[infinispan-dev] Classloader leaks?

[Dennis Reed]

Are those actually 2 million *unique* descriptors?

I've seen lsof output that listed many duplicates for the same file
descriptor (one for each thread?), making the list appear much larger
than it really was.

-Dennis


On 02/22/2017 02:25 PM, Sanne Grinovero wrote:
> Hi all,
> 
> our documentation suggest to raise the file limits to about 16K:
> http://infinispan.org/docs/stable/contributing/contributing.html#running_and_writing_tests
> 
> I already have this setup since years, yet I've been noticing errors such as:
> 
> "Caused by: java.io.IOException: Too many open files"
> 
> Today I decided to finally have a look, and I see that while running
> the testsuite, my system's consumption of file descriptor raises
> continuously, up to more than 2 millions.
> (When not running the suite, I'm consuming 200K - that's including
> IDEs and other FD hungry systems like Chrome)
> 
> Trying to get some samples of these file descriptors, it looks like
> it's really about open files. Jar files to be more precise.
> 
> What puzzles me is that taking just one jar - jgroups for example - I
> can count 7852 open instances of it, but distributed among a handful
> of processes only.
> 
> My guess is classloaders aren't being closed?
> 
> Also: why did nobody else notice problems? Do you all have
> reconfigured your system for unlimited FDs?
> 
> Thanks,
> Sanne
> _______________________________________________
> infinispan-dev mailing list
> infinispan-dev at lists.jboss.org
> https://lists.jboss.org/mailman/listinfo/infinispan-dev
>