Probably this a bug w/ Virtualbox's NAT setup. The stalling was only
happening to sites on the same subnet as the browser.
mike
On 4/6/2010 1:49 AM, Amos Jeffries wrote:
Mike Leong wrote:
I have this weird problem where if I go though squid, a page would
stall until it hits the read timeout limit. This only happens on
certain webservers.
eg:
read_timeout set to 120 seconds.
1. request page on browser
2. page load stalls
3. after 120 seconds, squid sends a fin, ack to http server
4. http server returns a HTTP 200 after receiving the fin,ack packet
5. browser gets the requested page.
Sometimes, squid would load the page fine.
Here's my squid build:
Squid Cache: Version 2.7.STABLE9
configure options: '--enable-linux-netfilter'
'--enable-follow-x-forwarded-for' '--enable-linux-tproxy'
'--enable-epoll' '--enable-async-io' '--with-pthreads'
'--enable-storeio=ufs,aufs,coss,diskd,null'
'--enable-removal-policies=lru,heap' '--enable-snmp'
'--with-maxfd=65536' 'CFLAGS=-march=pentium3 -O2 -fomit-frame-pointer
-pipe' 'CPPFLAGS=-march=pentium3 -O2 -fomit-frame-pointer -pipe'
You can download tcpdump and squid config from
http://mina.lolipower.org/mike/squid-stall.zip
If I access the page directly on the browser, the page stalling
doesn't happen.
any ideas on addressing this issue?
Looks like a TCP stack issue.
The only difference between those traces is the "bad" one has a FIN/ACK
packet playing double duty to ACK bytes of data and close the link. The
"good" only an empty FIN/ACK following the end of the data.
Amos
