TCP RST in varnish reply

[Carsten Ranfeld]

Hello David,

Am Freitag, den 11.09.2009, 18:40 +0200 schrieb David Martínez Moreno:
[...]
> > 4. varnish replies to the client - and here the problem occurs - in
> > the middle of the HTTP response the TCP connection is reset by 
> > varnish (or the machine). So content is not delivered fully. varnish
> > log doesn't show any errors. Tests using different clients show a
> > comparable behavior - TCP RST and partial delivered content, just
> > the size of the content delivered differs.
> >
> >
> >
> > Did anyone experience a similar problem? I searched through
> > documentation, mailing lists and Google. Unfortunately not a helpful
> > page was found.
> >
> > Here the important configuration parts:
[...]
> 
> Hello, Carsten, can you run and post the following in the proxy?
> 
> sysctl -a|grep tw
> 
Sure:

net.ipv4.tcp_max_tw_buckets = 180000
net.ipv4.tcp_tw_recycle = 0
net.ipv4.tcp_tw_reuse = 0

As far as I know there should be a log message created, if the
tw_bucket_limit was hit. We didn't found anything similar.

> Do you have any other sysctl setting altered from vanilla kernel?
> 
We have standard settings:

net.ipv4.conf.default.rp_filter=1
net.ipv4.icmp_echo_ignore_broadcasts = 1

and modified settings:

net.core.rmem_max=8388608
net.core.wmem_max=8388608
net.core.rmem_default=8388608
net.core.wmem_default=8388608
net.ipv4.tcp_rmem=4096 87380 8388608
net.ipv4.tcp_wmem=4096 16384 8388608
net.ipv4.tcp_mem=196608 262144 8388608


Best regards,
Carsten