| clayg | I'll be about 1hr-ish late to the PTG; need to drop kids off at school | 12:50 |
|---|---|---|
| timburke__ | good morning | 12:56 |
| *** timburke_ is now known as timburke | 12:58 | |
| kota | good morning timburke and clayg | 13:01 |
| mattoliver | o/, be there in a sec | 13:03 |
| timburke | https://review.opendev.org/c/openstack/swift/+/735359 | 13:13 |
| timburke | https://review.opendev.org/c/openstack/swift/+/736802 <- memcache: Config option to add a chance of skipping memcache | 13:18 |
| opendevreview | Pete Zaitcev proposed openstack/swift master: Change github.com to opendev.org so git-review works https://review.opendev.org/c/openstack/swift/+/814778 | 13:27 |
| timburke | 758605: Only count asyncs once per IP | https://review.opendev.org/c/openstack/swift/+/758605 | 13:31 |
| timburke | partial bug: https://bugs.launchpad.net/swift/+bug/1612168 | 13:34 |
| timburke | https://github.com/facebook/mcrouter/ | 13:39 |
| timburke | does anybody actually use the fact that ring.gz files are proper gzip streams? jokke_ had an idea to gzip each blob for v2 rings, and have the length be the compressed length (or maybe we write down both compressed and decompressed lengths, since the decompresed length may be useful on its own) | 15:25 |
| zaitcev | I don't know of anyone doing it. | 15:49 |
| zaitcev | The main reason is, suppose you uncompress it and get a .ring. Now what? Write a tool that examines it and does some consistency checking or statistics? I don't think anyone is into details of Swift enough. | 15:50 |
| zaitcev | clayg, timburke_, acoles: Is it useful to run replicator on EC policies by using its override option? Seems like that would split replication from reconstruction in a way. | 17:02 |
| timburke_ | i'm pretty sure our SREs sometimes run a separate replicator or reconstructor and specify just the one partition -- not sure about whether they'd ever run a separate handoffs_only reconstructor while the normal one is still trying to handle both primary syncing and handoff reverting... though maybe that could become more appealing if there were a primary_only mode | 17:48 |
| timburke_ | thinking some more about https://bugs.launchpad.net/python-swiftclient/+bug/1899495 -- it looks like we (attempt to) delete the manifest even if there were issues deleting the segments... so any retry won't be as simple as issuing the same request again :-/ i guess it's on the client to go back and retry each segment delete individually? | 17:57 |
| timburke_ | sticking a middleware after slo to give each delete a 50% chance of failing, i see swiftclient behaving like https://paste.opendev.org/show/810118/ -- which seems awful close to the reported bug | 18:01 |
| opendevreview | Tim Burke proposed openstack/python-swiftclient master: Detect and retry SLO delete partial failures https://review.opendev.org/c/openstack/python-swiftclient/+/814820 | 18:53 |
| timburke_ | jokke_, needs tests, but i think ^^^ might do what you need | 18:54 |
| opendevreview | Tim Burke proposed openstack/swift master: Fix some imports for py310 https://review.opendev.org/c/openstack/swift/+/814823 | 19:30 |
| jokke_ | timburke_: cool | 21:16 |
Generated by irclog2html.py 2.17.2 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!