[08:20:26] <dcausse>	 re: commons indices missing, hm I guess we should reroute traffic to codfw again?
[08:22:34] <dcausse>	 these indices are most probably empty
[08:44:50] <dcausse>	 manually recreated the geo field to save the switch
[08:45:20] <dcausse>	 because I just realized that we can't route only commons traffic
[09:21:16] <zpapierski>	 Ahh, I forgot that check in is in the morning 
[09:21:24] <zpapierski>	 Will miss it
[11:40:53] <dcausse>	 lunch
[12:48:08] <ejoseph>	 lunch
[13:19:16] <zpapierski>	 ejoseph: let me know once you're back (I'll probably be out for lunch soon as well, though)
[13:35:45] <zpapierski>	 dcausse: are you actively working on this atm - https://gerrit.wikimedia.org/r/c/wikidata/query/rdf/+/739916/5 ?
[13:36:46] <dcausse>	 zpapierski: I'm working on the next part after this, it's -1 by jenkins because one test depends on https://gerrit.wikimedia.org/r/c/schemas/event/secondary/+/737429
[13:37:13] <zpapierski>	 ah, so it's up for review, thx
[13:37:48] <dcausse>	 this patch should not change much unless I discover something completely wrong with this approach in the coming days
[13:39:37] <zpapierski>	 that's cool, judging on a size, it will take me some time anyway :)
[13:53:45] <zpapierski>	 lunch break
[14:38:26] <zpapierski>	 ejoseph: I'm back (at least until 4PM, meetings armageddon  ensues afterwards)
[16:02:44] <gehel>	 ryankemper, ebernhardson: triage meeting: https://meet.google.com/qho-jyqp-qos
[17:38:35] <ebernhardson>	 snapshots into swift started from codfw, also moved cirrus traffic to codfw
[17:38:50] <dcausse>	 thanks!
[17:40:11] <ebernhardson>	 archive took 1.8 minutes, slightly faster :)
[17:40:26] <ebernhardson>	 i guess thats cheating though, it's always tiny
[17:58:18] <ebernhardson>	 annoying limitation, you can't have multiple snapshots running. You can shapshot 1 index with 30 shards. But you can't snapshot 2 indexes with 4 shards each because "it's too hard on lucene freezeing all those segments"
[17:59:07] <ebernhardson>	 to be fair they built this with a different use case in mind, it supports incremental snapshots and you aren't intended to have them run all that long
[18:04:59] <dcausse>	 strange... seems pretty limiting esp for a large cluster
[18:52:48] <dcausse>	 dinner
[20:20:04] * ebernhardson waffles between minimum viable changes to fit wcqs into the notebooks, and making things more coherant...
[20:20:13] <ebernhardson>	 so many things have the name wdqs in them :P
[21:55:09] <Trey314159>	 Ugh. I just discovered that the Lucene Bengali normalizer uses a phonetic algorithm (like Soundex) to "normalize" Bengali tokens. That seems like a really aggressive way to find misspelled words. Thanks to tanny411 for her super helpful insight that made it a lot easier to figure out what's happening.