Hey Melanie, all, About OVH case (company I work for). We are digging into the issue. First thing, we do not limit anymore the IOPS. I dont remember when we removed this limit, but this is not new. However, the hypervisor are quite old now, and our policy on this old servers was to use some swap. And we think that the host may slow down when overcommitting on RAM (swapping on disk). Anyway, we also know that we can have better latency when upgrading QEMU. We are currently in the middle of testing a new QEMU version, I will push to upgrade your hypervisors first, so we will see if the latency on QEMU side can help the gate. Finally, we plan to change the hardware and stop doing overcommit on RAM (and swapping on disk). However, I have no ETA about that, but for sure, this will improve the IOPS. I'll keep you in touch. Cheers, -- Arnaud Morin On 18.03.20 - 22:50, melanie witt wrote:
Hey all,
We've been having a tough time lately in the gate hitting various bugs while our patches go through CI. I just wanted to mention a few of them that I've seen often in my gerrit notifications and give a brief status on fix efforts.
* http://status.openstack.org/elastic-recheck/#1813789
This one is where the nova-live-migration job fails a server evacuate test with: "Timeout waiting for [('network-vif-plugged', 'e3d3db3f-bce4-4889-b161-4b73648f79be')] for instance with vm_state error and task_state rebuild_spawning.: eventlet.timeout.Timeout: 300 seconds" in the screen-n-cpu.txt log.
lyarwood has a WIP patch here:
https://review.opendev.org/713674
and sean-k-mooney has a WIP patch here:
https://review.opendev.org/713342
* https://launchpad.net/bugs/1867380
This one is where the nova-live-migration or nova-grenade-multinode job fail due to n-cpu restarting slowly after being reconfigured for ceph. The server will fail to build and it's because the test begins before nova-compute has fully come up and we see this error: "Instance spawn was interrupted before instance_claim, setting instance to ERROR state {{(pid=3783) _error_out_instances_whose_build_was_interrupted" in the screen-n-cpu.txt log.
lyarwood has a patch approved here that we've been rechecking the heck out of that has yet to merge:
https://review.opendev.org/713035
* https://launchpad.net/bugs/1844568
This one is where a job fails with: "Body: b'{"conflictingRequest": {"code": 409, "message": "Multiple possible networks found, use a Network ID to be more specific."}}'"
gmann has a patch proposed to fix some of these here:
https://review.opendev.org/711049
There might be more test classes that need create_default_network = True.
* http://status.openstack.org/elastic-recheck/#1844929
This one is where a job fails and the following error is seen one of the logs, usually screen-n-sch.txt: "Timed out waiting for response from cell 8acfb79b-2e40-4e1c-bc3d-d404dac6db90".
The TL;DR on this one is there's no immediate clue why it's happening. This bug used to hit more occasionally on "slow" nodes like nodes from the OVH or INAP providers (and OVH restricts disk iops [1]). Now, it seems like it's hitting much more often (still mostly on OVH nodes).
I've been looking at it for about a week now and I've been using a DNM patch to add debug logging, look at dstat --disk-wait output, try mysqld my.cnf settings, etc:
https://review.opendev.org/701478
So far, what I find is that when we get into the fail state, we get no rows back from the database server when we query for nova 'services' and 'compute_nodes' records, and we fail with the "Timed out waiting for response" error.
Haven't figured out why yet, so far. The disk wait doesn't look high when this happens (or at any time during a run) so it's not seeming like it's related to disk IO. I'm continuing to look into it.
Cheers, -melanie
[1] http://lists.openstack.org/pipermail/openstack-discuss/2019-November/010505....