[lvs-users] Problems after upgrade to 2.6.33

To: lvs-users@xxxxxxxxxxxxxxxxxxxxxx
Subject: [lvs-users] Problems after upgrade to 2.6.33
From: Jan Kasprzak <kas@xxxxxxxxxx>
Date: Fri, 12 Mar 2010 22:14:54 +0100

I have two IPVS hosts (with heartbeat), managing about 20 virtual services
with ~150 workers. I use ldirectord, direct routing, wrr and wlc schedulers.
After upgrading one of the IPVS hosts to 2.6.33 (from 2.6.30),
I am experiencing problems - the workers are being removed and 
re-added by ldirectord frequently, and sometimes even the whole
virtual service stalls and is redirected to the fallback server.

        I am still not sure whether the kernel is the cause, but there is
a strange thing: "ipvsadm -L" reports 0 active connections for every
service and worker, the number of inactive connections reported is also
almost always 0, very rarely (about 1 worker out of 10-20) is reported
to have 1 inactive connection. This does not match the reality, and
indeed, "wc -l /proc/net/ip_vs_conn" reports a more reasonable number
of about 50-60 thousand connections total.

        Maybe this can confuse the wlc (and possibly also wrr) scheduler
enought to direct all connections to a single worker until it becomes
overloaded? Is it a kernel bug IPVS? Has anybody seen something similar?



| Jan "Yenya" Kasprzak  <kas at { - work | - private}> |
| GPG: ID 1024/D3498839      Fingerprint 0D99A7FB206605D7 8B35FCDE05B18A5E |
|    Journal: |
Please don't top post and in particular don't attach entire digests to your
mail or we'll all soon be using bittorrent to read the list.     --Alan Cox

Please read the documentation before posting - it's available at: mailing list - lvs-users@xxxxxxxxxxxxxxxxxxxxxx
Send requests to lvs-users-request@xxxxxxxxxxxxxxxxxxxxxx
or go to

<Prev in Thread] Current Thread [Next in Thread>