Nova compute servers unable to reach AMQP servers
Affects | Status | Importance | Assigned to | Milestone | |
---|---|---|---|---|---|
OpenStack Compute (nova) |
Incomplete
|
Undecided
|
Unassigned |
Bug Description
I have been running a stress test on a Nova cluster that creates VMs with m1.tiny images and destroys them randomly. Sample code is up on github (beware, the code is very rough): https:/
The Nova cluster has one Nova controller node and two Nova compute nodes. I have installed a recent version of Diablo on all three nodes from the stable/diablo branch of the nova repository (commit 1e3b88ba20b9f9a
In my test, I try to launch at most 32 or 64 machines with a sleep time of at least 600ms between launches. I turned off rate-limiting to simulate total load. I'm using carrot instead of kombu for the messaging. Occasionally, during these runs, on one of the two compute nodes, I will be presented with the following error in the log files:
2011-12-09 10:39:11,225 ERROR nova.rpc [d2910bd0-
AMQP server on 172.18.0.131:5672 is unreachable: Socket closed. Trying again in 1 seconds.
2011-12-09 10:39:11,226 ERROR nova.rpc [a88ee258-
AMQP server on 172.18.0.131:5672 is unreachable: Socket closed. Trying again in 1 seconds.
2011-12-09 10:39:11,227 ERROR nova.rpc [204c8797-
AMQP server on 172.18.0.131:5672 is unreachable: Socket closed. Trying again in 1 seconds.
2011-12-09 10:39:11,227 ERROR nova.rpc [c322da24-
AMQP server on 172.18.0.131:5672 is unreachable: Socket closed. Trying again in 1 seconds.
2011-12-09 10:39:11,227 ERROR nova.rpc [c90f3abc-
AMQP server on 172.18.0.131:5672 is unreachable: Socket closed. Trying again in 1 seconds.
I believe this is some kind of bug. If it isn't a big, it should not be an ERROR, but perhaps only a WARNING?
description: | updated |
description: | updated |
description: | updated |
Changed in nova: | |
status: | New → Incomplete |
Changed in nova: | |
status: | Incomplete → New |
Hard to say what's going on from that.
Could you run the compute node under 'strace -s8192 -f -v -o compute-trace.log' and provide the compute-trace.log somehow?