I just wanted to say, I’m probably more proud of this Ted talk than just about anything I’ve ever done so I’m gonna be absolutely useless for the next couple of weeks as I promote the shit out of this because I want you to watch it because it matters in the moment we are in youtu.be/dVG8W-0p6vg
The worst part to me is all that historical knowledge is sealed away. No web search will ever find that rare fix you need ever again
Who told companies that me buying one product from them means that I want multiple marketing emails from them every day for the rest of my life?
What's everyone so worried about this meteor for? Everything turned out fine after the last one 65 million years ago
Did #OpenAI just break their auth?
Judging by his past actions the new system will be 100% vision-based
I love when it goes back and forth between two wrong answers no matter how you try to correct it
DuckDuckGo was a refuge from this for a while, but ai has infected that too.
I'd definitely be up for a longer conversation. Trying to cram everything into 200 characters is rough :D
This database was already years into it's life when I came onboard and there's not a lot of will to fix the sins of the past. I was just telling my CIO the other day that it's probably worth engaging an expert for some amount of time to make sure we're doing things properly.
I really appreciate your input. I've worked in a lot of areas of IT over the last 20 years, but this is the first time I've had to manage Postgres. There are so many options to consider.
I hate that backups/restores take 8 hours. That feels like too big of an operational risk for me. My gut is saying to work on reducing that first and then worry about the upgrade. If i could get those down to 4 hours (or even lower) a dump/restore to upgrade feels like the cleanest option.
Oh that looks cool. I'll check it out.
I was thinking that, but I'm going to have to upgrade this OS in a few months anyways so I think I'm just going to bite the bullet and move to the new OS + new PG. In place OS upgrades always end up biting me down the line.
Thanks. I'll probably end up partitioning on date or an even split on the hash of a guid. The other option is dump a portion of the data (maybe the last 6 months) for the bigger tables and then fill the rest in later.
I was looking at logical replication, but there seem to be a lot of caveats around what's replicated and what isn't. The next path I was headed down was partitioning some of the biggest tables. It should help with some slower queries as well as helping the migration.
A 10 hour outage is acceptable. If I can get the dump <4 I'll definitely go that direction. It's 2 dbs in one server. One is 1TB, the other 125GB. dump/restore each take about 8 hours even with -Fd, no compress, and -j 16 on 32 cores. 2 tables account for 50% of the size, so that's my bottleneck.
That first time a Napster download was shorter than the length of the song 🤯🤯🤯
I'm loving distributed tracing in @sentry.io. Does anyone know if it's possible to show the duration of multiple transactions in a single trace? ie. A user requests a report, that triggers processing in celery and a report is emailed. Is there a way to get the total duration from click to delivery?
I upgraded all of our #python applications from 3.8 to 3.12 without incident and what's my reward? I now get to upgrade #postgres from 12 to 16. Any tips? Physical replication appears to be a no go. The life of a #cyberjanitor
All set with my Thanksgiving contribution
uv is in this weird space where it works so fast that I don't trust that it did everything right.
"It did all of that in 38ms? I must be doing something wrong"
We're going back to mainframes. Everyone just gets a dumb terminal to the web
The Bluesky algorithm is definitely giving me a better balance of the accounts I follow. On Twitter it felt like no matter how many techies I followed my feed would be awash with political posts.
My partner would say compressed air...
