[07:15:26] another quick run to the hospital with Oscar. I should be back in ~1h [08:25:04] dcausse: the BoostQueryFunctionTest doesn't pass in jenkins [08:25:28] ejoseph: yes, the change affects other tests [08:53:50] dcausse: I would be 20 mins late for our meeting today [08:58:41] ejoseph: np! [09:29:36] Back. This took way longer than expected [10:10:19] lunch [11:38:13] Errand [13:03:10] greetings [13:12:07] o/ [13:23:19] If anyone has time to look at puppet code with me, LMK. Running into issues in deployment-prep: https://phabricator.wikimedia.org/P25970 [14:54:13] inflatador: looks like a typo. The node is deployment-elastic-09.deployment-prep.eqiad1.wikimedia.cloud but is referenced in puppet/hieradata/cloud/eqiad1/deployment-prep/common.yaml as deployment-elastic09.deployment-prep.eqiad1.wikimedia.cloud [14:54:20] elastic-09 vs elastic09 [14:55:55] ah, thank you much! Let me get that fixed [14:58:36] \o [14:59:58] @team: retro time https://meet.google.com/eki-rafx-cxi [15:01:08] mpham, ryankemper, inflatador ^ [15:01:44] Trey314159: ^^ [15:49:36] hmm, i pulled up the hacker manifesto from '86 on phrack and i'm thinking of a different document :P This is semi-related, but the one i was thinking about was about how hacking isn't just breaking into things, but making things do what they were never intended in pursuit of your goals [15:51:21] I thought maybe the classic eric s raymond article had it, but not quite: http://www.catb.org/~esr/faqs/hacker-howto.html#what_is [15:55:07] ryankemper: thats much closer to what i was thinking about than the hacker manifesto at least, it's describing the same culture that i was thinking of i suppose [15:57:40] ryankemper, mpham: looks like you have another interview later today. Could you send me a quick note after the interview, even if you don't have time to write the full scorecard? [15:57:59] gehel: sure, will dm that to ya after [15:58:00] We have 2 finalists, so I'd like to move this as soon as possible. [15:58:01] Thanks! [15:58:25] gehel: sure. if i don't fill out the scorecard right away, i'll send you a message [16:07:58] dinner time, back later [16:23:29] * ebernhardson has surprisingly nothing for puppet deploy this week [16:28:03] initial tuning reports for wbsearchentities, if the analysis is correct it suggests we should run this more often than every 3 years: https://people.wikimedia.org/~ebernhardson/wbsearchentities_202203/ [16:29:49] many of the sensitivity reports suggest hyperopt didn't really find a local minima, might let it run with more rounds and restarts over the weekend [16:44:45] anyone need puppet code deployed? I was in the Meet but it apparently closed out [17:00:19] ebernhardson: I dig the sensitivity graphs! I wonder if coordinate descent (possibly just at the end) would find slightly better params. OTOH, the sensitivity reports generally show that the ones that aren't at minima only have a little room to move. [17:00:53] Do you know if the number of characters typed went up over the 3 years, or did re-running push it down a little more than the best from before? [17:02:49] Trey314159: we have old reports still, can compare to https://phab.wmfusercontent.org/file/download/cvhhts7mfe4zcnr734nx/PHID-FILE-iytjtx5ahsna4z6fb5q3/wikidata_wbsearchentities_completion_search_analysis_-_oct18-jan19_-_item_-_de)⏎ [17:33:12] ebernhardson: I found the old reports for de, fr, and es, but not en. I eyeballed the pre- and post-tuning mean char numbers: https://phabricator.wikimedia.org/P26014 [17:34:25] Some of the new post-tuning numbers are higher than the old pre-tuning numbers. That's odd. Maybe queries have gotten more "difficult" for some value of "difficult" in the last three years. [17:34:47] Trey314159: hmm, interesting. Delta's end up aquite similar. I do wonder about how they have increased, could indicate over-fitting to the particular time period? [17:35:21] as in, the old tuning worked well at that point in time but the specific values were fit to exactly what users were doing, but then again i don't suspect user behaviour has changed much...i dunno [17:36:13] I was wondering about overfitting, but also the similarity in the deltas might mean we got an easier sample or a harder sample—which *could* be seasonal, too, I guess. [17:36:37] What we need is a dashboard! (*sad laughing noises*) [17:37:34] in theory, this could probably run monthly or some such. I suppose we assumed it would break easily so didn't try before but it required very minor changes to run today. It has long term problem of needing a tensorflow 1.x -> 2.x upgrade though [17:39:37] For now, I'm glad you've re-tuned it! And hopefully it'll collectively save our users millions of keystrokes, even though none of them will notice their own half a keystroke saved here and there. ;-) [17:40:20] lol, yea i wonder a bit if those fractional key strokes mean anything. The reporting considers saving half a keystroke or more an improvement, and less than half a 'minor improvement' [17:41:33] Half a keystroke is one keystroke every other query. That's actually something! [17:50:14] lunch, back in ~30-45 [18:34:52] ebernhardson: would you have time to join us in https://meet.google.com/eki-rafx-cxi ? Talking about the ES restart process. [18:35:02] gehel: sure [18:42:49] do we know where the 6.8 version of the hebrew analyzer is? [19:44:59] Trey314159: hmm, the packaged plugin like https://people.wikimedia.org/~ejoseph/analysis-hebrew-6.8.23.zip or the source (still looking)? [19:49:36] Trey314159: probably https://gitlab.wikimedia.org/repos/search-platform/elasticsearch-analysis-hebrew [19:50:15] ebernhardson.. the packaged plugin is what I was looking for. Thanks! [20:42:24] well, I was able to define that cert name in the horizon hiera config, but now we're back to puppet saying the cert is invalid...which is exactly where we were before ;( [21:07:42] Have a good weekend everyone! (And remember not to come to work tomorrow!) [21:10:49] buh bye [22:14:15] good nightingtons