English
Jamie Turner
2.3K posts
Jamie Turner
@jamwt
Co-founder/CEO of @convex. Former Sr. Director/Principal Engineer of Storage & Databases at Dropbox. Shameless startup junkie.
Seattle, WA, USA 가입일 Nisan 2009
384 팔로잉7.2K 팔로워
@thdxr I still don’t know who is and isn’t and boomer and if i am
Will AI fix that
English
Why is it everyone with an absurdly futuristic AI take is someone who - as best I can tell - doesn’t work on (and often never has) real software that has real users and real requirements?
More so, why do you trust them?
English
No matter how bad your day is going, feel good you didn't dump $80b in an experiment that didn't pan out. Might make you feel a little better.
Polymarket@Polymarket
JUST IN: Meta announces they'll be shutting down the Metaverse, after pouring $80,000,000,000.00 into the project.
English
@jamwt Exactly what I was thinking when I saw this. Like seriously, only 64 connections?
English
It is so so embarrassing that mainstream databases still need this.
MySQL had some kinda epoll layer for awhile to park idle connections but I dunno if it ever went anywhere.
The transaction-over-network model isn’t helping this situation either.
Ben Dicken@BenjDicken
This is why you avoid direct connections to Postgres. Benchmarked PG running on a r8g.2xlarge (8 vCPU + 64GB ram) with connections ranging from 8 → 2048. Clearly a sweet spot at 64 with degrading perf thereafter. Apps often need 1000s of connections. Scale with a proxy!
English
@barre_of_lube Aka if an area the radius of a region is permanently lost, top story on the news will not be “petstore.ai lost 14 seconds of chew toy orders today…”
English
@barre_of_lube Cross AZ -> semi synchronous replication. Data loss not okay.
Cross region async. Can’t hide latency. Semi sync Introduce massive complexity to hide. Probability of disaster appropriate for bound-window data loss.
English
@AdamRackis This wouldn’t be as scary if it was good to begin with.
English
There's so many small details AI gets horribly wrong
Median software quality is gonna drop so hard next few years
English
@antho1404 @convex This data is in log streams, but admittedly that is a pro feature.
You can also find it in the dashboard if your click on any function call log entry there.
English
I love @convex, great DX, great product but man it doesn't allow for any mistakes in architecture.
One mistake on one app in dev + staging and all your usage is gone in a day. This is hard, problem solved but now I'm out of usage and might have to pay for the dev mode...
English
one thing to appreciate during hype cycles is it is very easy to spot people worth developing relationships with
there's people building in the ai space that are ruthlessly honest and care about being right in the end, not just momentarily
they stand out among the craziness
English
when a new model drops:
-> suddenly the previous model is trash
-> this is insane, we might actually be close to agi
2 weeks later:
-> it's worse than the previous model
English
@stefan_marsc @PDiTO @waynesutton @convex @theo Wanna dm me and we can talk it out? Easier on a call.
English
@jamwt @PDiTO @waynesutton @convex @theo Are these some that can be fixed or did I oversee something’s? Would honestly prefer to stay with Convex, also send sales team a form message few months ago with question regarding plan for higher limits but never any answer as well. At least AI writing back would’ve been nice 😄
English
@barre_of_lube I think the spanner model is incredibly clever but the wrong tradeoff for most apps.
English
@BenjDicken That knee in the curve at 64 connections is super common with Postgres - it's tied to how many backend processes can efficiently share CPU cores. pgbouncer in transaction pooling mode can handle 5000+ connections while maintaining only ~32 actual DB connections.
English
This is why you avoid direct connections to Postgres.
Benchmarked PG running on a r8g.2xlarge (8 vCPU + 64GB ram) with connections ranging from 8 → 2048. Clearly a sweet spot at 64 with degrading perf thereafter.
Apps often need 1000s of connections. Scale with a proxy!
English
Once again s/o to things like foundationdb and TiDB that do better here.
English
English
@jamwt @PDiTO @waynesutton @convex @theo Okay enough, still lots of love for the good parts of convex but doesn’t seem to work for us
English
English
English
@jamwt I like it but unless you increase the cost or perhaps the social cost of slop, slop architecture is inevitable precisely because its cheap and easy to build.
English
Convex's mission has recently become very clear: make excellent architecture inevitable.
It's the only way forward now.
James Cowling@jamesacowling
Whatever your take on this, writing code has always been tremendously easy compared to designing/maintaining large systems. The best engineers in the world are the best because of this stuff. Currently LLMs will not save you from bad architectural decisions. You need to constrain them to do the right thing.
English
