I'd like to use Neon with Hyperdrive, but I see conflicting documentation whether to turn on/off Neo

I'd like to use Neon with Hyperdrive, but I see conflicting documentation whether to turn on/off Neon's own pooling.

Cloudflare's docs says to turn it off:
https://developers.cloudflare.com/hyperdrive/examples/neon/#:~:text=uncheck%20the%20connection%20pooling

Neon's docs says to turn it on:
https://neon.tech/blog/hyperdrive-neon-faq#so-should-i-use-hyperdrive-together-with-neons-pooling

WHO DO I TRUST

AJR•2/25/25, 8:08 PM

Since you're asking in the Hyperdrive discord, I vote you trust us

AJR•2/25/25, 8:10 PM

Serious answer: if it works for you with both, it's fine to have it on. We discourage it because it isn't going to make a ton of difference for most users, and adds a lot of complexity to your traffic flow. In particular the poolers for some hosting services really struggle with prepared statements, and some popular JS/Rust postgres drivers make very heavy use of those.

JikyuOP•2/25/25, 8:10 PM

Thank you for the answer!!

AgentCord•2/27/25, 4:01 PM

Guys, why is this query not cached, is it possible to cache

users

users

?

It is fetching a user based on a provided API key, updating API key's "lastCall" field, and returning the user info:

WITH updated_keys AS (
  UPDATE keys
  SET "lastCall" = NOW()
  WHERE "key" = ${token} 
  RETURNING "key", "userId"
)
SELECT 
  uk.key,
  u.id      AS user_id,
  u.balance AS user_balance,
  u.mail    AS user_mail
FROM updated_keys uk
JOIN users u ON uk."userId" = u.id;

WITH updated_keys AS (
  UPDATE keys
  SET "lastCall" = NOW()
  WHERE "key" = ${token} 
  RETURNING "key", "userId"
)
SELECT 
  uk.key,
  u.id      AS user_id,
  u.balance AS user_balance,
  u.mail    AS user_mail
FROM updated_keys uk
JOIN users u ON uk."userId" = u.id;

AJR•2/27/25, 4:40 PM

Two reasons:

an UPDATE statement mutates database state, can't cache that.
the NOW() function is volatile, so that can't be cached either

If you were just querying the SELECT, separate from that CTE, that should cache fine.

kchro3•2/27/25, 7:54 PM

are there any recommendations on how to update hyperdrive if you want to rotate db passwords?

Kkchro3 are there any recommendations on how to update hyperdrive if you want to rotate ...

AJR•2/27/25, 7:55 PM

https://developers.cloudflare.com/hyperdrive/configuration/rotate-credentials/

Cloudflare Docs

Rotating database credentials · Hyperdrive

You can change the connection information and credentials of your Hyperdrive configuration in one of two ways:

Walshy•2/27/25, 7:57 PM

@Hyperdrive ^

Isaac McFadyen•2/27/25, 8:14 PM

@Nick I remember you asking about this before ^

JJikyu I'd like to use Neon with Hyperdrive, but I see conflicting documentation whethe...

AJ•2/28/25, 2:05 AM

I think Neon's doc says they do not recommend using Neon's pool with Hyperdrive, which is aligned with Hyperdrive's doc.

JikyuOP•2/28/25, 2:06 AM

Yup using hyperdrive with direct connection to Neon, got set up in like 10 minutes and working great

WWalshy @Hyperdrive ^

Xpertfxglobal•2/28/25, 3:32 AM

Hello

Heliodorofl42•3/1/25, 9:33 AM

@Walshy

HHeliodorofl42 @Walshy

Isaac McFadyen•3/1/25, 9:10 PM

?pings

SuperHelpflare•3/1/25, 9:10 PM

Please do not ping community members for non-moderation reasons. Doing so will not solve your issue faster and will make people less likely to want to help you.

matt•3/2/25, 10:43 AM

does hyperdrive over cloudflared tunnel require dns records?

matt•3/2/25, 10:44 AM

no wait, it seems to want SSL. shouldn't the tunnel already be encrypting traffic?

matt•3/2/25, 10:58 AM

can i use self-signed certs? connection still failing

matt•3/2/25, 10:59 AM

very little info being provided from the web ui just says 403 forbidden

matt•3/2/25, 10:59 AM

cant see any activity in the 2 cloudflared instance logs

matt•3/2/25, 11:02 AM

version of cloudflared is 2024.9.12024.9.1 is that an issue? im using the cloudflare-operator for k8s and unsure how to upgrade the binary

matt•3/2/25, 11:06 AM

code=2015
Failed to connect to the provided database: Connecting to database via Cloudflare Tunnel failed: 403 Forbidden

code=2015
Failed to connect to the provided database: Connecting to database via Cloudflare Tunnel failed: 403 Forbidden

matt•3/2/25, 11:23 AM

what the heck is going on. this is the tunnel configuration

- hostname: postgres.unweb.io
      service: tcp://postgresql-f4b1ec99.cloud-prod.svc:5432
      originRequest:
        noTLSVerify: true

- hostname: postgres.unweb.io
      service: tcp://postgresql-f4b1ec99.cloud-prod.svc:5432
      originRequest:
        noTLSVerify: true

ive confirmed multiple boxes in the cluster can access via psql, and im still not seeing anything in the cloudflared logs

matt•3/2/25, 12:10 PM

none of this has worked:

enable SSL on postgres
disable TLS verification for tunnel backend due to self-signed cert
upgrade version of cloudflared to 2025.2.12025.2.1

meanwhile the tunnel diagnostics are showing absolutely zero information. i cannot see hyperdrive attempting to make the connection. i've torn down the tunnel and recreated from scratch, nothing seems to work and the inability to see any diagnostics or logs is extremely frustrating

Mmatt none of this has worked: * enable SSL on postgres * disable TLS verification for...

AJR•3/2/25, 12:47 PM

Run the tunnel with loglevel=debug to see individual connection attempts and responses.

AJR•3/2/25, 12:48 PM

403 is usually an Access refusal. Check the logs on your access policy too? For service tokens you'll need to query them via API, the tunnel dash doesn't show service access login attempts

Mmatt does hyperdrive over cloudflared tunnel require dns records?

AJR•3/2/25, 12:48 PM

Yes it does

Mmatt version of cloudflared is `2024.9.1` is that an issue? im using the cloudflare-o...

AJR•3/2/25, 12:48 PM

Should be fine with this version

matt•3/2/25, 12:50 PM

I had to abandon hyperdrive, I noticed you're limited to 25 db connections which bundles auth and my per-user auth reaches to the db so i'd quickly hit the 25 limit

matt•3/2/25, 12:52 PM

one thought i had while i was prototyping is that it shouldn't require public DNS records

load balancer works without them and provides much more security by not exposing your db endpoint publically

matt•3/2/25, 12:55 PM

there probably needs to a bigger push to allow any cloudflare service to connect to tunnels without having them exposed publically

Mmatt one thought i had while i was prototyping is that it shouldn't require public DN...

AJR•3/2/25, 2:28 PM

This is something we're tinkering with. The trick is to still take advantage of Access for security without a hostname.

Mmatt I had to abandon hyperdrive, I noticed you're limited to 25 db connections which...

AJR•3/2/25, 2:29 PM

Hyperdrive is a transaction mode pooler, so those 25 go a lot further than you might imagine. Also the default limit is 60 after last month, with plans to make it adjustable down the road a bit.

Not saying you should reconsider, but just stating for posterity

kchro3•3/4/25, 8:11 PM

Hi, I am seeing a ton of logs like this:
Error: write CONNECTION_CLOSED <...>.hyperdrive.local:5432

can someone explain what this does

kchro3•3/4/25, 8:16 PM

we had as high as 20 errors per second and a couple hundred failures over a 10 min period

kchro3•3/4/25, 8:21 PM

brandonpsmith•3/5/25, 5:46 AM

i am using neon. i use neon branches for local development. when i set my neon branch connection string in .dev.vars (WRANGLER_HYPERDRIVE_LOCAL_CONNECTIONSTRING<BINDING>=neon branch connection string) and run wrangler dev. it recognizes my environment variable

Found a non-empty WRANGLER_HYPERDRIVE_LOCAL_CONNECTION_STRING variable for binding. Hyperdrive will connect to this database during local development.

Found a non-empty WRANGLER_HYPERDRIVE_LOCAL_CONNECTION_STRING variable for binding. Hyperdrive will connect to this database during local development.

now when i get the connection string from the binding (BINDING.connectionString), it returns a hyperdrive.localhyperdrive.local host instead of my neon connection string.

what am i doing wrong?

kchro3•3/5/25, 2:15 PM

has there been an uptick in the number of these Error: write CONNECTION_CLOSED <...>.hyperdrive.local:5432Error: write CONNECTION_CLOSED <...>.hyperdrive.local:5432 errors?

I've been getting tons of these starting yesterday, and even though I've added retry logic, it's still concerning & a waste of CPU time

AJR•3/5/25, 2:18 PM

There was an accelerated release yesterday that may have severed some connections. We'll need to investigate that, there shouldn't have been the level of disruption you're describing.

Are you still observing elevated error rates? I'm not seeing any such in our broader monitoring.

kchro3•3/5/25, 2:23 PM

yup, here's our 24hr view

kchro3•3/5/25, 2:24 PM

we haven't made any significant code changes before we started seeing errors, except to add retries

AJR•3/5/25, 2:24 PM

I think you were working with Adrian? Let me touch base with him

kchro3•3/5/25, 2:24 PM

yes, thank you

kchro3•3/5/25, 8:56 PM

still seeing this error... please let me know if there are things that I can try as mitigations or if there are any updates

for now, i'm trying to see what'll happen if i disable hyperdrive caching since i have a separate cache in front of my postgres client

AJR•3/5/25, 9:27 PM

We're investigating internally. We will share updates as we have them.

iano•3/5/25, 10:04 PM

We are also seeing increased CONNECTION CLOSED errors, although also are dealing with increased traffic so hard to tell if its the same % or not

AJR•3/5/25, 10:05 PM

same ID as in our thread?

Iiano We are also seeing increased CONNECTION CLOSED errors, although also are dealing...

AJR•3/5/25, 10:36 PM

Went looking. I do see a bump in disconnects for you, but it's not related to what kchro3 is dealing with.

Franco Romano Losada•3/6/25, 3:55 AM

Hey everyone, I'm testing hyperdrive. These are my results:
/api/test/ ---> 200ms first hit, 3-7ms subsequent calls. Great!
/api/test/:id ---> 200-600ms always.

I see in the Hyperdrive analytics that I have a lot of "Uncacheable" requests. I guess this is the dynamic endpoint.

I'm trying to understand why or how it works? Is there a workaround? Does hyperdrive always relay on caching? I'm wondering for the stale data too.

Thanks!

I'd like to use Neon with Hyperdrive, but I see conflicting documentation whether to turn on/off Neo

Similar Threads

Similar Threads

Similar Threads