private void AddReplicationPerformance(OutgoingReplicationStatsAggregator stats)
{
_lastReplicationStats.Enqueue(stats);
while (_lastReplicationStats.Count > 25)
{
_lastReplicationStats.TryDequeue(out _);
}
}
private void ReplicateToDestination()
{
try
{
AddReplicationPulse(ReplicationPulseDirection.OutgoingInitiate);
NativeMemory.EnsureRegistered();
if (_log.IsInfoEnabled)
{
_log.Info($"Will replicate to {Destination.FromString()} via {_connectionInfo.Url}");
}
using (_parent._server.ContextPool.AllocateOperationContext(out TransactionOperationContext context))
using (context.OpenReadTransaction())
{
var record = _parent.LoadDatabaseRecord();
if (record == null)
{
throw new InvalidOperationException($"The database record for {_parent.Database.Name} does not exist?!");
}
if (record.Encrypted && Destination.Url.StartsWith("https:", StringComparison.OrdinalIgnoreCase) == false)
{
throw new InvalidOperationException(
$"{record.DatabaseName} is encrypted, and require HTTPS for replication, but had endpoint with url {Destination.Url} to database {Destination.Database}");
}
}
var task = TcpUtils.ConnectSocketAsync(_connectionInfo, _parent._server.Engine.TcpConnectionTimeout, _log);
task.Wait(CancellationToken);
using (Interlocked.Exchange(ref _tcpClient, task.Result))
{
var wrapSsl = TcpUtils.WrapStreamWithSslAsync(_tcpClient, _connectionInfo, _parent._server.Server.Certificate.Certificate, _parent._server.Engine.TcpConnectionTimeout);
wrapSsl.Wait(CancellationToken);
using (_stream = wrapSsl.Result) // note that _stream is being disposed by the interruptible read
using (_interruptibleRead = new InterruptibleRead(_database.DocumentsStorage.ContextPool, _stream))
using (_buffer = JsonOperationContext.ManagedPinnedBuffer.LongLivedInstance())
{
var documentSender = new ReplicationDocumentSender(_stream, this, _log);
WriteHeaderToRemotePeer();
//handle initial response to last etag and staff
try
{
var response = HandleServerResponse(getFullResponse: true);
switch (response.ReplyType)
{
//The first time we start replication we need to register the destination current CV
case ReplicationMessageReply.ReplyType.Ok:
LastAcceptedChangeVector = response.Reply.DatabaseChangeVector;
break;
case ReplicationMessageReply.ReplyType.Error:
var exception = new InvalidOperationException(response.Reply.Exception);
if (response.Reply.Exception.Contains(nameof(DatabaseDoesNotExistException)) ||
response.Reply.Exception.Contains(nameof(DatabaseNotRelevantException)))
{
AddReplicationPulse(ReplicationPulseDirection.OutgoingInitiateError, "Database does not exist");
DatabaseDoesNotExistException.ThrowWithMessageAndException(Destination.Database, response.Reply.Message, exception);
}
AddReplicationPulse(ReplicationPulseDirection.OutgoingInitiateError, $"Got error: {response.Reply.Exception}");
throw exception;
}
}
catch (DatabaseDoesNotExistException e)
{
var msg = $"Failed to parse initial server replication response, because there is no database named {_database.Name} " +
"on the other end. ";
if (_external)
{
msg += "In order for the replication to work, a database with the same name needs to be created at the destination";
}
var young = (DateTime.UtcNow - _startedAt).TotalSeconds < 30;
if (young)
{
msg += "This can happen if the other node wasn't yet notified about being assigned this database and should be resolved shortly.";
}
if (_log.IsInfoEnabled)
{
_log.Info(msg, e);
}
AddReplicationPulse(ReplicationPulseDirection.OutgoingInitiateError, msg);
// won't add an alert on young connections
// because it may take a few seconds for the other side to be notified by
// the cluster that it has this db.
if (young == false)
{
AddAlertOnFailureToReachOtherSide(msg, e);
}
throw;
}
catch (OperationCanceledException e)
{
const string msg = "Got operation canceled notification while opening outgoing replication channel. " +
"Aborting and closing the channel.";
if (_log.IsInfoEnabled)
{
_log.Info(msg, e);
}
AddReplicationPulse(ReplicationPulseDirection.OutgoingInitiateError, msg);
throw;
}
catch (Exception e)
{
var msg = $"{OutgoingReplicationThreadName} got an unexpected exception during initial handshake";
if (_log.IsInfoEnabled)
{
_log.Info(msg, e);
}
AddReplicationPulse(ReplicationPulseDirection.OutgoingInitiateError, msg);
AddAlertOnFailureToReachOtherSide(msg, e);
throw;
}
DateTime nextReplicateAt = default(DateTime);
while (_cts.IsCancellationRequested == false)
{
while (_database.Time.GetUtcNow() > nextReplicateAt)
{
if (_parent.DebugWaitAndRunReplicationOnce != null)
{
_parent.DebugWaitAndRunReplicationOnce.Wait(_cts.Token);
_parent.DebugWaitAndRunReplicationOnce.Reset();
}
var sp = Stopwatch.StartNew();
var stats = _lastStats = new OutgoingReplicationStatsAggregator(_parent.GetNextReplicationStatsId(), _lastStats);
AddReplicationPerformance(stats);
AddReplicationPulse(ReplicationPulseDirection.OutgoingBegin);
try
{
using (var scope = stats.CreateScope())
{
try
{
if (Destination is InternalReplication dest)
{
_parent.EnsureNotDeleted(dest.NodeTag);
}
var didWork = documentSender.ExecuteReplicationOnce(scope, ref nextReplicateAt);
if (didWork == false)
{
break;
}
if (Destination is ExternalReplication externalReplication)
{
var taskId = externalReplication.TaskId;
UpdateExternalReplicationInfo(taskId);
}
DocumentsSend?.Invoke(this);
if (sp.ElapsedMilliseconds > 60 * 1000)
{
_waitForChanges.Set();
break;
}
}
catch (OperationCanceledException)
{
// cancellation is not an actual error,
// it is a "notification" that we need to cancel current operation
const string msg = "Operation was canceled.";
AddReplicationPulse(ReplicationPulseDirection.OutgoingError, msg);
throw;
}
catch (Exception e)
{
AddReplicationPulse(ReplicationPulseDirection.OutgoingError, e.Message);
scope.AddError(e);
throw;
}
}
}
finally
{
stats.Complete();
AddReplicationPulse(ReplicationPulseDirection.OutgoingEnd);
}
}
//if this returns false, this means either timeout or canceled token is activated
while (WaitForChanges(_parent.MinimalHeartbeatInterval, _cts.Token) == false)
{
//If we got cancelled we need to break right away
if (_cts.IsCancellationRequested)
{
break;
}
// open tx
// read current change vector compare to last sent
// if okay, send cv
using (_database.DocumentsStorage.ContextPool.AllocateOperationContext(out DocumentsOperationContext ctx))
using (var tx = ctx.OpenReadTransaction())
{
var etag = DocumentsStorage.ReadLastEtag(tx.InnerTransaction);
if (etag == _lastSentDocumentEtag)
{
SendHeartbeat(DocumentsStorage.GetDatabaseChangeVector(ctx));
_parent.CompleteDeletionIfNeeded();
}
else if (nextReplicateAt > DateTime.UtcNow)
{
SendHeartbeat(null);
}
else
{
//Send a heartbeat first so we will get an updated CV of the destination
var currentChangeVector = DocumentsStorage.GetDatabaseChangeVector(ctx);
SendHeartbeat(null);
//If our previous CV is already merged to the destination wait a bit more
if (ChangeVectorUtils.GetConflictStatus(LastAcceptedChangeVector, currentChangeVector) ==
ConflictStatus.AlreadyMerged)
{
continue;
}
// we have updates that we need to send to the other side
// let's do that..
// this can happen if we got replication from another node
// that we need to send to it. Note that we typically
// will wait for the other node to send the data directly to
// our destination, but if it doesn't, we'll step in.
// In this case, we try to limit congestion in the network and
// only send updates that we have gotten from someone else after
// a certain time, to let the other side tell us that it already
// got it. Note that this is merely an optimization to reduce network
// traffic. It is fine to have the same data come from different sources.
break;
}
}
}
_waitForChanges.Reset();
}
}
}
}
catch (AggregateException e)
{
if (e.InnerExceptions.Count == 1)
{
if (e.InnerException is OperationCanceledException oce)
{
HandleOperationCancelException(oce);
}
if (e.InnerException is IOException ioe)
{
HandleIOException(ioe);
}
}
HandleException(e);
}
catch (OperationCanceledException e)
{
HandleOperationCancelException(e);
}
catch (IOException e)
{
HandleIOException(e);
}
catch (Exception e)
{
HandleException(e);
}
void HandleOperationCancelException(OperationCanceledException e)
{
if (_log.IsInfoEnabled)
{
_log.Info($"Operation canceled on replication thread ({FromToString}). " +
$"This is not necessary due to an issue. Stopped the thread.");
}
if (_cts.IsCancellationRequested == false)
{
Failed?.Invoke(this, e);
}
}
void HandleIOException(IOException e)
{
if (_log.IsInfoEnabled)
{
if (e.InnerException is SocketException)
{
_log.Info($"SocketException was thrown from the connection to remote node ({FromToString}). " +
$"This might mean that the remote node is done or there is a network issue.", e);
}
else
{
_log.Info($"IOException was thrown from the connection to remote node ({FromToString}).", e);
}
}
Failed?.Invoke(this, e);
}
void HandleException(Exception e)
{
if (_log.IsInfoEnabled)
{
_log.Info($"Unexpected exception occurred on replication thread ({FromToString}). " +
$"Replication stopped (will be retried later).", e);
}
Failed?.Invoke(this, e);
}
}