andrew.morin1:

Leaking Lucene indexes can be caused by any number of things. While there have been some versions of JIRA with bugs that triggered this, it is almost always caused by a plugin that is doing something wrong, such as running searches from its own thread pool without using JiraThreadLocalUtil to take care of the cleanup.

Given that there are a wide variety of potential culprits and that it is usually related to a plugin that we do not ourselves test, there is no simple resolution for this issue. Customers that encounter it should open a support request so that we can work with you to track down what is causing it on your particular instance.

What was the resolution for this?

Hi jasbris,

Could you please file a support case at https://support.atlassian.com? that way we can take a look and troubleshoot what is happening in your instance.

Regards,

Oswaldo Hernández.
JIRA Bugmaster.
[Atlassian].

clepetit / matt.doar,

Were you able to resolve this as a support case? We see this both on JIRA 5.2.8 with ~600,000+ issues, and on 6.2.5 with ~500,000+ issues. We often see in the logs:

Tried to reopen the IndexReader, but it threw AlreadyClosedException. Opening a fresh IndexReader.

We also have to reboot our 5.2.8 instance every 3 weeks. We use Fusion I/O SSD cards for our storage on the 5.2.8 instance. This is the instance we really see the open files issue with, but both servers see the error message.

If you run lsof and see your total number of file descriptors increasing day over day, then check the names of the files from lsof. If they're all in the caches/issues directory and you see AlreadyClosedException in your logs files, then this would seem to match

I can't be sure we're seeing the same thing but would appreciate an update on how the investigation is progressing.

When a solution is found for this can you please post it here as well as in the support ticket as we are experiencing this problem also.

Chris, that's fine, thanks. I'd like to have this issue remain public (but Closed) so that others can find it for future reference if possible.

matt.doar, I'd like to move this over to Support if possible. It looks right now like it is something that the developers haven't been able to reproduce with the time and resources available to them (and since technically, we don't offer SLA's etc on jira.atlassian.com (JAC), whilst do we on support.atlassian.com (SAC), it will help us get to the bottom of this). Our Support team on the other hand have the time to work on setting up an environment and replicating it and identifying where the issue is before we finalise a JAC issue, and we're better setup there to ensure it gets resolved, rather than left to languish.

Would it be okay if we moved the investigation into this bug into a Support ticket?

Regards,

Chris Le Petit
Service Enablement

I'm still checking but this is the total list;

These plugins may be configured, enabled, disabled or uninstalled.

Copy Date *
CustomWare JIRA Utilities
Enterprise Mail Handler for JIRA (JEMH) (disabled)
JIRA Agile
JIRA Auditor (disabled)
JIRA Calendar Plugin
JIRA Charting Plugin
JIRA Misc Workflow Extensions
JIRA Project Role Tab
JIRA Subversion plugin
JIRA Suite Utilities
JIRA Toolkit Plugin
Local Helpdesk *
Script Runner
Show Saved Filter With Columns
Template Helper *
valiantys-base-plugin
VertygoSLA - Service Level Agreement

The ones with an asterisk are locally developed. I didn't find any reference to IndexSearcher or IndexManager. One of those add-ons defines a new searcher for finding issues where the last commenter is not the reporter