Hi, everyone I have a question with Hyperdrive. On this page: https://developers.cloudflare.com/hype

WWilliam Hi, everyone I have a question with Hyperdrive. On this page: https://developers...

A

AJR•5/4/25, 1:37 PM

That is a bit of an odd way to phrase it, I suppose.

Hyperdrive is a tx-mode connection pooler. It will agressively reuse connections, only opening a new one if a query arrives needing one and none of the currently open ones are available.

We currently have a soft limit of 60 connections to a pool, though can adjust that on request.

Hope that helps, please feel free to ask if you have any other questions!

AAJR That is a bit of an odd way to phrase it, I suppose. Hyperdrive is a tx-mode co...

W

WilliamOP•5/4/25, 2:24 PM

Assume that users of my website may come from around the world, including North America, South America, Europe, Africa, and Asia.
This means there could be 5 Hyperdrive connection pools active at the same time:

Pool1: North America
Pool2: South America
Pool3: Europe
Pool4: Africa
Pool5: Asia

I'm using a Postgres database, and let's say the maximum allowed number of connections to my database is 80.

I saw in the documentation that for the Paid Plan, the "Maximum potential origin database connections" is approximately ~100 connections.

So, with these 5 connection pools combined, will the total number of connections to my database be limited to below 80, or below 100?

WWilliam Assume that users of my website may come from around the world, including North ...

A

AJR•5/4/25, 2:28 PM

We refactored our architecture, there'll be only a single pool near wherever your database is hosted, capped to 60.

We prioritize availability over enforcing this limit, so if the system is stressed your traffic may spill to a second pool, which would be 120 total.

Edit: I see the bit I'm missing.

No, Hyperdrive does not currently consider the maximum connection GUC on your origin database.

WWilliam Assume that users of my website may come from around the world, including North ...

A

AJR•5/4/25, 2:29 PM

We go into some detail on this here, if you're curious: https://blog.cloudflare.com/it-it/how-hyperdrive-speeds-up-database-access/#better-faster-smarter-closer

Il blog di Cloudflare

Pools across the sea: how Hyperdrive speeds up access to databases ...

Hyperdrive, Cloudflare's global connection pooler, relies on some key innovations to make your database connections work. We go deep into the details of what that looks like, in celebration of announcing its availability for Free Plan customers.

AAJR We refactored our architecture, there'll be only a single pool near wherever you...

W

WilliamOP•5/4/25, 2:58 PM

I got it, thanks very much!

S

shay•5/4/25, 6:08 PM

i'm suddenly seeing what i believe to be increased connections from hyperdrive to my postgres server (there are many pg processes each using very little cpu), which is now pegging my server to 100% cpu usage, and that manifests in numerous "PostgresError: Timed out while waiting for an open slot in the pool." as well as CONNECTION_CLOSED errors in my logs. i don't think my traffic has increased proportionally to cause this--i've never noticed sustained CPU usage like this before and my daily requests are about what they've been for the past few weeks. what could be causing this? do i just need to scale up my server?

while writing this i'm noticing a few dips down into the 30-70% range which is more reasonable but the sustained highs are still concerning to me

O

oritwoen•5/5/25, 1:28 PM

Is this features will be available in the future?

https://developers.cloudflare.com/hyperdrive/reference/supported-databases-and-features/#unsupported-mysql-features

Ooritwoen Is this features will be available in the future? https://developers.cloudflare...

A

AJR•5/5/25, 1:29 PM

Some of them might be, depending on user needs. Which ones in particular would you be interested in?

AAJR Some of them might be, depending on user needs. Which ones in particular would ...

O

oritwoen•5/5/25, 1:30 PM

Non utf8 characters ;D

A

AJR•5/5/25, 1:31 PM

Heheh. I'll have to float that by our product folks. CC @thomasgauvin

Candidly I would consider that to not be a high priority right now, given the other things on our plate. I'll see what the team has to say, though.

T

thomasgauvin•5/5/25, 4:28 PM

Hey folks! We're actively talking to teams who are building cross-cloud apps to make sure that we shape Workers VPC (https://blog.cloudflare.com/workers-virtual-private-cloud/) to fit your specific needs. I figured I'd float the opportunity to this channel because I know lots of y'all are using Hyperdrive with Cloudflare tunnels and might be interested in how we're planning on making this more broadly available across Workers. If anyone wants to meet to chat about this, these are my availabilities: https://calendar.app.google/mJma67uLuf4n65Hq6

U

UB•5/5/25, 4:52 PM

I've mentioned it before but it'd still be really nice to have tls support when using local hyperdrive, so we can connect to IE: a planetscale db using local hyperdrive

U

UB•5/5/25, 4:54 PM

Currently this isn't possible because planetscale enforces tls, and doing:

const connection = await mysql.createConnection({
    host: host,
    user: username,
    password: password,
    database: database,
    port: port,
    disableEval: true,
    connectTimeout: 1000,
    ssl: {
        rejectUnauthorized: true,
    }
});

const connection = await mysql.createConnection({
    host: host,
    user: username,
    password: password,
    database: database,
    port: port,
    disableEval: true,
    connectTimeout: 1000,
    ssl: {
        rejectUnauthorized: true,
    }
});

const connection = await mysql.createConnection({
    host: host,
    user: username,
    password: password,
    database: database,
    port: port,
    disableEval: true,
    connectTimeout: 1000,
    ssl: {
        rejectUnauthorized: true,
    }
});

const connection = await mysql.createConnection({
    host: host,
    user: username,
    password: password,
    database: database,
    port: port,
    disableEval: true,
    connectTimeout: 1000,
    ssl: {
        rejectUnauthorized: true,
    }
});

Causes an infinite halt for some unknown reason

T

thomasgauvin•5/5/25, 5:07 PM

Thanks, we're aware of this and have it on our backlog, unfortunately, there are some runtime specific implementations specific to Hyperdrive that make this more complex that regular bindings but we're continuing to make progress on this and will update the discord when it's available

U

UB•5/5/25, 5:19 PM

Do you have a tracking issue on GitHub?

C

CanRau•5/10/25, 3:06 AM

Hi Hyperdrivers

Trying to use Hyperdrive with Planetscale & Kysely I've came across a couple questions and my first try just didn't work, no error, though haven't debugged anything so far, just disabled for the moment, so now to my questions:

Should I use the
```
Optimized Host
```
Optimized Host
```
Optimized Host
```
Optimized Host
:
```
aws.connect.psdb.cloud
```
aws.connect.psdb.cloud
```
aws.connect.psdb.cloud
```
aws.connect.psdb.cloud
or
```
Direct
```
Direct
```
Direct
```
Direct
which in my case is
```
eu-central.connect.psdb.cloud
```
eu-central.connect.psdb.cloud
```
eu-central.connect.psdb.cloud
```
eu-central.connect.psdb.cloud
Creating a Hyperdrive connection does test the connection settings, right? asking because pscale gives me the connection string with
```
?ssl={"rejectUnauthorized":true}
```
?ssl={"rejectUnauthorized":true}
```
?ssl={"rejectUnauthorized":true}
```
?ssl={"rejectUnauthorized":true}
at the end, should I keep it?
The Hyperdrive docs use
```
import { createConnection } from "mysql2/promise";
```
import { createConnection } from "mysql2/promise";
```
import { createConnection } from "mysql2/promise";
```
import { createConnection } from "mysql2/promise";
whereas kysely says
```
import { createPool } from 'mysql2' // do not use 'mysql2/promises'!
```
import { createPool } from 'mysql2' // do not use 'mysql2/promises'!
```
import { createPool } from 'mysql2' // do not use 'mysql2/promises'!
```
import { createPool } from 'mysql2' // do not use 'mysql2/promises'!
(https://www.kysely.dev/docs/getting-started?package-manager=pnpm&dialect=mysql#instantiation
What's the recommended
```
connectionLimit
```
connectionLimit
```
connectionLimit
```
connectionLimit
? Default in kysely docs is 10
think that's it for now also nice weekend to y'all

CCanRau Hi Hyperdrivers 🚗 💨 Trying to use Hyperdrive with Planetscale & Kysely I've ca...

I

Isaac McFadyen•5/10/25, 3:28 AM

Probably direct, because Hyperdrive will handle global routing for you and using the
```
Optimized Host
```
Optimized Host
```
Optimized Host
```
Optimized Host
will likely just add a layer of complexity since they'll be trying to also handle global routing.
It does test connection settings, yes.
You probably don't need a pool because Hyperdrive does pooling for you, so the
```
createConnection
```
createConnection
```
createConnection
```
createConnection
is correct. Note that because of the way Hyperdrive works, it's safe and recommended to create a new connection on every request (and terminate the connection after each request using
```
ctx.waitUntil
```
```
ctx.waitUntil
```
).
5 connections is the maximum, anything after that can result in deadlocking the Worker (not sure if the
```
createConnection
```
createConnection
```
createConnection
```
createConnection
even has a
```
connectionLimit
```
connectionLimit
```
connectionLimit
```
connectionLimit
param because it's not pooling anyway...)

IIsaac McFadyen 1. Probably direct, because Hyperdrive will handle global routing for you and us...

C

CanRau•5/10/25, 3:38 AM

hmm unfortunately kysely's

MysqlDialect

MysqlDialect

MysqlDialect

MysqlDialect only supports

pool

pool

pool

pool connection it seems

CCanRau hmm unfortunately kysely's `MysqlDialect` only supports `pool` connection it see...

I

Isaac McFadyen•5/10/25, 3:41 AM

Hmmm alright, it's probably fine as long as you keep it below 5 connections.

K

knickish•5/12/25, 5:27 PM

Version 5.4.0 of the Cloudflare Terraform Provider has been released. This update includes support for Hyperdrives with the MySQL database type, as well as Custom CA and mTLS support

https://github.com/cloudflare/terraform-provider-cloudflare/releases/tag/v5.4.0

.

.gwapes•5/14/25, 4:01 PM

I am using hyperdrive and I just want to clarify something. I'm using hono and then using the variables to set my db. Do I need to use ctx.waitUntil(sql.end()) or is it not 100% neccessary?

app.all("*", async (c, next) => {
    const sql = postgres(c.env.database.connectionString, {
        max: 5,
        fetch_types: false
    });

    const db = drizzle(sql);

    c.set("database", db);

    await next()
})

app.all("*", async (c, next) => {
    const sql = postgres(c.env.database.connectionString, {
        max: 5,
        fetch_types: false
    });

    const db = drizzle(sql);

    c.set("database", db);

    await next()
})

app.all("*", async (c, next) => {
    const sql = postgres(c.env.database.connectionString, {
        max: 5,
        fetch_types: false
    });

    const db = drizzle(sql);

    c.set("database", db);

    await next()
})

app.all("*", async (c, next) => {
    const sql = postgres(c.env.database.connectionString, {
        max: 5,
        fetch_types: false
    });

    const db = drizzle(sql);

    c.set("database", db);

    await next()
})

..gwapes I am using hyperdrive and I just want to clarify something. I'm using hono and t...

A

AJR•5/14/25, 6:07 PM

It is not strictly necessary, though would be preferred when possible.

AAJR It is not strictly necessary, though would be preferred when possible.

.

.gwapes•5/14/25, 6:11 PM

I added a

c.executionCtx.waitUntil(sql.end())

c.executionCtx.waitUntil(sql.end())

c.executionCtx.waitUntil(sql.end())

c.executionCtx.waitUntil(sql.end())

But then got an error that the client was closed or something before the request finished. I can get it again if you would want to see it

How necessary is it to close? What implications does it have if you don't

A

AJR•5/14/25, 6:13 PM

We'll hold the connection open until it times out Idle or the Worker itself disconnects, so it's just a bit of increased resource. Not the end of the world, but at a high enough scale or with particularly long-lived Workers/connections it may pile up enough to start slowing down your requests.

AAJR We'll hold the connection open until it times out Idle or the Worker itself disc...

.

.gwapes•5/14/25, 6:15 PM

Ok that makes sense, I remember reading the docs about that type of thing. Where would I add the c.executionCtx.waitUntil(sql.end()) to prevent the client being closed before the request completes?

..gwapes Ok that makes sense, I remember reading the docs about that type of thing. Where...

A

AJR•5/14/25, 6:16 PM

Just at the end of your fetch handler, as late as possible in the Worker's control flow.

Y

Yashu Mittal•5/24/25, 5:37 PM

I want to confirm if I understand this correctly, before i make a mistake by exposing the db publicly.

I have a db running on third-party cloud provider, along with a

cloudflared

cloudflared

container in the same network.
Now, from this docs: https://developers.cloudflare.com/hyperdrive/configuration/connect-to-private-database/#12-connect-your-database-using-a-public-hostname

I need to create a public hostname

<random_name>.<my_domain>.com/

<random_name>.<my_domain>.com/

<random_name>.<my_domain>.com/

<random_name>.<my_domain>.com/
and point it to db

mysq://...

mysq://...

mysq://...

mysq://... with TCP type

Question:

If I link my db to make it accessible publicity via endpoint wouldn't anyone access it?
Do I need to enter the db username and password in type TCP url as well?

When I create a hyperdrive configuration, I can see it does ask for the db name, username and password.
Does that mean, that I do not have to add username, password and db name in the tcp url?

A

AJR•5/26/25, 12:36 AM

We can do it for you on request, please feel free to DM.

Making this self-service is a known feature gap, and we do plan to address it. However we have not prioritized it onto our roadmap for the upcoming few months. We do not yet have a committed timeline for when we will deliver this feature.

K

kindaraffy•5/26/25, 12:55 PM

Does anybody else have issues with their drizzle transaction's being complete cache misses? I know hyperdrive struggles to cache transactions, but my transactions are completely independent queries (I have a root transaction carrying individual, child transactions. It's these child transactions I'm expecting to be cached). I'm using

postgres.js

postgres.js

postgres.js

postgres.js and

{prepare: true}

{prepare: true}

{prepare: true}

{prepare: true}. Are transactions known not to work with hyperdrive at all?

I wanted to create a reproducible repo, but I wanted to see if anybody else had this problem.

A

AJR•5/26/25, 1:51 PM

We do not cache any queries conducted within a transaction, by design.

That said, Hyperdrive should be perfectly capable of handling transactions, with all the state correctly handled throughout.

Are you having any issues with things besides caching?

A

AJR•5/26/25, 1:56 PM

As an aside, I genuinely wish the Postgres wire protocol had commit/rollback status encoded into its ReadyForQuery messages. It would make caching tx queries so much less risky.

D

Dylan•5/27/25, 2:05 AM

hello everyone,

i'm having this error that I haven't quite know how to solve yet. it's related to using neon with postgres.js and hyperdrive
i'm trying to set up a connection as follows:

import { Context } from 'hono';
import postgres from 'postgres';

const db = (c: Context) => {
  console.log('Connecting to database...');
  console.log(c.env.HYPERDRIVE.connectionString);

  const connectionString = c.env.HYPERDRIVE?.connectionString;

  return postgres(connectionString, {
    fetch_types: false
  });
};

export default db;

import { Context } from 'hono';
import postgres from 'postgres';

const db = (c: Context) => {
  console.log('Connecting to database...');
  console.log(c.env.HYPERDRIVE.connectionString);

  const connectionString = c.env.HYPERDRIVE?.connectionString;

  return postgres(connectionString, {
    fetch_types: false
  });
};

export default db;

import { Context } from 'hono';
import postgres from 'postgres';

const db = (c: Context) => {
  console.log('Connecting to database...');
  console.log(c.env.HYPERDRIVE.connectionString);

  const connectionString = c.env.HYPERDRIVE?.connectionString;

  return postgres(connectionString, {
    fetch_types: false
  });
};

export default db;

import { Context } from 'hono';
import postgres from 'postgres';

const db = (c: Context) => {
  console.log('Connecting to database...');
  console.log(c.env.HYPERDRIVE.connectionString);

  const connectionString = c.env.HYPERDRIVE?.connectionString;

  return postgres(connectionString, {
    fetch_types: false
  });
};

export default db;

but this is the output that I'm getting in the console:

Connecting to database...
postgresql://neondb_owner:*************@91a7****************d1e4c29.hyperdrive.local:5432/neondb?sslmode=disable
✘ [ERROR] PostgresError: connection is insecure (try using `sslmode=require`)

Connecting to database...
postgresql://neondb_owner:*************@91a7****************d1e4c29.hyperdrive.local:5432/neondb?sslmode=disable
✘ [ERROR] PostgresError: connection is insecure (try using `sslmode=require`)

Connecting to database...
postgresql://neondb_owner:*************@91a7****************d1e4c29.hyperdrive.local:5432/neondb?sslmode=disable
✘ [ERROR] PostgresError: connection is insecure (try using `sslmode=require`)

Connecting to database...
postgresql://neondb_owner:*************@91a7****************d1e4c29.hyperdrive.local:5432/neondb?sslmode=disable
✘ [ERROR] PostgresError: connection is insecure (try using `sslmode=require`)

i'm not sure what I'm supposed to do since the url provided by hyperdrive has sslmode=false.
can anyone point me in the right direction?

AAJR Local dev mode (i.e. `wrangler dev`) does not currently support connecting to re...

D

Dylan•5/27/25, 2:36 AM

oooh this explains it, shouldve joined the discord server sooner.

wrangler dev --remote

wrangler dev --remote

wrangler dev --remote

wrangler dev --remote works perfectly

S

splon•5/28/25, 6:45 PM

Hi, probably a noob question, but I'm connecting to a distant Neon db, and the latencies I'm seeing suggest that my query always takes 2 round trips (350ms latency, while pinging the Neon server takes 150ms and VPNing to a location near the server drops the latency to 50ms). This is even if I just hit Neon a few seconds earlier. I connect using

postgres(connStr, {max: 5, fetch_types: false,})

postgres(connStr, {max: 5, fetch_types: false,})

postgres(connStr, {max: 5, fetch_types: false,})

postgres(connStr, {max: 5, fetch_types: false,}) (not that I need more than one connection but the docs suggested setting max to 5) and I have the cache disabled since the purpose of my query is to catch updates to a counter due to mutations by other clients. My app is not yet publicly released, so there is very little traffic.

Ssplon Hi, probably a noob question, but I'm connecting to a distant Neon db, and the l...

A

AJR•5/28/25, 11:06 PM

postgres.js will often take multiple round trips for a query. For your use case I'd try switching to node-postgres, it should do each of your queries in a single round trip.

AAJR postgres.js will often take multiple round trips for a query. For your use case...

I

Isaac McFadyen•5/29/25, 4:35 AM

that's kinda interesting, any idea why postgres.js does and node-postgres doesn't? I feel like I've seen it mentioned before but don't remember what the reason given was

A

AJR•5/29/25, 1:34 PM

Sure, I can talk through that. It's very "in the weeds" and is definitely the sort of thing we prefer to avoid making our users have to deal with, but I find it interesting.

So all of this has to do with Bind messages (https://www.postgresql.org/docs/current/protocol-message-formats.html) and how you handle the parameter values for each bound parameter. There are two formats available,

binary

binary

and

text

text

. Postgres.js uses the binary format, while node-postgres uses the text format.

Binary format avoids needing text parsing, is smaller on the wire, and can in some scenarios be more precise. However, you generally need to get the actual format codes for each type from the server. To do this, postgres.js structures all new prepared statements (and therefore most new queries), as a Parse-Describe-Flush sequence, which will get back the ParameterDescription containing the binary format codes for each parameter. It then uses that information to construct the Bind message, and sends a followup Bind-Execute-Sync sequence, to get the actual results.

Node-postgres just builds the whole thing in one shot, as a Parse-Bind-Execute-Sync sequence. This pulls back the results in a single round-trip, at the cost of making the origin DB do some extra work handling the Bind message, and with the risk of some loss of precision in types like

float

float

or

timestamp

timestamp

.

Which is most appropriate, as always, depends

PostgreSQL Documentation

53.7. Message Formats

53.7. Message Formats # This section describes the detailed format of each message. Each is marked to indicate that it can …

AAJR Sure, I can talk through that. It's very "in the weeds" and is definitely the so...

I

Isaac McFadyen•5/30/25, 4:44 AM

That's kinda interesting. I assume the overhead is minimal for serverful applications that run all the time but is worse for Hyperdrive? Is that message needed only at driver connection time or for all queries?

TTyveris With Aurora DSQL now generally available, any chance we will get improved Hyperd...

A

AJR•5/30/25, 1:18 PM

That's not currently on our roadmap, but it's something we've kicked around internally a bit.

Ultimately we don't get a lot of requests for it relative to observability/stability/other features, so it's unlikely to happen very soon unless we hear a bunch of people wanting it.

CC @thomasgauvin , in case he has more/different thoughts to share here.

A

AJR•5/30/25, 1:19 PM

That does raise a good question, though.

How many folks around here are using DSQL in anger? How's your experience been, and is it something you'd be particularly interested in having more robust Hyperdrive support for? I haven't heard a ton about it after the initial launch, aside from the official AWS blog posts, and I'd be very curious to know what everyone is seeing about it.

IIsaac McFadyen That's kinda interesting. I assume the overhead is minimal for serverful applica...

A

AJR•5/30/25, 3:04 PM

I think that's more or less correct, the round trips are most punishing for short-lived connections that don't re-use the prepared statement much.

That attribute fetch is needed for each new prepared statement, so if you re-run the same query with different params multiple times you only pay the overhead once. If you run N different extended-protocol queries you'll pay the overhead on each one.

D

dave•5/30/25, 4:06 PM

Still no support for connecting to Hyperdrive from a non-Worker source, right?

Ddave Still no support for connecting to Hyperdrive from a non-Worker source, right?

A

AJR•5/30/25, 4:07 PM

Correct. Though we're actively keeping an eye on cloudchamber and how we can best integrate with it once it's in a good spot to do so, if you count that.

AAJR That does raise a good question, though. How many folks around here are using...

D

dave•5/30/25, 4:12 PM

I was interested in DSQL at first, but IMO sticking "regular" AWS RDS Aurora Serverless v2 is way easier. There's a bunch of extra PostgreSQL features that DSQL doesn't support.

D

dave•5/30/25, 4:16 PM

DSQL also seems like 10x more expensive per query, and 3x more expensive on storage?

Ddave I was interested in DSQL at first, but IMO sticking "regular" AWS RDS Aurora Ser...

D

dave•5/30/25, 4:17 PM

the lack of foreign keys is pretty terrible

D

dave•5/30/25, 4:18 PM

imo the main benefit I see of DSQL over RDS Aurora Serverless v2 is support for public IPv6 lol

Hi, everyone I have a question with Hyperdrive. On this page: https://developers.cloudflare.com/hype

Similar Threads