We started using MongoDB pretty recently. We are on version 2.6.3. We have a replica set of 4 nodes (1 Primary + 3 Secondary)

We are currently seeing some latency in replication. The Secondary nodes seem to be falling behind primary by a few minutes (2 mins to 20 minutes).

Looking at the log file of Secondary nodes, I do see socket timeouts.

2014-10-01T08:42:03.485-0400 [rsBackgroundSync] Socket recv() timeout 2014-10-01T08:42:03.485-0400 [rsBackgroundSync] SocketException: remote: error: 9001 socket exception [RECV_TIMEOUT] server [] 2014-10-01T08:42:03.485-0400 [rsBackgroundSync] DBClientCursor::init call() failed 2014-10-01T08:42:03.485-0400 [rsBackgroundSync] replSet sync source problem: 10276 DBClientBase::findN: transport error: XSNW10K832W:27017 ns: local.oplog.rs query: {} 2014-10-01T08:42:03.485-0400 [rsBackgroundSync] replSet syncing to: XSNW10K832W:27017 2014-10-01T08:42:23.110-0400 [rsBackgroundSync] replset setting syncSourceFeedback to XSNW10K832W:27017

Here is what I see in the log file of Primary.

2014-10-01T08:42:23.260-0400 [conn492807] getmore local.oplog.rs cursorid:95505391617 ntoreturn:0 keyUpdates:0 numYields:1 locks(micros) r:8846 nreturned:1 reslen:1629780 108434ms 2014-10-01T08:42:23.260-0400 [conn492807] end connection (551 connections now open) 2014-10-01T08:42:23.276-0400 [conn496810] getmore local.oplog.rs cursorid:96066510233 ntoreturn:0 keyUpdates:0 numYields:2 locks(micros) r:4403 nreturned:1 reslen:1629780 108438ms 2014-10-01T08:42:23.276-0400 [conn496810] end connection (550 connections now open) 2014-10-01T08:42:23.276-0400 [conn496812] getmore local.oplog.rs cursorid:95180580732 ntoreturn:0 keyUpdates:0 numYields:1 locks(micros) r:2960 nreturned:1 reslen:1629780 108440ms 2014-10-01T08:42:23.276-0400 [conn496812] end connection (549 connections now open) 2014-10-01T08:42:23.291-0400 [conn496843] end connection (548 connections now open)

Any insight on what is causing the socket timeouts and how to identify/fix the problem?

  • We'll need more information to diagnose this, so it probably won't fit into a DBA Q&A and may be more suited to a thread on mongodb-user. For example: is this behind a mongos (sharded) or is it a standalone replica set? Is this Linux? Do you see any unable to allocate resource (or similar) errors in the primary/secondary logs? What are your ulimit settings? Are all the instances on separate hosts? Is your network oversubsribed, have you tested? - and that's just a start
    – Adam C
    Oct 4, 2014 at 11:52
  • One more comment - even numbers in a replica set is generally a bad idea (see stackoverflow.com/a/26179419/1148648) - you should add an arbiter to bring it up to 5
    – Adam C
    Oct 4, 2014 at 11:54

1 Answer 1


Just a shot in the dark. Could it be a problem with the primary servicing too many connectionss simultaneously?

enter image description here

This information comes from the mongodb 202 training.

If a primary can no longer service requests, there will be

  1. socket exceptions

  2. timeouts

  3. the over servers can no longer ping the primary possibly causing an election --maxConns limit connections to mongos can insulate mongods from getting into unpredicatble states

From MongoDB best practices

Connections MongoDB drivers implement connection pooling to facilitate efficient use of resources. Each connection consumes 1MB of RAM, so be careful to monitor the total number of connections so they do not over-whelm the available RAM and reduce the availablememory for the working set. This typically happenswhen client applicationsdo not properly close their connections, or with Java in particular, that relies ongarbage collection to close the connections.

If there are too many connections open on the primary at any one given time, secondaries, administrative, clients, etc, then your primary could also be running out of memory.

Here is an example layout, with connections coming from drivers, secondaries, to and from different processes. This can all quickly add up depending on the number of clients and machines in your infrastructure!

enter image description here

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.