| mnederlof | hi, does anyone know why only the primary container nodes are queried for the async_pending objects? I looked into the code and noticed that in the past a 404 was considered a success (but that caused issues if all containers are placed on handoff nodes). So why are we not querying the handoff nodes, if not all primaries are a success? | 08:50 |
|---|---|---|
| opendevreview | Shreeya Deshpande proposed openstack/swift master: Split statsd client from logger https://review.opendev.org/c/openstack/swift/+/915483 | 13:49 |
| opendevreview | Yan Xiao proposed openstack/swift master: proxy-logging: Add real-time transfer bytes counters https://review.opendev.org/c/openstack/swift/+/930918 | 13:59 |
| opendevreview | Alistair Coles proposed openstack/swift master: sq? expirer: don't try to delete non-existent task containers https://review.opendev.org/c/openstack/swift/+/930925 | 15:21 |
| timburke | mnederlof, the primary locations are where we expect the container DBs to be, and the async pending means we're *already* putting up with a widened consistency window -- so may as well wait for the container-replicator to get the DBs where they ought to be | 17:54 |
| timburke | asyncs already add to container-server load -- if we can do the update inline, the assigned object-server (typically, depending on replica counts for obj and cont layers) only needs to update one container-server. but if any of them need to write an aysnc, it'll want to contact all primaries to resolve it | 17:56 |
| timburke | (hopefully, most of those updates should be pretty quick: container-server sees that yep, i already know about that, and doesn't have to do any writing) | 17:57 |
| timburke | but it still ties up connections. going out to handoffs doesn't seem likely to improve the situation much | 17:58 |
| timburke | that being said, i could maybe see us growing a couple more knobs: *maybe* we could have an option to look out in handoffs on an explicit 404 (DB really, truly does not exist) from a primary, but i wonder if the better value would be something akin to handoff_delete for the replicators -- if an async gets to 2/3 primaries, say, go ahead and clean it up (with the assumption that container replication will eventually get the | 18:02 |
| timburke | last guy on the same page) | 18:02 |
| timburke | i'd forgotten all about that pre-bug #1328735 behavior! | 18:06 |
| patch-bot | https://bugs.launchpad.net/swift/+bug/1328735 - Object-updater gives up updating container with no success if all containers are placed at handoff (Fix Released) | 18:06 |
| opendevreview | Clay Gerrard proposed openstack/swift master: wip: rotate cycle and skip after awhile https://review.opendev.org/c/openstack/swift/+/930954 | 22:18 |
Generated by irclog2html.py 2.17.3 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!