| *** diablo_rojo_phone is now known as Guest7964 | 00:30 | |
| rcmgleite | Has anyone recently tried to remove the .pending files from the broker impl in favor of sqlite WAL? I saw that 15 years ago this was tried but reverted but I failed for find context on why. I'm going to assume that at that point we had only sqlte2 with performance issues related to WAL but unclear how useful the current external .pending implementation is. We are going to try a few things out and maybe propose to remove it | 01:50 |
|---|---|---|
| rcmgleite | again here.. just wanted to see if others tried this recently | 01:50 |
| rcmgleite | and related to the sqlite usage still - we use a bunch of timestamps (like normal unix timestamps) for things like deciding which metadata key is newer when updating metadata. That means that we use LWW approach for conflicts in updates based on wall clocks. Did I get that wrong in any sense? I'm just trying to get a clear mental model of what are the unexpected behaviors we could get based on this mechanism | 01:54 |
| opendevreview | Alistair Coles proposed openstack/swift master: Tighten up encrypter tests re. request footers https://review.opendev.org/c/openstack/swift/+/940651 | 11:31 |
| jianjian | rcmgleite, I was thinking about trying sqlite WAL or mattoliver's sharded .pending files when we worked on container performance last year, but after detailed profilings, it turned out the suspected .pending files was not the bottleneck, and we worked on optimizing several key queries and saw huge improvement. have you seen any issues with current .pending files, or just question in general? | 15:10 |
| jianjian | I think it's last-write-win, acoles knows better | 15:12 |
| opendevreview | Jianjian Huo proposed openstack/swift master: Object-server: return 409 if x-delete-at is newer than queue entry https://review.opendev.org/c/openstack/swift/+/932447 | 17:48 |
| opendevreview | Tim Burke proposed openstack/swift master: py3: Fix unicode-header-name handling https://review.opendev.org/c/openstack/swift/+/938631 | 20:08 |
| opendevreview | Jianjian Huo proposed openstack/swift master: Object-server: return 409 if x-delete-at is newer than queue entry https://review.opendev.org/c/openstack/swift/+/932447 | 21:45 |
| Callum027 | Hi there, I was wondering if anyone had any thoughts on my proposed change here? Adding new values to GET account has been done in the past, I'm expecting most well-behaved Swift clients to be compatible with it but it'd be good to hear people's thoughts, and whether more needs to be done to get this into the next major release. | 23:09 |
| Callum027 | https://review.opendev.org/c/openstack/swift/+/940601 | 23:09 |
| rcmgleite | @jianjian - was more of a general question. At this point all of our performance issues seem to be related to bad daemon default configs... auditors, updaters and replicators just run too frequently and end up overwhelming dbs or disks.. | 23:26 |
| rcmgleite | But the .pending file just feel wrong.. and I wanted to understand how much ppl saw problems with that | 23:26 |
| rcmgleite | On your analysis, anything worth sharing about where your bottlenecks where? | 23:26 |
| rcmgleite | and I'll bring up the last write wins thing tomorrow during the weekly sync to make sure I didn't misunderstand everything | 23:27 |
Generated by irclog2html.py 2.17.3 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!