[08:46:10] <gehel>	 ebernhardson, ryankemper: I moved our 1:1 for tomorrow (I need to go to a SexEd presentation for Oscar this evening)
[08:46:22] <gehel>	 ebernhardson: let's try to do our ITC tomorrow if that's OK for you
[11:29:08] <dcausse>	 lunch
[14:14:01] <inflatador>	 <o/
[14:20:18] <dcausse>	 o/
[14:20:24] <gehel>	 o/
[14:20:41] <inflatador>	 looks like wdqs1009 corrupted itself, wdqs1010 is still going
[14:28:04] <dcausse>	 yes, I think you can stop the import on wdqs1009 :(
[14:28:16] <inflatador>	 I just restarted it
[14:29:26] * inflatador crosses fingers
[15:51:59] <ebernhardson>	 \o
[15:52:57] <ebernhardson>	 perhaps interesting, a bunch of yandex source repos leaked and someone put up a list of ~1900 ranking factors they use: https://yandex-explorer.herokuapp.com/search?q=&o=all
[16:02:18] <gehel>	 ryankemper: triage: https://meet.google.com/eki-rafx-cxi
[17:07:51] <inflatador>	 dcausse FYI , this updates the JRE. Don't know if this explains the increased memory requirements you were seeing in yarn but FYI https://gerrit.wikimedia.org/r/c/operations/docker-images/production-images/+/884351
[17:16:22] <dcausse>	 inflatador: thanks for the heads up! it's unrelated because the mem increase I saw was in yarn (it does not use docker images) and it runs java8
[17:17:27] <inflatador>	 oh yeah, I know it doesn't use docker but didn't know it was using java8
[18:39:17] <inflatador>	 lunch, back in ~1h
[19:55:08] <inflatador>	 back
[20:39:48] * ebernhardson realizes i have no idea what git commit the version of mjolnir used in yarn is
[20:40:18] <ebernhardson>	 guessing it's the commit 'Increase evrsion to 1.1', but it's not actually recorded in the wheel anywhere
[21:28:44] <ebernhardson>	 meh...i changed the conda environment to remove specific versioning so it could choose on its own. It's been "solving environment" for 30 minutes now :S
[22:12:38] <ebernhardson>	 finally got some things running...it doesn't start spark yet for some reason, but at least it's installing deps and doing something now :)
[22:15:57] <ebernhardson>	 some sort of scala mismatch..should be fun: java.lang.NoSuchMethodError: 'scala.collection.mutable.ArrayOps scala.Predef$.booleanArrayOps(boolean[])'
[23:11:58] <ebernhardson>	 interesting, i hadn't previously looked but pulled some numbers.  taking the avg across instances of the p95 per-node full_text latency, it dropped from 180-195ms to 160-170ms when the incoming links jobs were turned of. Similarly moved more_like from 220-235ms down to 195-205ms
[23:12:23] <ebernhardson>	 active thread usage in the search pool was reduced by 200 threads