[12:55:56] Oh very nice, I do want to see that! [12:56:07] Also, happy Monday all [13:18:16] starting the ES65->68 upgrade in eqiad shortly [13:37:25] gehel or anyone else, can you show me how to "revert the reversion" in gerrit? Re: https://gerrit.wikimedia.org/r/c/operations/puppet/+/763484/1 . The reversion is not needed [13:37:55] the reversion is not in the git log in production [13:38:08] so I think I just want to discard it? [13:38:26] brb [13:53:19] (oh yeah, he's off today) [14:26:59] ebernhardson when you get in, could you take a look at https://gerrit.wikimedia.org/r/c/operations/puppet/+/783848 ? Just need to get puppet fixed up [14:44:30] mpham are we still doing sprint planning today? [14:55:39] \o [14:57:03] inflatador: seems reasonable to me [14:59:06] mpham just messaged me that he's locked out and may not make it to planning [14:59:21] we have a pretty small group of folks in today, so thinking maybe we just cancel [14:59:47] cbogen_ gotcha. I think most of our European associates are out today due to bank holiday [14:59:58] yep. I'm gonna cancel now! [15:00:07] Cool. [15:00:20] actually I don't have access to cancel the event [15:00:23] Thanks ebernhardson ! [15:00:26] so if anyone shows up and is wondering why no one's there :) [15:11:56] aww, this wbsearchentities optimization has been working so well with minimal errors, but now running the step that actually does the optimization step it's failing while trying to read a dataset it built in previous round, some lockfile not found :( i guess i have to remember how tensroflow works :) [15:16:29] actually not so bad, turns out i had a system specific tmp dir (/run/user/$(id -u)) when we ran this previously and of course hard coded that in [15:24:46] brb [15:32:26] back [15:52:13] maybe i should reboot my desktop, it turns out pushing a laptop with all-cores to 100% gets throttled to 2.1Ghz :P [16:13:34] back [16:14:07] Disable those c-states and watch it cook ;) [17:05:47] haircut/lunch, back in ~45 [17:18:29] so, this optimizers claims that the production-deployed ranking algorithm for item_en achieved a mean characters-typed of 6.68, post optimization brings it to 6.09 [17:18:36] * ebernhardson doesn't remember how that compares to before [17:19:15] std increases from 2.00 to 2.27, so less characters typed but slightly more variance [17:22:40] i guess i gotta find the notebooks from before that graphed improvements vs query percentiles (like https://phab.wmfusercontent.org/file/download/cvhhts7mfe4zcnr734nx/PHID-FILE-iytjtx5ahsna4z6fb5q3/wikidata_wbsearchentities_completion_search_analysis_-_oct18-jan19_-_item_-_de) [17:23:25] based on that, last time de improved from 6.27 to 5.75, so same general scale of improvement (if the metric is legit) [17:24:00] * ebernhardson wonders why he chose to use dec 25-jan 7 as a test dataset last time, thats going to be a biased dataset [18:14:14] Back. Since most of us are out of office today, I'm going to postpone the ES upgrad until tomorrow [19:04:20] ebernhardson you still running a job on cloudelastic? Was thinking of trying your PR ( https://gerrit.wikimedia.org/r/c/operations/software/spicerack/+/781009 ) in that env [19:15:14] inflatador: should be all done [19:15:18] err, s/should/is/ [19:15:35] Cool, will give your PR a try shortly [20:11:59] Also created https://phabricator.wikimedia.org/T306389 to track the update [20:21:41] makes sense, i should make more tickets for things i do. It's an ongoing problem and you're not the first to notice :) [20:23:14] I have the opposite problem. I make tickets and never do them ;P [20:27:55] forgot i scheduled something w/ the doc today, heading out a few hours [20:32:54] OK, just started rolling restart of cloudelastic, will letcha know how it goes [21:25:09] Rolling restart successful using ebernhardson 's patch; more details at https://phabricator.wikimedia.org/T306389 [21:40:21] Out a little early, my son has his 2nd recital in less than wk ;0 [22:15:54] back