[09:59:10] <dcausse>	 errand+lunch
[13:05:40] <inflatador>	 greetings
[13:41:21] <mpham>	 I'd appreciate the time back from canceling retro today, but if anybody else would like to have it, we can still do it
[13:41:37] <dcausse>	 mpham: fine by me to skip
[14:55:06] <Trey314159>	 I'm okay to skip the retro, too. Does anyone have anything they feel we need to talk about?
[14:57:17] <ebernhardson>	 \o
[14:57:22] <ebernhardson>	 i guess no retro
[14:59:29] <inflatador>	 I can't make it either, sorry
[15:00:04] <dcausse>	 o/
[15:05:37] <ebernhardson>	 got relforge_wbsearchentities running in a docker env, passing tests. Seems plausible can get it going again. It's targeting elastic 5.5 though so will see how the explains go
[15:06:03] <dcausse>	 nice!
[15:06:12] <ebernhardson>	 it also emits alot of deprecation warnings, can't read the output without muting them  :P
[15:07:34] <ebernhardson>	 it also seems at some point i knew how to write complicated Makefile's. reminds me of the old quote - Debugging is twice as hard as writing the code in the first place. Therefore, if you write the code as cleverly as possible, you are, by definition, not smart enough to debug it.
[15:18:04] <dcausse>	 this somehow makes me discover this page: https://en.wikipedia.org/wiki/Write-only_language :)
[15:19:32] <dcausse>	 love the warning on the page: "This article contains APL source code.  Without proper rendering support, you may see question marks, boxes, or other symbols instead of APL symbols."
[15:19:38] <ebernhardson>	 lol, yea seems about right :)
[15:20:13] <ebernhardson>	 that APL2 is inscrutible, i wonder if anyone actually programs with symbols instead of text
[15:26:41] <dcausse>	 I guess it was a great step forward compared to punching holes in cards :)
[15:37:11] <ebernhardson>	 :)
[15:41:21] <dcausse>	 there was a lot more than I thought on translate... nothing really bad but these are little things we could have fixed earlier I think
[16:08:44] <dcausse>	 just realized that elastica 7 is "beta1"
[16:09:16] <ebernhardson>	 lol, really?
[16:09:34] <ebernhardson>	 It does seem like elastica isn't seen as the future by elastic, they built the lower level library and support that primarily
[16:10:04] <dcausse>	 yes...
[16:10:23] <dcausse>	 but I wonder about the "type" in classes like "Document"
[16:10:25] <ebernhardson>	 if i had to guess, elastica is still ruflin's side project that maybe gets some 10% time
[16:10:53] <dcausse>	 yes probably
[16:11:20] <ebernhardson>	 in 7 they still accept types everywhere, it just has to be the _doc constant.  based on what i've seen of the 6->7 transition i guess they wont kill the types in documents until the 8 transition
[16:11:21] <dcausse>	 but seems tedious to support the type removal properly at the Elastica level
[16:11:27] <ebernhardson>	 yea, probably
[16:11:36] <dcausse>	 I'm getting: [types removal] Specifying types in bulk requests is deprecated.
[16:11:41] <ebernhardson>	 oh :S
[16:11:43] <dcausse>	 with 7
[16:12:05] <dcausse>	 I guess we can continue silencing the deprecation warnings
[16:12:43] <ebernhardson>	 yea, probably have to
[16:14:51] <dcausse>	 well... one thing at a time, I should mix deprecation warnings I see when running es 7.... I'll mess up something, I've been switching between es6 and es7 constantly this afternoon
[16:16:19] <dcausse>	 *should not*
[16:17:55] <ebernhardson>	 yea, makes sense
[16:23:29] <ebernhardson>	 i wonder why this collects the search requests via cirrusDumpQuery, but then runs them directly against elasticsearch to collect the explains instead of running through prod directly
[16:23:32] <ebernhardson>	 maybe we didn't have that yet?
[16:24:57] <dcausse>	 you mean use the explain output from cirrusDumpResults?
[16:25:27] <ebernhardson>	 yea, instead of using the cirrusExplain output to get cirrus to run the explain, this requires port forwarding into an elastic instance (i'm using cloudelastic)
[16:25:54] <ebernhardson>	 i suppose stronger guarantee that the query didn't change and both sides are the same, maybe
[16:25:57] <dcausse>	 note that now analytics can directly contact prod if that helps
[16:26:30] <ebernhardson>	 ahh, maybe. I suppose i don't actually know if i can run docker containers in analytics.  This needs old python, tensorflow 1 doesn't install against py 3.8+
[16:26:43] <dcausse>	 oh I see
[16:26:57] <ebernhardson>	 maybe others, i went with 3.5 because i had an old venv that said this used to work with 3.5
[16:27:42] <ebernhardson>	 more proof i should have done this with perhaps numpy math directly instead of something that changes so quickly like tf
[16:32:15] <dcausse>	 I thought there was a need to have a kind of "fit" function to adjust all the params
[16:32:24] <dcausse>	 but numpy might have that
[16:34:08] <ebernhardson>	 in these we really aren't doing any kind of gradient optimization, actually hyperopt is doing all the tuning :)
[16:34:23] <ebernhardson>	 we feed the parameter space into hyperopt, and only use tensorflow to evaluate the equation
[16:34:29] <dcausse>	 ah ok I see
[16:35:21] <ebernhardson>	 i had asked that professor from canada (forgetting name) about actual gradient decent against the lucene equation but (in the brief moment over beers) he didn't have any reasonable ideas of how that would be helpful over brute force
[16:45:22] <inflatador>	 dcausse ebernhardson sorry for being late, if y'all need anything for puppet deploy I'm her
[16:45:24] <inflatador>	 or here
[16:46:13] <ebernhardson>	 inflatador: it's not puppet, but lets ship the whitelist.txt for wdqs? lemme find patch
[16:46:25] <dcausse>	 inflatador: nothing on my side, you merged everything I had yesterday :)
[16:46:27] <ebernhardson>	 inflatador: https://gerrit.wikimedia.org/r/c/wikidata/query/deploy/+/779069
[16:46:54] <inflatador>	 cool, I was looking at this one yesterday, 1 sec
[16:47:46] <inflatador>	 OK, merged. I assume we'll need to do a wdqs deploy to pick up the changes?
[16:48:58] <ebernhardson>	 inflatador: doh, actually that patch was in wrong repo
[16:49:20] <ebernhardson>	 inflatador: i was thinking we need to build, then deploy the build, but realized this is already in the deploy repo. Will take one sec to make new patch :)
[16:49:21] <inflatador>	 ebernhardson should I revert?
[16:49:30] <ebernhardson>	 inflatador: nah, it will be overwritten in a sec
[16:49:56] <ebernhardson>	 well, maybe. Maybe this is supposed to only be in deploy? I gotta look
[16:50:50] <ebernhardson>	 inflatador: ok i'm just totally confused. The first patch is correct :) While many of the root files in that deploy repo are sourced from the rdf repo, this file is only found in deploy so i guess this is the canonical location. 
[16:51:56] <ebernhardson>	 inflatador: sorry for making everything confusing. So we should only require a deploy of the repo, and maybe a restart of the instances (unclear, maybe david knows when whitelist.txt is loaded)
[16:53:23] <inflatador>	 np, not sure what you mean by deploy repo? this is what I typically do to deploy WDQS https://phabricator.wikimedia.org/P18246
[16:53:35] <inflatador>	 I'm also back in the chat room, was just getting coffee
[16:53:57] <ebernhardson>	 inflatador: there is wikidata/query/deploy, most of the contents of that comes from wikidata/query/rdf repo either copied from dist/src/script or emit from the java build
[16:54:04] <ebernhardson>	 sec, joining
[18:08:03] <inflatador>	 yeah, we need to add some guard rails
[18:11:10] <inflatador>	 lunch, back in ~30
[18:28:16] <ebernhardson>	 hmm, close but i'm suspicious of this test case. So i added assert True is False as first line of test, told pytest to only run that test. And it passes :S
[18:35:15] <Trey314159>	 It was a bright cold day in April, and the clocks were striking thirteen.... True *is* False!
[18:39:04] <inflatador>	 BTW, "The Punch Escrow" is on its way to my local library! 
[18:40:09] <ebernhardson>	 turns out the PEBKAC is strong on this one :) I copied another test as a skeleton and never renamed it, and in python whichever function is defined last wins 
[18:40:10] <ebernhardson>	 :)
[18:53:33] <Trey314159>	 inflatador: Cool! Now I won't feel too bad if you don't love it. As a book hoarder, I sometimes forget libraries are a thing—even though I sort of run one out of my basement!
[18:55:07] <inflatador>	 The San Antonio library is pretty good. Mainly I borrow digital audiobooks, but we go up there sometimes. Even renting (gasp) DVDs now and again
[18:57:14] <Trey314159>	 BTW, I'll pass on a vaguely similar recommendation from gehel: https://en.wikipedia.org/wiki/The_Quantum_Thief —the whole trilogy was good, with philosophically interesting premises (and a fun sprinkle of multilinguial etymology throughout—though you don't have to notice or care to enjoy the excellent story).
[20:57:07] <ebernhardson>	 lunch
[21:42:52] <ebernhardson>	 back
[21:45:25] <inflatador>	 see you tomorrow