[Pkg-varnish-devel] Bug#709954: Bug#709954: varnish: Varnish consumes 100% CPU without handling requests

Stig Sandbeck Mathisen ssm at debian.org
Mon May 27 05:47:30 UTC 2013


Stefan Freudenberg <stefan at agaric.com> writes:

> During a stress test one of two load balanced varnish servers started
> to consume 100% of one CPU. The instance seemed to be functional, but
> it continued to keep one CPU 100% busy until a restart of varnish.

Hi, thanks for reporting a problem.

I'd ask you to look at the first two parts of
https://www.varnish-cache.org/trac/wiki/DebuggingVarnish for the minimum
required information to debug a varnish problem.

Having the VCL is essential, as well as the request which triggered the
problem, if you have it. Output from the "varnishlog" program would be
helpful with showing what the child process was busy doing with the
request.

I see you added an strace of the main varnish process. In normal
operation, the main process will just "ping" the worker process, and I
assume that it was the child process which spent 100% cpu. At least, the
child process appeared to be responding to pings from the master, and as
you write: it "seemed functional".

Is the problem reproducable? If so, how?

-- 
Stig Sandbeck Mathisen



More information about the Pkg-varnish-devel mailing list