[09:42:36] I read a ticket where AW was influenced negatively by a parser cache that was local to a DC. [09:42:36] I found out that WMF seems to have decided to go multiDC with active-active. [09:42:38] I'm interested in knowing who made that decision and what the motivation was? [09:42:39] To me it seems like a ton of complexity at little gain over active-passive. [09:46:06] I'm working in a government agency that prepares for war and we don't even have active-active DCs so I have a hard time seeing why WMF would need it. [09:46:06] Also both datacenters are in the US which seems like a bad idea given the recent changes to the political climate (oligarchy, authoritarian measures, weak/failing democracy with illegal lottery influenced votes, etc) [09:55:08] I don’t know whether this will answer your question [09:55:09] https://techblog.wikimedia.org/2023/05/08/around-the-world-how-wikipedia-became-a-multi-datacenter-deployment/ (re @Npriskorn: I'm working in a government agency that prepares for war and we don't even have active-active DCs so I have a hard time seeing w...) [10:02:18] Grokipedia vs Wikipedia: Quantitative Analysis (video) ← Lewoniewski [10:02:18] https://en.lewoniewski.info/2025/grokipedia-vs-wikipedia-quantitative-analysis-video/ [10:04:28] this is not really new but maybe relevant to AW. (re @Npriskorn: Grokipedia vs Wikipedia: Quantitative Analysis (video) ← Lewoniewski [10:04:29] https://en.lewoniewski.info/2025/grokipedia-vs-wikipedia-qu...) [10:11:44] I'm curious, which ticket? (re @Npriskorn: I read a ticket where AW was influenced negatively by a parser cache that was local to a DC. [10:11:44] I found out that WMF seems to have...) [10:19:22] this one (re @davidsantamariame: We are working on that here: [10:19:23] https://phabricator.wikimedia.org/T405461 [10:19:24] Hopefully we will fix it soon) [10:27:13] Having caches spread out is not a problem. [10:27:14] What I'm reacting to is this: [10:27:15] Operating multiple application data centers also reduces organizational risk from catastrophic damage or connectivity loss to a single data center. To achieve this redundancy, each application data center must contain all hardware, databases, and services required to handle the full worldwide volume of our backend traffic. [10:27:17] AFAIK both of the current application datacenters are in the US and currently the orchestrator is only deployed in one of them which led to the issue. [10:27:19] So either Wikifunctions had to be redeployed to 2 datacenters or redesigned so they ralk together (complex). [10:27:20] It's way simpler to switch to active-passive and only have 1 application DC up at a given point in time. [10:27:21] WMware support hotswitching datacenters which is very cool and means no downtime during the hand-off. (re @Al: I don’t know whether this will answer your question [10:27:23] https://techblog.wikimedia.org/2023/05/08/around-the-world-how-wikipedia-be...) [10:35:13] in any case the more or less zero downtime of Wikipedia as long as I remember (joined in 2008) is very impressive. So kudos to the SRE team! [10:35:14] In most of the time since 2008 there was only a single primary DC. [10:35:15] So I would argue if WMF didn't need it 2008-2018 it's not worth the trouble now either. [11:23:50] As I read it, the set-up is active-active only for reads, which would make it “geo-distributed primary-replica”. Of course, the geographical distribution is currently limited, but that makes managing latency issues more tractable. I can’t see WMF crossing an ocean for replication any time soon. (re @Npriskorn: Having caches spread out is not a problem. [11:23:50] What I'm reacting to is this: [11:23:51] Operating multiple application data centers also reduce...) [16:22:06] Hallo [16:22:30] People sometimes reuse implementation pages and change the programming the language, or change from code to composition. [16:23:15] Is it a good practice? In Wikidata, completely repurposing an item is frowned upon, and the recommendation is to create a new item and to delete the old one (or just leave it alone). [16:23:28] Example: https://www.wikifunctions.org/view/en/Z10889 [16:23:37] It's composition now, but it was Python. [16:23:56] oh, this looks bad [16:23:56] I've never seen it but I guess we should revert and copy the code in a new implementation... (re @amire80: People sometimes reuse implementation pages and change the programming the language, or change from code to composition.) [16:24:57] that said, in this case, the same person did the creation and the change, so maybe there is a reason behind it (like the original code was not working), you should ask them first [16:26:11] for me, the same logic applies more or less on Wikifunctions [16:26:12] (a bit more flexible tho, if the code does the same think it's less bad than changing a Wikidata item prupose) (re @amire80: Is it a good practice? In Wikidata, completely repurposing an item is frowned upon, and the recommendation is to create a new it...) [16:30:20] Yes, it’s not a common practice. We would generally expect an additional implementation, but I don’t think that’s documented anywhere. [16:33:22] I don't contribute enough to write such a thing myself, but perhaps someone should document what's the best practice here somewhere. [16:34:24] And maybe even make an abuse filter that prevents it. Or prevent it in the WikiLambda extension code. Or not. I don't have a strong opinion. Just thinking out loud. [16:36:36] we should talk about it first, but yes ;) (re @amire80: I don't contribute enough to write such a thing myself, but perhaps someone should document what's the best practice here somewh...) [16:36:50] that would be overkill (at least for now) (re @amire80: And maybe even make an abuse filter that prevents it. Or prevent it in the WikiLambda extension code. Or not. I don't have a str...) [16:37:59] We have [[Wikifunctions talk:Best practices]] for such thoughts, but the activity is very low. (re @amire80: I don't contribute enough to write such a thing myself, but perhaps someone should document what's the best practice here somewh...)