Re: [squid-users] Server outputs "TCP: time wait bucket table overflow" in syslog and crashes

On 31.05.2012 01:26, Christoph Fink wrote:
> Hi Squid Users,
>
> I have a server running Debian Squeeze and Squid3 (version 3.1.6 from
> Debian repo). It is configured using winbind+ntlm for authentication.
> The server ran fine for a while now (I'd say 6 months more or less)
> but
> now I get a lot of "TCP: time wait bucket table overflow" errors in
> syslog. After ~30 minutes the system then begins killing processes
> and
> reboots. First I thought that the problem was that the server had not
> enough RAM. I upgraded from 512MB to 1GB but the problem still
> existed.

Something unspecified is still consuming all the RAM you gave the box.

Might be Squid, 3.1.6 has several memory leaks fixed in the latest
releases. Try installing squid 3.1.19 package from the Wheezy/Testing
repositories. It has a lot of NTLM connection fixes as well as better
memory management.

> Then I controlled on the VMware Host (the server is a virtual machine
> running on a ESX Cluster) about the performance datas and I saw that
> performance was not the issue (Memory, CPU and all the other counters
> are at 30% max.). About 50 people are using that proxy Server.

Number of end-users is irrelevant to HTTP. Number of requests per
second is the metric which counts.
How many requests per minute/second is Squid cache manager reporting?

>
> There are no other services running on that Server. It's just squid3,
> samba (for winbind) and apache (only hosting the proxy.pac).
>
> Squid Configuration file:
> http://pastebin.com/5yngPt3c
>

What is this "use_negotiate_ntlm" configuration? Not something squid
supports.

Amos
Received on Thu May 31 2012 - 00:22:49 MDT