On 4/07/2013 7:00 p.m., Grant wrote:
Are your file descriptors limit correctly configured ? Let squid have 6
descriptors per client minimum to be cool.
Is squid limit under user limit ?
I have:
# ulimit -n
1024
And even as latency skyrockets I only have:
# lsof -u squid | wc -l
80
Is it a problem that the client is behind a router which I have no
control over? Does squid need to establish an inbound connection to
the client?
No, and no.
Is your Squid using the kqueue I/O module of Squid by chance? there is a
SSL hanging bug in there since forever which only just got fixed in 3.3.6.
The usual cause of these type of issues is forwarding loops, although
your low of socket usage indicates that is probably not the problem.
It may also be the machine virtual memory swapping. Each transaction
through Squid utilizes around 256KB of RAM, plus the cache_mem size,
plus about 10-15MB of cache index data for each GB of the total cache size.
If you are able to replicate it easily, please try to locate an strace
or a cache.log at level-9 to see what Squid is doing during the
unresponsive period.
Amos
