| zaitcev | Guys | 01:00 |
|---|---|---|
| zaitcev | I'm looking at Clay's patch https://review.opendev.org/c/openstack/swift/+/103779 | 01:01 |
| zaitcev | Actually, let's poke clayg. | 01:01 |
| zaitcev | So the question I'm having is this: under what circumstances the performance of the reconciler matters? | 01:02 |
| zaitcev | Do we know of multi-region clusters that operate for a long time under a network partition? Wouldn't this kind of situation create way worse issues than the scenario where two containers with the same name have different policies? | 01:04 |
| timburke_ | zaitcev, the trouble we've been having isn't so much that the reconciler doesn't keep up (we haven't had much in the way of complaints) | 01:04 |
| timburke_ | but rather that we have too many workers fighting over the same work | 01:04 |
| zaitcev | Okay, thanks | 01:05 |
| timburke_ | with hundreds of nodes running the reconciler, we'll regularly have multiple nodes read the same item off the queue, move the same data with the same timestamps, and write down the same tombstones with the same timestamps | 01:05 |
| timburke_ | which leads to tracebacks like http://paste.openstack.org/show/806718/ during a part power increase, or problems like https://bugs.launchpad.net/swift/+bug/1910804 | 01:07 |
| zaitcev | Interesting scale. | 01:13 |
| opendevreview | Tim Burke proposed openstack/swift master: Add support for multiple container-reconciler https://review.opendev.org/c/openstack/swift/+/103779 | 01:16 |
| opendevreview | Merged openstack/swift master: Run in-process func tests under py38 https://review.opendev.org/c/openstack/swift/+/795425 | 01:39 |
| opendevreview | Matthew Oliver proposed openstack/swift master: Add object-expirer new mode to execute tasks from general task queue https://review.opendev.org/c/openstack/swift/+/517389 | 04:18 |
| opendevreview | Tim Burke proposed openstack/swift master: Include status code when we fail to parse an SLO-delete response https://review.opendev.org/c/openstack/swift/+/796734 | 22:26 |
| timburke | this seems really weird to me: https://github.com/openstack/swift/blob/master/swift/proxy/controllers/obj.py#L1399-L1400 | 22:43 |
| timburke | and i think it prevents us from supporting more than 32 fragmensts total :-/ | 22:44 |
| opendevreview | Tim Burke proposed openstack/swift master: Quiet EPIPE tracebacks https://review.opendev.org/c/openstack/swift/+/796960 | 23:41 |
Generated by irclog2html.py 2.17.2 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!