public Task <bool> EnqueueOperationAsync(ShardWriteOperation data)
{
lock (queueLock)
{
OperationQueue.Add(SystemExtension.Clone(data));
}
return(Task.FromResult(true));
}
public Task <bool> DeleteOperationFromQueueAsync(ShardWriteOperation operation)
{
lock (queueLock)
{
OperationQueue.Remove(operation);
}
return(Task.FromResult(true));
}
public async void ReverseLocalTransaction(Guid shardId, string type, int pos)
{
var operation = await _shardRepository.GetShardWriteOperationAsync(shardId, pos);
if (operation != null)
{
await _shardRepository.AddDataReversionRecordAsync(new DataReversionRecord()
{
OriginalOperation = operation,
NewData = operation.Data,
OriginalData = await _dataRouter.GetDataAsync(type, operation.Data.Id),
RevertedTime = DateTime.Now
});
ShardData data = operation.Data;
SortedDictionary <int, ShardWriteOperation> allOperations;
ShardWriteOperation lastObjectOperation = null;
if (operation.Operation == ShardOperationOptions.Update || operation.Operation == ShardOperationOptions.Delete)
{
allOperations = await _shardRepository.GetAllObjectShardWriteOperationAsync(data.ShardId.Value, data.Id);
lastObjectOperation = allOperations.Where(ao => ao.Key < operation.Pos).Last().Value;
}
switch (operation.Operation)
{
case ShardOperationOptions.Create:
data = await _dataRouter.GetDataAsync(type, operation.Data.Id);
if (data != null)
{
_logger.LogInformation(_nodeStateService.GetNodeLogId() + "Reverting create with deletion of " + data);
await _dataRouter.DeleteDataAsync(data);
}
break;
// Put the data back in the right position
case ShardOperationOptions.Delete:
await _dataRouter.InsertDataAsync(lastObjectOperation.Data);
break;
case ShardOperationOptions.Update:
await _dataRouter.UpdateDataAsync(lastObjectOperation.Data);
break;
}
await _shardRepository.RemoveShardWriteOperationAsync(shardId, operation.Pos);
}
//Update the cache
_shardLastOperationCache.Remove(operation.Data.ShardId.Value, out _);
}
public async Task <bool> EnqueueOperationAsync(ShardWriteOperation transaction)
{
lock (queueLock)
{
OperationQueue.Add(transaction);
}
if (_persistToDisk)
{
return(await _operationCacheRepository.EnqueueOperationAsync(transaction));
}
return(true);
}
public async Task <bool> AddShardWriteOperationAsync(ShardWriteOperation operation)
{
try
{
await _shardWriteOperations.InsertOneAsync(operation);
}
//Return true if the operation already exists
catch (MongoDuplicateKeyException e)
{
return(true);
}
return(true);
}
public async void ApplyOperationToDatastore(ShardWriteOperation operation)
{
switch (operation.Operation)
{
case ShardOperationOptions.Create:
await _dataRouter.InsertDataAsync(operation.Data);
break;
case ShardOperationOptions.Delete:
await _dataRouter.DeleteDataAsync(operation.Data);
break;
case ShardOperationOptions.Update:
await _dataRouter.UpdateDataAsync(operation.Data);
break;
}
}
public async Task <WriteShardDataResponse> WriteShardData(ShardData data, ShardOperationOptions operationType, string operationId, DateTime transactionDate)
{
ShardWriteOperation operation = new ShardWriteOperation()
{
Data = data,
Id = operationId,
Operation = operationType,
TransactionDate = transactionDate
};
ShardWriteOperation lastOperation = await GetOrPopulateOperationCache(operation.Data.ShardId.Value);
//Start at 1
operation.Pos = lastOperation == null ? 1 : lastOperation.Pos + 1;
var hash = lastOperation == null ? "" : lastOperation.ShardHash;
operation.ShardHash = ObjectUtility.HashStrings(hash, operation.Id);
_logger.LogDebug(_nodeStateService.GetNodeLogId() + "writing new operation " + operationId + " with data " + Environment.NewLine + JsonConvert.SerializeObject(data, Formatting.Indented));
//Write the data
var writeOperation = await _shardRepository.AddShardWriteOperationAsync(operation); //Add shard operation
if (writeOperation)
{
ApplyOperationToDatastore(operation);
var shardMetadata = _stateMachine.GetShard(operation.Data.ShardType, operation.Data.ShardId.Value);
//Mark operation as applied
await _shardRepository.MarkShardWriteOperationAppliedAsync(operation.Id);
//Update the cache
UpdateOperationCache(operation.Data.ShardId.Value, operation);
ConcurrentBag <Guid> InvalidNodes = new ConcurrentBag <Guid>();
//All allocations except for your own
var tasks = shardMetadata.InsyncAllocations.Where(id => id != _nodeStateService.Id).Select(async allocation =>
{
try
{
var result = await _clusterClient.Send(allocation, new ReplicateShardWriteOperation()
{
Operation = operation
});
if (result.IsSuccessful)
{
_logger.LogDebug(_nodeStateService.GetNodeLogId() + "Successfully replicated all " + shardMetadata.Id + "shards.");
}
else
{
throw new Exception("Failed to replicate data to shard " + shardMetadata.Id + " to node " + allocation + " for operation " + operation.ToString() + Environment.NewLine + JsonConvert.SerializeObject(operation, Formatting.Indented));
}
}
catch (TaskCanceledException e)
{
_logger.LogError(_nodeStateService.GetNodeLogId() + "Failed to replicate shard " + shardMetadata.Id + " on node " + allocation + " for operation " + operation.Pos + " as request timed out, marking shard as not insync...");
InvalidNodes.Add(allocation);
}
catch (Exception e)
{
_logger.LogError(_nodeStateService.GetNodeLogId() + "Failed to replicate shard " + shardMetadata.Id + " for operation " + operation.Pos + " with error " + e.Message + ", marking shard as not insync..." + Environment.NewLine + e.StackTrace);
InvalidNodes.Add(allocation);
}
});
await Task.WhenAll(tasks);
if (InvalidNodes.Count() > 0)
{
await _clusterClient.Send(new ExecuteCommands()
{
Commands = new List <BaseCommand>()
{
new UpdateShardMetadataAllocations()
{
ShardId = data.ShardId.Value,
Type = data.ShardType,
StaleAllocationsToAdd = InvalidNodes.ToHashSet(),
InsyncAllocationsToRemove = InvalidNodes.ToHashSet()
}
},
WaitForCommits = true
});
_logger.LogInformation(_nodeStateService.GetNodeLogId() + " had stale virtual machines.");
}
return(new WriteShardDataResponse()
{
Pos = operation.Pos,
ShardHash = operation.ShardHash,
IsSuccessful = true
});
}
else
{
return(new WriteShardDataResponse()
{
IsSuccessful = false
});
}
}
private void UpdateOperationCache(Guid shardId, ShardWriteOperation operation)
{
operation.Applied = true;
_shardLastOperationCache.AddOrUpdate(shardId, operation, (key, val) => operation);
}
public async Task <bool> ReplicateShardWriteOperationAsync(ShardWriteOperation operation)
{
// Get the last operation
ShardWriteOperation lastOperation;
var startTime = DateTime.Now;
if (operation.Pos != 1)
{
while ((lastOperation = await GetOrPopulateOperationCache(operation.Data.ShardId.Value)) == null || !lastOperation.Applied || lastOperation.Pos < operation.Pos - 1)
{
//Assume in the next 3 seconds you should receive the previos requests
if ((DateTime.Now - startTime).TotalMilliseconds > 3000)
{
return(false);
}
await Task.Delay(100);
}
if (lastOperation.Pos != operation.Pos - 1)
{
lastOperation = await _shardRepository.GetShardWriteOperationAsync(operation.Data.ShardId.Value, operation.Pos - 1);
}
string newHash;
if ((newHash = ObjectUtility.HashStrings(lastOperation.ShardHash, operation.Id)) != operation.ShardHash)
{
//Critical error with data concurrency, revert back to last known good commit via the resync process
_logger.LogError(_nodeStateService.GetNodeLogId() + "Failed to check the hash for operation " + operation.Id + ", marking shard as out of sync...");
throw new Exception(_nodeStateService.GetNodeLogId() + "Failed to check the hash for operation " + operation.Id + ", marking shard as out of sync...");
}
}
// if (operation.Pos > 1)
//{
// lock (_shardLastOperationCache[operation.Data.ShardId.Value])
// {
//The operation must already be being applied
/* if (_shardLastOperationCache[operation.Data.ShardId.Value].Pos >= operation.Pos)
* {
* throw new Exception("There must be a replication process being run");
* }*/
_shardRepository.AddShardWriteOperationAsync(operation).GetAwaiter().GetResult();
//Run shard operation
ApplyOperationToDatastore(operation);
_shardRepository.MarkShardWriteOperationAppliedAsync(operation.Id).GetAwaiter().GetResult();
//Update the cache
UpdateOperationCache(operation.Data.ShardId.Value, operation);
//}
/* }
* else
* {
* await _shardRepository.AddShardWriteOperationAsync(operation);
* //Run shard operation
* ApplyOperationToDatastore(operation);
* await _shardRepository.MarkShardWriteOperationAppliedAsync(operation.Id);
* //Update the cache
* UpdateOperationCache(operation.Data.ShardId.Value, operation);
* }*/
return(true);
}
public async Task <bool> EnqueueOperationAsync(ShardWriteOperation data)
{
await _shardWriteOperationsQueue.InsertOneAsync(data);
return(true);
}
public async Task <bool> DeleteOperationFromQueueAsync(ShardWriteOperation operation)
{
return((await _shardWriteOperationsQueue.DeleteOneAsync(swoq => swoq.Id == operation.Id)).IsAcknowledged);
}
public async Task <bool> AddOperationToTransitAsync(ShardWriteOperation operation)
{
await _shardWriteOperationsTransit.InsertOneAsync(operation);
return(true);
}
public Task <bool> AddShardWriteOperationAsync(ShardWriteOperation operation)
{
return(Task.FromResult(ShardWriteOperations.TryAdd(operation.Id, SystemExtension.Clone(operation))));
}
public Task <bool> AddOperationToTransitAsync(ShardWriteOperation operation)
{
return(Task.FromResult(TransitQueue.TryAdd(operation.Id, SystemExtension.Clone(operation))));
}
public async Task <bool> SyncShard(Guid shardId, string type)
{
var lastOperation = await _shardRepository.GetShardWriteOperationAsync(shardId, _shardRepository.GetTotalShardWriteOperationsCount(shardId));
int lastOperationPos = lastOperation == null ? 0 : lastOperation.Pos;
var shardMetadata = _stateMachine.GetShard(type, shardId);
var result = await _clusterClient.Send(shardMetadata.PrimaryAllocation, new RequestShardWriteOperations()
{
ShardId = shardId,
From = lastOperationPos,
To = lastOperationPos + 1
});
if (result.IsSuccessful)
{
var totalPositions = _shardRepository.GetTotalShardWriteOperationsCount(shardId);
//If the primary has less operations
if (result.LatestPosition < lastOperationPos)
{
while (totalPositions != result.LatestPosition)
{
_writer.ReverseLocalTransaction(shardId, type, totalPositions);
totalPositions--;
}
}
else
{
//Check whether the hash is equal, if not equal roll back each transaction
var currentPosition = lastOperationPos;
ShardWriteOperation currentOperation = null;
if (lastOperationPos != 0 && !(result.Operations[lastOperationPos].ShardHash == lastOperation.ShardHash))
{
//While the shard position does not match
while (!((await _clusterClient.Send(shardMetadata.PrimaryAllocation, new RequestShardWriteOperations()
{
ShardId = shardId,
From = currentPosition,
To = currentPosition
})).Operations[currentPosition].ShardHash != (currentOperation = await _shardRepository.GetShardWriteOperationAsync(shardId, currentPosition)).ShardHash))
{
_logger.LogInformation("Reverting transaction " + currentOperation.Pos + " on shard " + shardId);
_writer.ReverseLocalTransaction(shardId, type, currentOperation.Pos);
currentPosition--;
}
}
result = await _clusterClient.Send(shardMetadata.PrimaryAllocation, new RequestShardWriteOperations()
{
ShardId = shardId,
From = currentPosition + 1,
To = currentPosition + 50
});
var totalShards = (_shardRepository.GetTotalShardWriteOperationsCount(shardId));
//If you have more operations
if (result.LatestPosition < totalShards)
{
_logger.LogWarning("Detected more nodes locally then primary, revering to latest position");
while (totalShards != result.LatestPosition)
{
_writer.ReverseLocalTransaction(shardId, type, totalShards);
totalShards--;
}
}
while (result.LatestPosition != (_shardRepository.GetTotalShardWriteOperationsCount(shardId)))
{
foreach (var operation in result.Operations)
{
_logger.LogDebug(_nodeStateService.Id + "Replicated operation " + operation.Key + " for shard " + shardId);
await _writer.ReplicateShardWriteOperationAsync(operation.Value);
}
currentPosition = result.Operations.Last().Key;
result = await _clusterClient.Send(shardMetadata.PrimaryAllocation, new RequestShardWriteOperations()
{
ShardId = shardId,
From = currentPosition + 1,
To = currentPosition + 50
});
}
}
_logger.LogInformation("Successfully recovered data on shard " + shardId);
return(true);
}
else
{
throw new Exception("Failed to fetch shard from primary.");
}
}
