i see there's a difference in buffer contents as well, prod had to read entire recording table from disk where it was in cache on wolf.
alastairp
I think they might both be hard drives
zas: md3 on aretha is rebuilding, not sure if that's expected or known
lucifer
you mean both aretha disks are HDD or both wolf and aretha are HDDs?
alastairp
lucifer: both servers have hdds
lucifer
ah i see.
i'll dump the settings from both instances and do a compare to see if i can figure out some difference.
mayhem
aaargh.
the signup process for the airbnb has become decisively spanish. type on lots of data, in triplicate and sign it.
zas
alastairp: that's not a rebuild, but a check. Those are done on regular basis. That's still a resync.
atj_mb
Will there be an iris scan on arrival?
lucifer
alastairp: its the cache difference. mapping.canonical_release_redirect needs to read entirely from disk and takes about 15mins to read. rerunning the query after cache is built take 43s.
alastairp
zas: right thanks, I didn't look into it in too much detail, only the .....>
lucifer: right, makes sense
atj_mb: only for people who left the EU
mayhem
reosarevok: is you want to eat at this restaurant at 10pm, you'll wait until 1am before you get in.
atj_mb
alastairp: as ever, thanks for the reminder
lucifer
alastairp: i guess the 3.5hours runtime makes sense then. not sure anything can be done about this.
[critiquebrainz] 14dependabot[bot] opened pull request #470 (03master…dependabot/npm_and_yarn/glob-parent-5.1.2): build(deps): bump glob-parent from 3.1.0 to 5.1.2 https://github.com/metabrainz/critiquebrainz/pu...
[critiquebrainz] 14dependabot[bot] opened pull request #471 (03master…dependabot/npm_and_yarn/json-schema-and-jsprim-0.4.0): build(deps): bump json-schema and jsprim https://github.com/metabrainz/critiquebrainz/pu...
lucifer
also +1 on adding run_sql_script and the non transaction version to BU.
BrainzGit
[critiquebrainz] 14dependabot[bot] opened pull request #472 (03master…dependabot/npm_and_yarn/ajv-6.12.6): build(deps): bump ajv from 6.10.0 to 6.12.6 https://github.com/metabrainz/critiquebrainz/pu...
alastairp
lucifer: hi
lucifer
hey!
alastairp
consul config for LB has ` "websockets_server_url": "https://sockets.listenbrainz.org",`
but this seems obviously wrong
lucifer
yes that config url is unused.
alastairp
I'm just talking to monkey now about it
we see that in js we use window.location.path and append socket.io to the end
alastairp: apparently the cleanup process is taking ~24 seconds for 105k rows 🗿
And was without factoring in the r/w time
I suspect that's because the functions I am using (pandas.series.map / pandas.series.apply) are just glorified for loops. I am currently trying out ways to vectorize the operations somehow
lucifer
alastairp: i don't think we can do ws in same containers in local development. flask dev server serves only 1 request at a time iirc so incompatible with having a ws server running. unless we want to replace to flask dev server with something else.
(regarding the last point in that ticket)
alastairp
lucifer: new flask actually uses threads I think
lucifer
oh, nice.
alastairp
you're right that we'd need to test is
doing this would mean that our ws server was on the same host as the flask server, allowing us to continue using window.location
lucifer
yes makes sense. also 1 less container is better
alastairp
yep, right
but there is the magic in the ws container with the extra thread for rmq connections
which makes me a bit worried about how we do that
lucifer
yeah, i wonder if we should add celery and move the magic to it.
there are probably better resources available on how celery interacts with flask/threads etc.
alastairp
I think I had another thought about how we could do it
OK, for now let's delete all of the websocket data in the config file (I'll do that now, to assist zas with some cleanup he's doing). This isn't going to make anything _more_ broken than it is now
and if we need to add something back in when we try and fix local dev, we can do that
I would put money on more than 50% being duplicates and other other unlistenable shit.
lucifer
MB has 27 M recordings and 40 M tracks. it would be odd if 60M tracks were missing from MB but in apple music. so a lot more than 50% in dupes i think.
mayhem
agreed.
40M tracks. wow.
seems it was just yesterday we celebrated 1M tracks.