odd. I know that they have the same ip pool so i was wondering because for about a minute i had conn
odd. I know that they have the same ip pool so i was wondering because for about a minute i had connection timeouts
BSeeing an interesting problem with workers, may or may not be a result of hyperdrive. I have an API implemented in workers with typescript and I'm running React in Pages. Some of my components load 2 different requests simultaneously. Sometimes, works fine, a lot of times, one of the requests receives a connection timeout from the DB. If I disable hyperdrive, the problem goes away. If I stagger the request by a second or two then there is no problem. Is it possible hyperdrive or workers cannot locate a connection in such quick succession? Should I serialize the requests?
Interestingly, I also see the problem locally with wrangler dev. Again, stagger the requests, no issue.
Interestingly, I also see the problem locally with wrangler dev. Again, stagger the requests, no issue.
MIf you’re seeing it in dev - where Hyperdrive’s connection logic isn’t in place - it might be something else. Are you doing something odd with your driver/client code?
BNothing unique and no problems normally. The only issue is if I send off 2 simultaneous connections from react at roughly the same time. I see it in wrangler and I see it in workers production. Using Hono and Kysley. I'll dig a bit more, as of right now, just staggering the requests.
PI think its possible I'm experiencing this as well
Pup here
Pstarted getting this error a couple days ago
Pthat looks like a bug
Pand i suppose it is
Pbut whats interesting is
PPCONNECT_TIMEOUT?
P its a bug in postgres.js, caused by CONNECTION_TIMEOUTs
PFurthermore, I can corroborate that the issue is solved when hyperdrive is disabled
PI can tell that hyperdrive was properly disabled because requests now take about 200ms longer
AHow interesting. Can you share a code snippet that induces this behavior? How often/reliably does it reproduce for you?
AAside, you should be seeing better caching behavior now, btw, @Ping for toast . How's it looking on your side?
PBy a code snippet, do you mean you're asking for a minimum reproducible example? My configuration is very much like bs_sfalls
- I use solid-start and hyperdrive, and my backend is elysia with hyperdrive.
- The backend is hosted on a separate worker.
- In SSR, two requests are fired simultaneously — one to retrieve data about the current user, and one to fetch the data presented on the page, using service bindings
- This can obviously be triggered by doing a page reload.
- Until said page reload happens, requests work fine
- After n seconds (where n is unknown; greater than 30s and less than 5 minutes?), presumably after hyperdrive disconnects // the worker cools down, it begins working again
PCaching is a lot better — 50% or so
PI also literally just switched db hosts as I thought this would be a good opportunity, so I will test it again to see if it reproduces
AInteresting. We do have some rate limiting in place but it wouldn't be triggered by a single pair of simultaneous requests, and even if it was it would only lock out for a couple hundred ms.
I'm going to have to dig at this a bit. I'll put together a ticket for it
You said this only started within the last week or so?
@bs_sfalls same timeframe for you, just the last week or so?
I'm going to have to dig at this a bit. I'll put together a ticket for it
You said this only started within the last week or so?
@bs_sfalls same timeframe for you, just the last week or so?
Pyup, you can actually see the first time i noticed it
Pright here :)
Pit could have been up to a week tho
PReproduced
PThe reason that it looks like the html has loaded is because of streaming
IYes followed the instructions, using the exact same code as the examples in the documentation. Getting the error 'PostgresError: server requires encryption' when postgres in the Cloudflare worker attempts to set up a connection.
PAny updates on this? If not, would you mind pinging me when there's updates 
, it would be very appreciated
, it would be very appreciatedAWill ping you when we have anything to share on this one. You'd better have toast.
Palways 
Bhi gusy does worker support node-postgres's pg.Pool ? when i building with svelte Cloudflare adapter, i got fs,dns,...etc not supported error
IMake sure you have nodejs_compat enabled in your wrangler.toml.
Icompatibility_flags = ["nodejs_compat"]Morning all! Yep, that scenario is why we sunk some time into getting postgres-js supported, and now recommend it.
Boh ok, maybe it is good idea to put a warning on this page https://developers.cloudflare.com/workers/tutorials/postgres/
Cloudflare Docs
In this tutorial, you will learn how to create a Cloudflare Workers application and connect it to a PostgreSQL database using TCP Sockets. The Workers application you create in this tutorial will interact with a product database inside of PostgreSQL.
AI'll take a look through and add that. Appreciate the suggestion
Thas anyone used hyperdrive with neondb? What is the latency that you guys are getting?
INeon databases are hosted in a single region so you need to use smart placement in the same region to get acceptable latency.
EI personally use a neon database for testing, and get sub 100 millisecond latency on hot connections without caching -- but your results will vary based on where your requests are coming from and where you neon database is located
TI created one in Frankfurt and I'm based in Malta, so I was getting 200-300ms for a single query (not cached I think), using the hyperdrive with postgres.js. Btw, I haven't managed making hyperdrive work with the serverless neondb driver (via drizzle orm), but only with postgres.js. Maybe the serverless driver of neondb is not supported in cloudflare?
Talso, which driver do you recommend using with hyperdrive? Are you using postgres.js yourself as well, instead of the serverless driver of neon?
MThe serverless Neon driver is supported… it’s just slower than Hyperdrive: https://blog.cloudflare.com/hyperdrive-making-regional-databases-feel-distributed/
The Cloudflare Blog
Hyperdrive makes accessing your existing databases from Cloudflare Workers, wherever they are running, hyper fast.
MAll of our docs recommend Postgres.js and our teams internally use it.
M
Cloudflare Docs
Hyperdrive supports PostgreSQL and PostgreSQL-compatible databases, popular drivers and Object Relational Mapper (ORM) libraries that use those drivers.
TThanks for your kind reply. I was referring that I did not manage to use the serverless Neon driver together with Hyperdrive (not separately!). I don't think hyperdrive can work together with the serverless driver without changing it to postgres.js. More info here:
https://orm.drizzle.team/docs/get-started-postgresql#neon-postgres
https://orm.drizzle.team/docs/get-started-postgresql#neon-postgres
Drizzle ORM is a lightweight and performant TypeScript ORM with developer experience in mind.
MThe Neon serverless driver uses WebSockets — because most serverless providers don't have something like Hyperdrive — so they aren't compatible and can't be.
WebSocket-per-request is still slower than using Hyperdrive's connection pooling (by a fair bit, as the blog lays out)
WebSocket-per-request is still slower than using Hyperdrive's connection pooling (by a fair bit, as the blog lays out)
TThank you very much for your detailed reply, this clears a lot of the confusion I had
May I ask about any future plans for Hyperdrive? It's a really amazing product!
Thank you once again! I'll proceed with using postgres.js.
May I ask about any future plans for Hyperdrive? It's a really amazing product!
Thank you once again! I'll proceed with using postgres.js.
LHello there! Has anyone managed to make a NextJS app hosted on Cloudflare Pages to fetch its Postgres database through Hyperdrive?
AThere was an issue with prepared statements and extra round trips to the DB with postgres.js. Has that been resolved?
AWe saw a reduction in latency by using the node-pg driver
MQuite some time ago - we discussed that here too. https://developers.cloudflare.com/hyperdrive/platform/changelog/#2024-05-22