartitionKeyRangeId, dcex.Message));
await Task.Delay(dcex.RetryAfter != TimeSpan.Zero ? dcex.RetryAfter : this.options.FeedPollDelay, cancellation.Token);
}
}
if (response != null)
{
if (response.Count > 0)
{
List <Document> docs = new List <Document>();
docs.AddRange(response);
try
{
await observer.ProcessChangesAsync(context, docs);
}
catch (Exception ex)
{
TraceLog.Error(string.Format("IChangeFeedObserver.ProcessChangesAsync exception: {0}", ex));
closeReason = ChangeFeedObserverCloseReason.ObserverError;
throw;
}
// Checkpoint after every successful delivery to the client.
lease = await CheckpointAsync(lease, response.ResponseContinuation, context);
}
else if (string.IsNullOrEmpty(lease.ContinuationToken))
{
// Checkpoint if we've never done that for this lease.
lease = await CheckpointAsync(lease, response.ResponseContinuation, context);
}
}
}while (query.HasMoreResults && this.isShutdown == 0);
if (this.isShutdown == 0)
{
await Task.Delay(this.options.FeedPollDelay, cancellation.Token);
}
} // Outer while (this.isShutdown == 0) loop.
}
catch (TaskCanceledException)
{
Debug.Assert(cancellation.IsCancellationRequested, "cancellation.IsCancellationRequested");
TraceLog.Informational(string.Format("Cancel signal received for partition {0} worker!", context.PartitionKeyRangeId));
}
}
catch (LeaseLostException)
{
closeReason = ChangeFeedObserverCloseReason.LeaseLost;
}
catch (Exception ex)
{
TraceLog.Error(string.Format("Partition {0} exception: {1}", context.PartitionKeyRangeId, ex));
if (!closeReason.HasValue)
{
closeReason = ChangeFeedObserverCloseReason.Unknown;
}
}
if (closeReason.HasValue)
{
TraceLog.Informational(string.Format("Releasing lease for partition {0} due to an error, reason: {1}!", context.PartitionKeyRangeId, closeReason.Value));
// Note: this has to be a new task, because OnPartitionReleasedAsync awaits for worker task.
await Task.Factory.StartNew(async() => await this.partitionManager.TryReleasePartitionAsync(context.PartitionKeyRangeId, true, closeReason.Value));
}
TraceLog.Informational(string.Format("Partition {0}: worker finished!", context.PartitionKeyRangeId));
});
var newWorkerData = new WorkerData(workerTask, observer, context, cancellation);
this.partitionKeyRangeIdToWorkerMap.AddOrUpdate(context.PartitionKeyRangeId, newWorkerData, (string id, WorkerData d) => { return(newWorkerData); });
}