CLOSE_WAIT

Question

Our self-hosted Confluence instance becomes unresponsive and the only solution is to restart it. There's barely any CPU usage and the logs don't show errors, besides the usual (macros and stuff).

There's a large number of connection in CLOSE_WAIT state:

tcp6 775 0 ::1:8090 ::1:32970 CLOSE_WAIT 
tcp6 794 0 ::1:8090 ::1:33048 CLOSE_WAIT 
tcp6 16787 0 ::1:8090 ::1:60580 CLOSE_WAIT 
tcp6 994 0 ::1:8090 ::1:60404 CLOSE_WAIT 
tcp6 16787 0 127.0.0.1:8090 127.0.0.1:36412 CLOSE_WAIT 
tcp6 428 0 127.0.0.1:8090 127.0.0.1:38934 CLOSE_WAIT 
tcp6 703 0 127.0.0.1:8090 127.0.0.1:37592 CLOSE_WAIT

I believe this could indicate an issue in Confluence itself. How should we proceed?

This issue just happened with v6.7.0 but we faced it on earlier versions too.

AnnWorley · Answer

There is usually, but not always, an error in the <Confluence_home>/logs/atlassian-confluence.log that reads something like "cannot open connection". My guess was partly based on the line in your description "This issue just happened with v6.7.0 but we faced it on earlier versions too."

The database connection pool default for new installs of Confluence 6.7 is 60 and that line made me think yours was an upgraded instance. :)

I hope the performance issue is resolved. We shall see in time.

Giovanni Tirloni · Answer

Thanks for the pointers. The DB pool was indeed too low (at 30) and I've increased it. If that ends up being the root cause, I just wish there error messages but maybe I've missed something. Thanks again!

AnnWorley · Answer

Based on the symptoms, I am guessing that Confluence could be running out of database connections. To diagnose and resolve this, please see Confluence slows and times out during periods of high load due to database connection pool

The article recommends taking thread dumps during the performance issue. If you don't find the database connection issue described in the article, please see HOW TO: Analyse Thread Dumps in case the thread dumps can point you to another diagnosis.

If that doesn't turn out to be the issue, please also check for errors in the <Confluence_home>/logs/atlassian-confluence.log and let us know what you find during the unresponsive times.