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 InitializeAsync()
{
this.documentClient = new DocumentClient(this.collectionLocation.Uri, this.collectionLocation.MasterKey, this.collectionLocation.ConnectionPolicy);
Uri databaseUri = UriFactory.CreateDatabaseUri(this.collectionLocation.DatabaseName);
Database database = await this.documentClient.ReadDatabaseAsync(databaseUri);
Uri collectionUri = UriFactory.CreateDocumentCollectionUri(this.collectionLocation.DatabaseName, this.collectionLocation.CollectionName);
ResourceResponse <DocumentCollection> collectionResponse = await this.documentClient.ReadDocumentCollectionAsync(
collectionUri,
new RequestOptions { PopulateQuotaInfo = true });
DocumentCollection collection = collectionResponse.Resource;
this.collectionSelfLink = collection.SelfLink;
// Grab the options-supplied prefix if present otherwise leave it empty.
string optionsPrefix = this.options.LeasePrefix ?? string.Empty;
// Beyond this point all access to collection is done via this self link: if collection is removed, we won't access new one using same name by accident.
this.leasePrefix = string.Format(CultureInfo.InvariantCulture, "{0}{1}_{2}_{3}", optionsPrefix, this.collectionLocation.Uri.Host, database.ResourceId, collection.ResourceId);
var leaseManager = new DocumentServiceLeaseManager(
this.auxCollectionLocation,
this.leasePrefix,
this.options.LeaseExpirationInterval,
this.options.LeaseRenewInterval);
await leaseManager.InitializeAsync();
this.leaseManager = leaseManager;
this.checkpointManager = (ICheckpointManager)leaseManager;
if (this.options.DiscardExistingLeases)
{
TraceLog.Warning(string.Format("Host '{0}': removing all leases, as requested by ChangeFeedHostOptions", this.HostName));
await this.leaseManager.DeleteAllAsync();
}
// Note: lease store is never stale as we use monitored colleciton Rid as id prefix for aux collection.
// Collection was removed and re-created, the rid would change.
// If it's not deleted, it's not stale. If it's deleted, it's not stale as it doesn't exist.
await this.leaseManager.CreateLeaseStoreIfNotExistsAsync();
var ranges = new Dictionary <string, PartitionKeyRange>();
foreach (var range in await this.EnumPartitionKeyRangesAsync(this.collectionSelfLink))
{
ranges.Add(range.Id, range);
}
TraceLog.Informational(string.Format("Source collection: '{0}', {1} partition(s), {2} document(s)", this.collectionLocation.CollectionName, ranges.Count, GetDocumentCount(collectionResponse)));
await this.CreateLeases(ranges);
this.partitionManager = new PartitionManager <DocumentServiceLease>(this.HostName, this.leaseManager, this.options);
await this.partitionManager.SubscribeAsync(this);
await this.partitionManager.InitializeAsync();
}
async Task AddLeaseAsync(T lease)
{
if (this.currentlyOwnedPartitions.TryAdd(lease.PartitionId, lease))
{
bool failedToInitialize = false;
try
{
TraceLog.Informational(string.Format("Host '{0}' opening event processor for PartitionId '{1}' and lease token '{2}'", this.workerName, lease.PartitionId, lease.ConcurrencyToken));
await this.partitionObserverManager.NotifyPartitionAcquiredAsync(lease);
TraceLog.Informational(string.Format("Host '{0}' opened event processor for PartitionId '{1}' and lease token '{2}'", this.workerName, lease.PartitionId, lease.ConcurrencyToken));
}
catch (Exception ex)
{
TraceLog.Informational(string.Format("Host '{0}' failed to initialize processor for PartitionId '{1}' and lease token '{2}'", this.workerName, lease.PartitionId, lease.ConcurrencyToken));
failedToInitialize = true;
// Eat any exceptions during notification of observers
TraceLog.Exception(ex);
}
// We need to release the lease if we fail to initialize the processor, so some other node can pick up the parition
if (failedToInitialize)
{
await this.RemoveLeaseAsync(lease, true, ChangeFeedObserverCloseReason.ObserverError);
}
}
else
{
// We already acquired lease for this partition but it looks like we previously owned this partition
// and haven't completed the shutdown process for it yet. Release lease for possible others hosts to
// pick it up.
try
{
TraceLog.Warning(string.Format("Host '{0}' unable to add PartitionId '{1}' with lease token '{2}' to currently owned partitions.", this.workerName, lease.PartitionId, lease.ConcurrencyToken));
await this.leaseManager.ReleaseAsync(lease);
TraceLog.Informational(string.Format("Host '{0}' successfully released lease on PartitionId '{1}' with lease token '{2}'", this.workerName, lease.PartitionId, lease.ConcurrencyToken));
}
catch (LeaseLostException)
{
// We have already shutdown the processor so we can ignore any LeaseLost at this point
TraceLog.Informational(string.Format("Host '{0}' failed to release lease for PartitionId '{1}' with lease token '{2}' due to conflict.", this.workerName, lease.PartitionId, lease.ConcurrencyToken));
}
catch (Exception ex)
{
TraceLog.Exception(ex);
}
}
}
async Task InitializeAsync()
{
this.documentClient = new DocumentClient(this.collectionLocation.Uri, this.collectionLocation.MasterKey, this.collectionLocation.ConnectionPolicy);
Uri databaseUri = UriFactory.CreateDatabaseUri(this.collectionLocation.DatabaseName);
Database database = await this.documentClient.ReadDatabaseAsync(databaseUri);
Uri collectionUri = UriFactory.CreateDocumentCollectionUri(this.collectionLocation.DatabaseName, this.collectionLocation.CollectionName);
DocumentCollection collection = await this.documentClient.ReadDocumentCollectionAsync(collectionUri);
this.collectionSelfLink = collection.SelfLink;
// Beyond this point all access to colleciton is done via this self link: if collection is removed, we won't access new one using same name by accident.
this.leasePrefix = string.Format(CultureInfo.InvariantCulture, "{0}_{1}_{2}", this.collectionLocation.Uri.Host, database.ResourceId, collection.ResourceId);
var leaseManager = new DocumentServiceLeaseManager(
this.auxCollectionLocation,
this.leasePrefix,
this.options.LeaseExpirationInterval,
this.options.LeaseRenewInterval);
await leaseManager.InitializeAsync();
this.leaseManager = leaseManager;
this.checkpointManager = (ICheckpointManager)leaseManager;
if (this.options.DiscardExistingLeases)
{
TraceLog.Warning(string.Format("Host '{0}': removing all leases, as requested by ChangeFeedHostOptions", this.HostName));
await this.leaseManager.DeleteAllAsync();
}
// Note: lease store is never stale as we use monitored colleciton Rid as id prefix for aux collection.
// Collection was removed and re-created, the rid would change.
// If it's not deleted, it's not stale. If it's deleted, it's not stale as it doesn't exist.
await this.leaseManager.CreateLeaseStoreIfNotExistsAsync();
string[] rangeIds = await this.EnumPartitionKeyRangeIds(this.collectionSelfLink);
Parallel.ForEach(rangeIds, async rangeId => {
this.statsSinceLastCheckpoint.AddOrUpdate(
rangeId,
new CheckpointStats(),
(partitionId, existingStats) => existingStats);
await this.leaseManager.CreateLeaseIfNotExistAsync(rangeId);
});
this.partitionManager = new PartitionManager <DocumentServiceLease>(this.HostName, this.leaseManager, this.options);
await this.partitionManager.SubscribeAsync(this);
await this.partitionManager.InitializeAsync();
}
/// <summary>
/// Asynchronously checks the current existing leases and calculates an estimate of remaining work per leased partitions.
/// </summary>
/// <returns>An estimate amount of remaining documents to be processed</returns>
public async Task <long> GetEstimatedRemainingWork()
{
await this.InitializeAsync();
long remaining = 0;
ChangeFeedOptions options = new ChangeFeedOptions
{
MaxItemCount = 1
};
foreach (DocumentServiceLease existingLease in await this.leaseManager.ListLeases())
{
options.PartitionKeyRangeId = existingLease.PartitionId;
options.RequestContinuation = existingLease.ContinuationToken;
IDocumentQuery <Document> query = this.documentClient.CreateDocumentChangeFeedQuery(this.collectionSelfLink, options);
FeedResponse <Document> response = null;
try
{
response = await query.ExecuteNextAsync <Document>();
long parsedLSNFromSessionToken = TryConvertToNumber(ParseAmountFromSessionToken(response.SessionToken));
long lastSequenceNumber = response.Count > 0 ? TryConvertToNumber(response.First().GetPropertyValue <string>(LSNPropertyName)) : parsedLSNFromSessionToken;
long partitionRemaining = parsedLSNFromSessionToken - lastSequenceNumber;
remaining += partitionRemaining < 0 ? 0 : partitionRemaining;
}
catch (DocumentClientException ex)
{
ExceptionDispatchInfo exceptionDispatchInfo = ExceptionDispatchInfo.Capture(ex);
DocumentClientException dcex = (DocumentClientException)exceptionDispatchInfo.SourceException;
if ((StatusCode.NotFound == (StatusCode)dcex.StatusCode && SubStatusCode.ReadSessionNotAvailable != (SubStatusCode)GetSubStatusCode(dcex)) ||
StatusCode.Gone == (StatusCode)dcex.StatusCode)
{
// We are not explicitly handling Splits here to avoid any collision with an Observer that might have picked this up and managing the split
TraceLog.Error(string.Format("GetEstimateWork > Partition {0}: resource gone (subStatus={1}).", existingLease.PartitionId, GetSubStatusCode(dcex)));
}
else if (StatusCode.TooManyRequests == (StatusCode)dcex.StatusCode ||
StatusCode.ServiceUnavailable == (StatusCode)dcex.StatusCode)
{
TraceLog.Warning(string.Format("GetEstimateWork > Partition {0}: retriable exception : {1}", existingLease.PartitionId, dcex.Message));
}
else
{
TraceLog.Error(string.Format("GetEstimateWork > Partition {0}: Unhandled exception", ex.Error.Message));
}
}
}
return(remaining);
}
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();
}
}
private static long TryConvertToNumber(string number)
{
if (string.IsNullOrEmpty(number))
{
return(0);
}
long parsed = 0;
if (!long.TryParse(number, NumberStyles.Any, CultureInfo.InvariantCulture, out parsed))
{
TraceLog.Warning(string.Format(CultureInfo.InvariantCulture, "Cannot parse number '{0}'.", number));
return(0);
}
return(parsed);
}
/// <summary>
/// Create leases for new partitions and take care of split partitions.
/// </summary>
private async Task CreateLeases(IDictionary <string, PartitionKeyRange> ranges)
{
Debug.Assert(ranges != null);
// Get leases after getting ranges, to make sure that no other hosts checked in continuation for split partition after we got leases.
var existingLeases = new Dictionary <string, DocumentServiceLease>();
foreach (var lease in await this.leaseManager.ListLeases())
{
existingLeases.Add(lease.PartitionId, lease);
}
var gonePartitionIds = new HashSet <string>();
foreach (var partitionId in existingLeases.Keys)
{
if (!ranges.ContainsKey(partitionId))
{
gonePartitionIds.Add(partitionId);
}
}
var addedPartitionIds = new List <string>();
foreach (var range in ranges)
{
if (!existingLeases.ContainsKey(range.Key))
{
addedPartitionIds.Add(range.Key);
}
}
// Create leases for new partitions, if there was split, use continuation from parent partition.
var parentIdToChildLeases = new ConcurrentDictionary <string, ConcurrentQueue <DocumentServiceLease> >();
await addedPartitionIds.ForEachAsync(
async addedRangeId =>
{
this.statsSinceLastCheckpoint.AddOrUpdate(
addedRangeId,
new CheckpointStats(),
(partitionId, existingStats) => existingStats);
string continuationToken = null;
string parentIds = string.Empty;
var range = ranges[addedRangeId];
if (range.Parents != null && range.Parents.Count > 0) // Check for split.
{
foreach (var parentRangeId in range.Parents)
{
if (gonePartitionIds.Contains(parentRangeId))
{
// Transfer continiation from lease for gone parent to lease for its child partition.
Debug.Assert(existingLeases[parentRangeId] != null);
parentIds += parentIds.Length == 0 ? parentRangeId : "," + parentRangeId;
if (continuationToken != null)
{
TraceLog.Warning(string.Format("Partition {0}: found more than one parent, new continuation '{1}', current '{2}', will use '{3}'", addedRangeId, existingLeases[parentRangeId].ContinuationToken, existingLeases[parentRangeId].ContinuationToken));
}
continuationToken = existingLeases[parentRangeId].ContinuationToken;
}
}
}
bool wasCreated = await this.leaseManager.CreateLeaseIfNotExistAsync(addedRangeId, continuationToken);
if (wasCreated)
{
if (parentIds.Length == 0)
{
TraceLog.Informational(string.Format("Created lease for partition '{0}', continuation '{1}'.", addedRangeId, continuationToken));
}
else
{
TraceLog.Informational(string.Format("Created lease for partition '{0}' as child of split partition(s) '{1}', continuation '{2}'.", addedRangeId, parentIds, continuationToken));
}
}
else
{
TraceLog.Warning(string.Format("Some other host created lease for '{0}' as child of split partition(s) '{1}', continuation '{2}'.", addedRangeId, parentIds, continuationToken));
}
},
this.options.DegreeOfParallelism);
// Remove leases for splitted (and thus gone partitions) and update continuation token.
await gonePartitionIds.ForEachAsync(
async goneRangeId =>
{
await this.leaseManager.DeleteAsync(existingLeases[goneRangeId]);
TraceLog.Informational(string.Format("Deleted lease for gone (splitted) partition '{0}', continuation '{1}'", goneRangeId, existingLeases[goneRangeId].ContinuationToken));
CheckpointStats removedStatsUnused;
this.statsSinceLastCheckpoint.TryRemove(goneRangeId, out removedStatsUnused);
},
this.options.DegreeOfParallelism);
}
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 LeaseRenewer()
{
while (this.isStarted == 1 || !this.shutdownComplete)
{
try
{
TraceLog.Informational(string.Format("Host '{0}' starting renewal of Leases.", this.workerName));
ConcurrentBag <T> renewedLeases = new ConcurrentBag <T>();
ConcurrentBag <T> failedToRenewLeases = new ConcurrentBag <T>();
List <Task> renewTasks = new List <Task>();
// Renew leases for all currently owned partitions in parallel
foreach (T lease in this.currentlyOwnedPartitions.Values)
{
renewTasks.Add(this.RenewLeaseAsync(lease).ContinueWith(renewResult =>
{
if (renewResult.Result != null)
{
renewedLeases.Add(renewResult.Result);
}
else
{
// Keep track of all failed attempts to renew so we can trigger shutdown for these partitions
failedToRenewLeases.Add(lease);
}
}));
}
// Renew leases for all partitions currently in shutdown
List <T> failedToRenewShutdownLeases = new List <T>();
foreach (T shutdownLeases in this.keepRenewingDuringClose.Values)
{
renewTasks.Add(this.RenewLeaseAsync(shutdownLeases).ContinueWith(renewResult =>
{
if (renewResult.Result != null)
{
renewedLeases.Add(renewResult.Result);
}
else
{
// Keep track of all failed attempts to renew shutdown leases so we can remove them from further renew attempts
failedToRenewShutdownLeases.Add(shutdownLeases);
}
}));
}
// Wait for all renews to complete
await Task.WhenAll(renewTasks.ToArray());
// Update renewed leases.
foreach (T lease in renewedLeases)
{
bool updateResult = this.currentlyOwnedPartitions.TryUpdate(lease.PartitionId, lease, lease);
if (!updateResult)
{
TraceLog.Warning(string.Format("Host '{0}' Renewed lease {1} but failed to update it in the map (ignorable).", this.workerName, lease));
}
}
// Trigger shutdown of all partitions we failed to renew leases
await failedToRenewLeases.ForEachAsync(
async lease => await this.RemoveLeaseAsync(lease, false, ChangeFeedObserverCloseReason.LeaseLost),
this.options.DegreeOfParallelism);
// Now remove all failed renewals of shutdown leases from further renewals
foreach (T failedToRenewShutdownLease in failedToRenewShutdownLeases)
{
T removedLease = null;
this.keepRenewingDuringClose.TryRemove(failedToRenewShutdownLease.PartitionId, out removedLease);
}
await Task.Delay(this.options.LeaseRenewInterval, this.leaseRenewerCancellationTokenSource.Token);
}
catch (OperationCanceledException)
{
TraceLog.Informational(string.Format("Host '{0}' Renewer task canceled.", this.workerName));
}
catch (Exception ex)
{
TraceLog.Exception(ex);
}
}
this.currentlyOwnedPartitions.Clear();
this.keepRenewingDuringClose.Clear();
TraceLog.Informational(string.Format("Host '{0}' Renewer task completed.", this.workerName));
}
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); });
}