Hi,
Le 18/05/2012 00:28, Yehuda Sadeh a écrit :
Hi,
We have a setup where there's haproxy (1.4.15) in front of a few
apache servers. We have a test suite that occasionally fails with 502
when going through haproxy (this doesn't happen when going directly to
one of the apache servers). I was able to have a tcp capture while that
was happening, and I do see some strange behavior. From what I can see
haproxy sends a request to the apache server, then hangs up the tcp
connection (while response is arriving), and reopens a second tcp
connection on which it sends nothing. After 20 seconds apache closes
down the second connection and haproxy sends 502 to the client.
The haproxy log shows the following:
May 17 14:03:20 localhost haproxy[3695]: 10.214.131.22:38909
<http://10.214.131.22:38909> [17/May/2012:14:02:59.347] api-http-ipv4
radosgw-http/plana93-1 0/0/2/-1/21020 502 204 - - SH-- 0/0/0/0/1 0/0
"GET /jenkins-d15dxs9og2dxhfhlywf8i-182/testobj HTTP/1.1"
Which means, if I decipher it correctly, that the server hung the
connection after more than 20 seconds. However, looking at the server
logs I couldn't find evidence for that happening. I do see the server
responding immediately.
I was taking a tcpdump while that was happening, and I do see some
strange behavior. The haproxy log shows the following (captured on the
haproxy machine):
>
(...)
We also see the same symptoms on a different environment that runs
1.5dev7, though I didn't take any network capture so it might be a
different issue.
Before any other analysis, can you retry with better timeout values ?
From the configuration you provided :
timeout check 6
timeout connect 2
timeout http-request 3
Those 3 timeouts are in *milliseconds*, I think you wanted to define
them in seconds.
--
Cyril Bonté
