Tickets will go offline in 5 minutes from now for maintenance.
monkey[m]
lucifer: something went wrong with the user stats.
The stats I am seeing for myself (other users reported issue too, see forum link above) look like old stats, and switching between time periods doesn’t change anything.
Also hello :)
outsidecontext[m
Interesting. On my account it looks sensible.
mayhem[m]
moooin!
lucifer: ready for a chat.
lucifer[m]
mayhem: yes, debugging stats right now. will ping you sson.
monkey: does this week stats look any better now?
artists this week
monkey[m]
lucifer: negative. Definitely stats from some time ago
The top three artists for every rabge look suspicious, and don't change. The rest could be correct
Almost looks like in interspersed some of my top artists from last year into every other range
mayhem: for listen deletion, we add listened_at, recording_msid, user_id to the `listen_delete_metadata` without checking if the listen actually exists. sometimes there can also be duplicate rquests for the same listen's deletion. at the moment, we try to delete all requests and record the max_id of the listen that was attempted to delete. then after deletion is complete we removed the requests until that max_id row. and if there
were any duplicate or non-existent listen deletion requests they would go away too.
however, with the latest PR i made yesterday this changes. we now do not clear the table immediately but set a deleted column to true and for invalid/non-existent requests deleted remains false, just like for a new request. after a full dump has been produced, we removed rows from the table where deleted is true. but since we can't differentiate between invalid or new requests, the table clutters with the invalid requests. i am
thinking about how to fix it and was wondering if i should add a new table to store just the actually deleted listens and restore the original cleanup method for existing table. but while writing this i realised i can change deleted from boolean to an enum and add a invalid state to differentiate.
Jigen
wait, wtf why is that question closed and downvoted? it seems clear enough? even to a guy liek me who don't code
lucifer[m]
>Post Closed as "Not suitable for this site" - lol
mayhem[m]
wow, just closed it.
fuck me. I better go ask someplace not run by community nazis. gees.
Jigen
god it's probably becasue this is the first post by mayhem. some sites liek this are very "u must be this experienced before we take u seriously"
fuck
maybe reddit 🤔
lucifer[m]
yeah maybe scikit reddit or some community channel is a better option
mayhem[m]
they specifically tell people to ask on SO. sigh.
lucifer[m]
lolol
mayhem[m]
> but while writing this i realised i can change deleted from boolean to an enum and add a invalid state to differentiate.
that was the first thing that came to my mind too. probably the easiest.
Jigen
i'venever seen an useful stackoverflow post tbh, most of the time i look up some thing. and like the guy explains it, and thne there are 30 mesgs saying "lol u are stupid for asking this slightly wrong" and thne one, highly usefull post that actually help (and aswers my question as well) in small it also has like atleast 4 "well ackshually" refuting comments underneath
mayhem[m]
"and the shortest code necessary to reproduce the problem." that was fully ignored. I guess I need to inline that.
Jigen
i signed up with github but i can't vote or comment becasue they want me to onboard or whatever. fack
can't beassed right now
yea i cant until i.. uh get 15 upvotes. i don't care to even bother with writing a post. god useless site
monkey[m]: you around :)
monkey[m]
Jigen (IRC): Finishing lunch, will be back shortly
SigHunter has quit
Jigen
👍
SigHunter joined the channel
SigHunter has quit
SigHunter joined the channel
outsidecontext[m
mayhem: weird how your question was handled. But I'd suspect the low CPU usage is still somewhat related to Python's global interpreter lock. The assumption would be that the performance relevant parts of the nmslib calls run mostly outside the GIL, but still there might be enough Python code execution in build_index when preparing the data to have this slowed down by the GIL.
mayhem[m]
oh. GIL. feck, that might be it.
outsidecontext[m
Would be interesting to compare this when running the code with a Python 3.13 build with GIL disabled
pite joined the channel
mayhem[m]
indeed -- something worth trying. but... I think I may have found a workable solution.
be lazy.
SigHunter has quit
just build the indexes on demand, since it is very quick to build them.
I spent a bit of time on SO back in 2018 and did some questions and answers there, but haven't really come back to it because it's not been useful to me lol
but even though I just logged on for the first time since like 2022 my reputation just kept going up
Jade[m] uploaded an image: (13KiB) < https://matrix.chatbrainz.org/_matrix/media/v3/download/ellis.link/e602WU6b6lohJ5wTfnE8fJvMhD63Ez02/image.png >
because it's not about who engages in the site in a positive way, or gives useful answers, but who got there first
bitmap: LB just wobbled with a lot of "server closed the connection unexpectedly" errors from PG. I restarted the container and all seems well. could you have a look to see if anything odd is happening on the PG side of things?
bitmap[m]
mayhem: sorry that was almost certainly my fault since I'm moving some pgbouncer services in preparation for the switchover tomorrow
I'm not entirely sure why it affected LB and not MB though .. I put everything into maintenance mode and issued a PAUSE to pgbouncer first (which waits for all transactions / sessions to complete)
mayhem[m]
ah, ok. can you do me a favor and keep an eye out for more LB wobbles? If so, restart lb-web-prod on kiss, please.
ok, off to bed!
bitmap[m]
yeah I'll watch over it, good night!
2025-02-20 05133, 2025
hmm, I think the problem is that the LB web container didn't reload uwsgi when I put the old pgbouncer container into maintenance mode. this template block in /etc/consul-template-uwsgi.conf looks suspect since it has no command/exec setting:... (full message at <https://matrix.chatbrainz.org/_matrix/media/v3/...>)
lucifer: is my understanding of the consul-template config correct? if so, can it be changed to reload uwsgi when config.py changes?
that seems problematic since the workers have to be restarted if the config changes, right?
zas: hey, can you configure gmail to allow exim relay on isaac? (logging off now but I can move them tomorrow)
lucifer[m]
yes, i think so. i'll discuss with monkey on the rationale.
aerozol[m]
mayhem: 🥳 v
Aerozol[m]1 joined the channel
Aerozol[m]1 uploaded an image: (17KiB) < https://matrix.chatbrainz.org/_matrix/media/v3/download/chatbrainz.org/zcAxxQAkKnBzmftKQlQTuboT/image.png >
pite has quit
dabeglavins3 has quit
dabeglavins3 joined the channel
BrainzGit
[bookbrainz-site] 14shalini-bhandari opened pull request #1154 (03master…replace-twitter): fix: Replaced Twitter logo and link with Mastodon and Bluesky https://github.com/metabrainz/bookbrainz-site/p...
