[09:51:47] <dcausse>	 lunch
[13:01:01] <inflatador>	 <o/
[13:03:23] <dcausse>	 o/
[13:22:36] <ebernhardson>	 \o
[13:27:35] <inflatador>	 .o/
[13:27:38] <dcausse>	 o/
[13:28:10] <inflatador>	 I set the number of shard recoveries for codfw from 4 to 7 , hopefully that will speed up the reboot stuff. I'm keeping an eye out but if you notice P95 creeping up LMK
[13:28:16] <dcausse>	 concerning index upgrades so we have now only have .ltr and .tasks indices left
[13:28:55] <dcausse>	 should we schedule a quick switch-over to do the upgrades, not sure we can upgrade those without interruption
[13:30:17] <inflatador>	 dcausse sure, we can do it now if you think it's low risk. Otherwise happy to schedule some time next week. I'll be out for the next 10 days after next week
[13:31:18] <dcausse>	 I think the .tasks indices can be dropped, the .ltr ones I suspect the models are cached but that sounds risky
[13:31:41] <dcausse>	 inflatador: sure
[13:31:47] <ebernhardson>	 hmm, yea i think the ltr ones we have to move traffic for, wouldn't want to risk it
[13:32:09] <inflatador>	 ACK, sounds like the preference is to schedule time next week then? Otherwise I can just failover now
[13:32:59] <dcausse>	 perhaps not on a friday, I don't have any scripts ready to do it, next week is fine
[13:33:24] <inflatador>	 OK, let's schedule for 30m before Monday's standup if that works?
[13:33:38] <ebernhardson>	 +1
[13:33:49] <dcausse>	 monday is a holiday for me and IIRC in the us too?
[13:33:55] <ebernhardson>	 oh duh, yea it is
[13:34:02] <inflatador>	 Ooops
[13:34:12] <inflatador>	 So much for my bright ideas. Tuesday it is ;P
[13:34:16] <dcausse>	 :)
[13:40:36] <inflatador>	 OK, I sent the invite but feel free to suggest a different time if that doesn't work
[13:42:42] <dcausse>	 thanks!
[14:40:17] <inflatador>	 OK, I updated my crappy maintenance dashboards to use tmuxp . Feel free to try out/offer suggestions if interested: https://gitlab.wikimedia.org/repos/search-platform/es-maint-viewer
[16:14:55] <ebernhardson>	 i was curious about how effective the Soundness test was for regexes.  Pulling it back to the previous implementation it gets 25 fails in 5k iterations.  It might actually be useful :)
[16:15:54] <ebernhardson>	 ex: Expression ((((bcc AND ccd) OR ((dba OR eba) AND bac AND acd)) AND cdb) OR (bab AND aba AND bad AND adb AND dba) OR (bab AND aba AND bad AND adb AND dbc) OR (bab AND aba AND bad AND add) OR (bac AND aca) OR ((((dba OR eba) AND bac AND aca AND cad) OR ((((dba OR eba) AND bac AND aca AND caa) OR (bab AND aba AND bad AND ada AND daa) OR (bab AND aba AND baa AND aaa)) AND aad) OR (bab AND
[16:15:56] <ebernhardson>	 aba AND bad AND ada AND dad) OR (bab AND aba AND baa AND aad)) AND adb AND dbc)) not satisfied by 'ebcLbcebc񞓺bcc;bcadbccb' (regex: (([d-e]bac|.bc[c-d]?|babad*)+a*|(baca.)+)dbccb, rewritten: (([d-e]bac|[^﷐﷑]bc[c-d]?|babad*)+a*|(baca[^﷐﷑])+)dbccb, gramSize: 3)
[16:15:59] * ebernhardson should have chose a shorter one
[16:29:19] <ebernhardson>	 better example (looking for examples to put in notes for presenting this):  (bae OR (((acc AND cca) OR aca OR (adc AND dca)) AND cae) OR adc) not satisfied by 'ι꭯bcae' (regex: ..+[b-c]ae, rewritten: [^﷐﷑][^﷐﷑]+[b-c]ae, gramSize: 3)
[16:34:43] <ebernhardson>	 and it confirms (by skipping rewriting) that this is not due to rewriting, rewriting just made it more obvious
[17:11:24] <dcausse>	 nice! the dfa approach had strong limitations indeed
[17:11:35] <dcausse>	 heading out, have a nice week end
[17:51:09] <inflatador>	 .o/
[17:51:38] <inflatador>	 FYI, it does look like the choice of qlever for WDQS is public now: https://www.wikidata.org/wiki/Wikidata:Wikidata_Platform_team/Newsletter#Backend_replacement_and_architectural_design_proposals
[17:56:17] <ebernhardson>	 nice!