_changes feed timeouts an SG stuck

We are running test with 200 clients and 2000 docs. Each client pulls the 200 docs and then pushes a conflicting change to every doc.

What happens to sync_gateway during this test is that _changes longpoll starts taking ages, like 10 minutes or so and effectively the whole test just gets stucks, because longpoll timeouts everytime and never gets any changes.

If we stop the test and then wait a very long time (like 2 hours), we start getting changes using cURL eventually. So it seems that some “housekeeping” is going on during the test and it saturates the whole thing?

Is this a known issue, that large amount of conflicts jams the SG?


Can you upload your Sync Gateway logs to a github gist (or similar) and add a link here.

I was wondering would it be more usefull the run the go profiling on the sync_gateway when it gets stuck like this? I already tried but the problem was that the official SG 1.2 I am using might not have the symbols? To be able to profile using the admin REST api _profile system, do I need to build SG by myself?

(We are running on Ubuntu 14.04 btw.)