[ https://issues.apache.org/jira/browse/TS-3778?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14640539#comment-14640539 ]
Susan Hinrichs commented on TS-3778: ------------------------------------ Reverted TS-3640 and TS-3584 > Memory leak caused by the TS-3640. > ---------------------------------- > > Key: TS-3778 > URL: https://issues.apache.org/jira/browse/TS-3778 > Project: Traffic Server > Issue Type: Bug > Components: Core > Reporter: Sudheer Vinukonda > Assignee: Sudheer Vinukonda > Priority: Blocker > Fix For: 6.1.0 > > > TS-3640 removed the code that was otherwise always doing a half-close on the > ua_session for POST transactions that had error'ed on the origin connection. > It seems like this allows to preserve the keep alive client connections as > well as drain any body that was still in the buffers. This piece of code is > very important when there's heavy POST involved. > On a related note, I see that TS-3640 was added to fix regressions caused by > TS-3584. But, TS-3584 looks questionable to me too. > We have observed heavy memory leaks with hosts crashing in less than a day > after TS-3640's commit has been deployed (We are in the process of confirming > that the problem is completely resolved by reverting TS-3640. The memory > looks stable enough after about 12 hrs run without TS-3640, running a little > longer to be safe) -- This message was sent by Atlassian JIRA (v6.3.4#6332)