Perlbal memory leak?

dormando dormando at
Fri Feb 10 02:48:05 UTC 2006


I have to run in five minutes, but I wanted to throw this out;

If perbal cannot contact any servers in a pool for a service, it seems 
to bloat memory like crazy then use up 99% CPU. I've seen this happen on 
our busy load balancers because our core switch has crashed twice in the 
last month. Once everything is working again the processes have grown to 
somewhere between 600 megabytes and 1.6 gigabytes, and are no longer 
responding very fast.

I could reproduce it with a pool with a single server in it that is 
unavailable, then hammering it with requests.

Haven't had a moment (nor do I really have a direction) on tracing it 
further right now, though I'm willing to.

Running perlbal 1.39 with the latest Danga::Socket as of that release, 
on top of debian sarge running a custom kernel above 2.6.14. Changes 
inbetween don't seem to be related to this problem, but I am planning on 
upgrading them in a week or two.


More information about the perlbal mailing list