Every once in a while, our varnish daemon stops responding and has to be restarted. Before I restart it, I dump the contents of the shared memory segment to disk using this command:
varnishlog -w /tmp/blah.log
When I take a look at the contents of the log later, it consists almost exclusively of lines that look like this (~1350 in total out of ~1380 lines):
0 WorkThread - 0x48302cf0 start
Is there something that this is telling me? Previously we've had problems with bots bogging down the backend connections to Apache and causing similar hangs, but in this case, the Apache logs are clean.