Notices by SolSoCoG (solsocog@ieji.de), page 13

Ok wow preview cards purging is at 301000/7780000. Looks like it will take closer to half a month for it to be done.

CPU is still at 160% usage while preview_cards are being dealt with, the fastest mastodon performance ever witnessed is delayed by a day or two because of this. In fact it might take a second or two to toot right now because of how busy the DB is.

@samuel oh god that sounds very very uncomfortable, wish you the best of luck there :schmusekadser: . Got a non python matrix server you recommend? I consider setting a private matrix up again.

@samuel but well right, version differences is something else, isn't as easy 😂

@samuel I shoved /var/lib/postgresql/16/main/ from Server A to Server B, both are 95% identical (except for the mainboard), no issues. a pgdump would have taken like 50x the time :tom_scott:

Purging around 7,5 million preview card images. I believe thats what fills up the 1tb of space. This does cause postgres to get quite sweaty hands, checking if these several years old preview cards expired. What an oversight to not remove old preview cards. It wasn't causing issues or anything, it's just pure data trash. So for the next couple of hours or days, the file-/database server cpu will be stressed out.

@Rabenalt hab ich auch nicht so aufgefasst, wollte dir fürs faven nur extra infos geben und mich doch noch entschuldigen weils länger gedauert hat als geplant :tom_scott:

@Rabenalt die 1-2 stunden wurden leider zu 5-6 stunden, mir ist irgendwie entfallen das sich 250gb preview cards angestaut hatten, sorry :crychkat3r:

ok the huge 5h backlog was finished within a couple of minutes though, now it calmed down to using around 6 cores.

I was like "huh thats odd, on the second, identical server, postgres only uses like 5% of the cpu, how?" and then this showed up. Yep the whole Ryzen 7950x strangled by postgres. By just postgres, nothing else. Nice.

@Moon :( that sucks, my condolences

Checking if minio works

Will be commencing updates on the structure 19:00 BST, sorry for EU primetime interruption but I can't do a late night shift at the moment. Will take approximately 1 to 2 hours to move the files and do a dump/import of postgres.

Expect a stupid fast performance starting thursday morning. Then 2 AMD Ryzen 9 7950X will be used to host ieji.de (well and some other stuff, just for mastodon would be over the top). Server 1 will have the sidekiq, redis and puma processes, server 2 will have files (via minio) and postgres. Is this way too much and could I improve it by just using pgbouncer or cleaning up the db for example? Yes probably, did I opt in to just throw hardware at it until Mastodon is 200% satisfied? Yes indeed.

Sorry for the short interruption, updated the system to easily migrate between two identical setups.

I am at about 80% of configuration. Won't finish this night unfortunately. I will migrate stuff from tuesday to wednesday night.

New server arrived, will prepare things, if i manage to set things up until this evening I will briefly stop the instance, move the db and images and start up things again.

Does anyone else have a hobby of buying domain names and then doing nothing with them? For like years? Just me?

Quelle: https://youtube.com/shorts/NoXPOgYt5sE

EVERY SINGLE TIME I'm like *huh haven't read into current attack schemes a lot* a humble bundle with recent books about hacking and defending against it appears, I buy it and forget about it. wtf