[12:03:46] Krinkle: if you're around, this is tiny bit important: https://gerrit.wikimedia.org/r/1173364 When I looked at logs, it turned out it was doing wayyyy too much deletes, it turned out the deletion part repeats on every table so it gets repeated on 256 tables in parser cache. The patch should fix it without breaking other cases. Would you mind taking a look or if you don't have time right now, I'll revert the config patch [12:08:55] if anyone else is feeling like reviewing that patch too, I'd be grateful [12:09:59] Looking [12:12:02] Amir1: this doesn't change maint script behaviour? [12:13:28] hmm, it does [12:14:02] OTOH, I want to fully remove that script [12:14:59] Krinkle: I can add a check to ignore this condition if limit is INF [12:15:07] for now, until we get rid of the script [12:15:48] It's used by third parties presumably so not until we remove the option to turn it off. Which... sounds good to me long term if we prove this viable. The order is shuffled so should be fine [12:16:05] But the purge period ratio does affect this [12:16:23] Might need to do more per turn given it is less likely to reach stuff now [12:16:28] We'll see. [12:16:37] I'd love for it to be one number instead of two [12:16:46] Sounds good to me [12:16:52] RE: inf [12:17:44] you mean I should change the condition to instead of dividing by ten, divide by the ratio config? I like the idea [12:17:51] (on top of the INF) [12:22:17] Maybe later yes. It seems to me controlling them separately is giving sysadmin more then they need as there is basically only one right way, if you don't want to build up a backlog or take on more load than needed. [12:22:51] yeah [12:22:53] Finding the right balance is tricky. You want to account for some spikes where a single request or cli script might write more than one row. [12:23:08] But it doesn't seem like the kind of thing where every site needs to do that itself. [12:24:15] I don't know. Maybe as a first step we can make it null and do it automatically while we build confidence. [12:24:39] For example parsoid doubles writes [12:26:38] There's a few things that can increase it by config choice so one might not fit all. But if we can find a reasonable margin that isn't too much on small sites and clears everything quick enough, that could work. Basically it comes down to what we assume is the number of rows added on average per cache-writing process. [12:27:12] added/replaced [12:27:45] It might actually be less than 1, because of updates/replaces. [12:29:07] I'll monitor and see how it goes [12:29:55] Landing [12:30:15] thanks! [12:46:10] Amir1: wikidiff2 Phan warning fixed in master in the last few days [12:46:24] Prolly needs backporting [12:46:32] thanks [12:47:36] https://gerrit.wikimedia.org/r/c/mediawiki/core/+/1172658