On Tue, 31 May 2011 11:28:49 -0700, Rich Rauenzahn wrote:
Version : 3.1.9
We're getting (fairly consistently) hanging transfers on a particular
resource. We're using squid as an accelerator, and it is all
internal
traffic within our corporate intranet.
I found this,
http://wiki.squid-cache.org/KnowledgeBase/BrokenWindowSize,
but I'm skeptical because (a) we shouldn't have any firewalls
intercepting the traffic,
NP: There is a firewall on every box the packets flow through. Whether
you know its there or not. These issues are most noticeable when active
alterations are done to the packet, but can occur due to TCP stack bugs
as well.
and (b) when it happens, restarting the
squid service fixes the issue.
Also, it will just be a particular resource/URL -- a user reports the
URL and I can reproduce it with wget -- the transfer will hit around
7% then pretty much stall indefinitely. If I hit another squid in
our
cluster, it works fine.
Anyone see this kind of behavior before?
Only on 3.2 betas (now resolved) or when using ICAP (couple of open
bugs there).
Can you replicate this with 3.1.12 or later? I think you will enjoy the
reduction in memory leaks since 3.1.9 either way :)
Amos
