Hyperdrive is a connection pooler and avoids this exact thing. A few connections are opened from Hyp

Hyperdrive is a connection pooler and avoids this exact thing. A few connections are opened from Hyperdrive, and then every request opens a new connection to Hyperdrive (but not the DB) so that your DB doesn't get overloaded.

Isaac McFadyenOP•10/5/23, 1:43 PM

Because connections are held open from Hyperdrive, the connections to Hyperdrive are very very quick and won't overload things.

IIsaac McFadyen Hyperdrive is a connection pooler and avoids this exact thing. A few connections...

Sam•10/5/23, 1:43 PM

Ahh ok, that's what I was thinking but I was not sure. Thanks for clarifying

Sam•10/5/23, 1:58 PM

For local dev, I'm trying this (pasting a string, if HYPERDRIVE is not available in the env), but I'm getting SSL issues. Is this happening for more people?

Or is my (managed by digital ocean) database cluster not set up correctly?

Sam•10/5/23, 2:08 PM

Also wrangler defaults to 3.0.0 after installing, should I update to 3.10.1?

Llithdew Ran into an issue just now trying to create a Hyperdrive over a PostgreSQL datab...

Matt Silverlock•10/5/23, 3:31 PM

Can you DM me your database URL? There’s not enough info in the issue to debug unfortunately.

SSam Also wrangler defaults to 3.0.0 after installing, should I update to 3.10.1?

Matt Silverlock•10/5/23, 3:31 PM

Yes

JJordan (UNCVRD)hey! I was wondering if this statement is accurate about being able to use Postg...

Matt Silverlock•10/5/23, 8:54 PM

You need to use the latest version of Postgres.jsPostgres.js

we are waiting for them to publish 3.4.03.4.0

JJordan (UNCVRD)also...i was wondering if we'd be able to control the max number of connections ...

Matt Silverlock•10/5/23, 8:54 PM

It's not currently configurable: 20 is very, very low though and would impact performance (fewer conns in the pool == more waiting)

mattfreeman•10/6/23, 6:49 AM

Is the connection string returned by env.HYPERDRIVE.connectionString usable outside of workers, is it accessible on from workstation when connecting over Warp Teams into the same account ?

kian•10/6/23, 8:54 AM

Nope

Nikil•10/7/23, 3:07 PM

So am I understanding Hyperdrive correctly? It's basically the same idea as Cloudflare DNS w/ CDN-style caching but for DBs? That sounds cool...

Kkian Nope

mattfreeman•10/9/23, 5:53 AM

Any plans for this? It would be awesome use case to connect into a zero trust with Cloudflare Warps for team, point DataGrip at the hyperdrive endpoint and presume it gets hyperdriven from a near edge, and enjoy the acceleration of less chat between me and live DB 300ms away.

kian•10/9/23, 7:04 PM

I don't think there's a timeframe on it hyperdrive

Kkian It'd be nice if Wrangler made a warning/error when you use local development w/ ...

kian•10/9/23, 7:06 PM

So it turns out the docs callout I was thinking of is actually in Workers AI, and there isn't one for Hyperdrive. Oops

Kkian Who will win, docs aside or https://github.com/cloudflare/workerd/pull/1266/

kian•10/9/23, 7:13 PM

My bets have shifted over to the PR now meowlul

kian•10/9/23, 7:14 PM

I guess docs asides are pretty easy to miss - it'd be cool for Wrangler to do the warnings (like it does for Queues in remote mode currently).

kian•10/9/23, 7:14 PM

I guess that's hard for remote mode which can support them independently of a new Wrangler release, but local should be fine.

Max (@rozenmd)•10/9/23, 7:16 PM

Converted the hyperdrive example to work with Hono if anyone's curious:

import { Client } from 'pg';
import { Hono, MiddlewareHandler } from 'hono';

type Bindings = {
    HYPERDRIVE: Hyperdrive;
};

const app = new Hono<{
    Bindings: Bindings;
    Variables: {
        DB: Client;
    };
}>();

const dbMiddleware: MiddlewareHandler = async (c, next) => {
    c.set('DB', new Client({ connectionString: c.env.HYPERDRIVE.connectionString }));
    await next();
};

app.get('/', dbMiddleware, async (c) => {
    try {
        // Connect to our database
        await c.var.DB.connect();

        // Test query
        let result = await c.var.DB.query({ text: 'SELECT * FROM pg_tables' });

        // Return result rows as JSON
        return c.json(result);
    } catch (e) {
        console.log(e);
        return Response.json({ error: JSON.stringify(e) }, { status: 500 });
    }
});

export default app;

import { Client } from 'pg';
import { Hono, MiddlewareHandler } from 'hono';

type Bindings = {
    HYPERDRIVE: Hyperdrive;
};

const app = new Hono<{
    Bindings: Bindings;
    Variables: {
        DB: Client;
    };
}>();

const dbMiddleware: MiddlewareHandler = async (c, next) => {
    c.set('DB', new Client({ connectionString: c.env.HYPERDRIVE.connectionString }));
    await next();
};

app.get('/', dbMiddleware, async (c) => {
    try {
        // Connect to our database
        await c.var.DB.connect();

        // Test query
        let result = await c.var.DB.query({ text: 'SELECT * FROM pg_tables' });

        // Return result rows as JSON
        return c.json(result);
    } catch (e) {
        console.log(e);
        return Response.json({ error: JSON.stringify(e) }, { status: 500 });
    }
});

export default app;

Note that I've renamed "client" to "DB", otherwise its the same

meai1•10/10/23, 2:52 PM

I keep getting " not_entitled [code: 10000]" even though I can connect to the db via datagrip just fine. Do I need to create a token permission for hyperdrive or do I need a paid workers plan? It logs me into cloudflare via oauth so I dont know what I'm not entitled for

Mmeai1 I keep getting " not_entitled [code: 10000]" even though I can connect to the db...

Cyb3r-Jak3•10/10/23, 2:53 PM

Workers paid plan is required for hyperdrive

meai1•10/10/23, 2:53 PM

oh okay, thank you

alex (he/him)•10/11/23, 6:39 PM

if hyperdrive keeps connections in a warm pool, does that mean that serverless dbs e.g. neon wouldn't suspend to not pay for compute?

Unsmart•10/11/23, 6:41 PM

If they charge by connection hours then I would assume that would be the case yes

kian•10/11/23, 6:41 PM

There's an initial connection timeout of 15 seconds and an idle connection timeout of 10 minutes.

kian•10/11/23, 6:42 PM

Neon's default auto-suspend delay is 5 minutes

kian•10/11/23, 6:42 PM

I imagine Hyperdrive might let you configure the timeout periods down the line

alex (he/him)•10/11/23, 6:42 PM

Ok that means neon will stay online 10 minutes instead of the 5 minutes?

alex (he/him)•10/11/23, 6:42 PM

Or are idle connection not forcing the db to stay online?

Colin•10/11/23, 10:52 PM

hey y'all - wanted to ask if anyone had opinions regarding Neon w/ Neon's connection pooling vs Neon w/ Hyperdrive (aside from the "one is GA and one isn't")

Colin•10/11/23, 10:53 PM

It seems like the main Hyperdrive-specific features are query caching + analytics (not yet but soon)

Colin•10/11/23, 10:54 PM

does anyone have experience with it / have any thoughts i might be missing?

CColin hey y'all - wanted to ask if anyone had opinions regarding Neon w/ Neon's connec...

Isaac McFadyenOP•10/11/23, 10:55 PM

Hyperdrive will be more efficient. The reason for that is:

Hyperdrive:
Worker --(very low latency)--> Hyperdrive --(persistent connection)--> Neon, so connection establishment from Worker to Hyperdrive is fast
Neon:
Worker --(higher latency)--> Neon Pooler, so while Neon might handle connection pooling the connections will still take a longer amount of time to start

Workers are rapidly spun up/down based on requests, so initial connection time is very important.

Isaac McFadyenOP•10/11/23, 10:56 PM

However, your Neon DB might not sleep very well with inactivity on Hyperdrive. See the conversation above.

IIsaac McFadyen Hyperdrive will be more efficient. The reason for that is: **Hyperdrive:** Work...

Colin•10/11/23, 10:58 PM

I definitely understand this for the

pg

pg

driver, but if we're using

@neondatabase/serverless

@neondatabase/serverless

or similar I'd expect their optimizations to reduce round trips would make it closer at least

Colin•10/11/23, 10:59 PM

i'm not sure how close that'd be to hyperdrive, if it's comparable at all

CColin I definitely understand this for the `pg` driver, but if we're using `@neondatab...

Isaac McFadyenOP•10/11/23, 10:59 PM

Ah - missed that you were using the Serverless driver. That should be fast to start since it's Websockets to the closest Neon node which then uses a long-lived TCP connection.

Isaac McFadyenOP•10/11/23, 11:01 PM

Check out https://neon.tech/demos/ping-thing if you want some preliminary numbers.

Ping Thing

A @neondatabase/serverless demo

Isaac McFadyenOP•10/11/23, 11:01 PM

In particular you want the Seconds field.

Colin•10/11/23, 11:03 PM

yep I totally forgot to mention the serverless driver, my bad. so probably a better way to phrase what I was asking about is "how should we expect hyperdrive w/ standard SQL drivers to perform in comparison to neon w/ serverless drivers"

Colin•10/11/23, 11:04 PM

it still seems like Hyperdrive might be faster as it's much closer to the worker despite having more round trips when establishing a connection(?)

Colin•10/11/23, 11:05 PM

but it seems like there's a lot of Neon fans in here who might've tried this instead of just me theorizing

CColin it still seems like Hyperdrive might be faster as it's much closer to the worker...

Matt Silverlock•10/12/23, 10:39 AM

The Neon Serverless driver still needs to make 3-4 round trips for almost every Workers invocation.

CColin but it seems like there's a lot of Neon fans in here who might've tried this ins...

Matt Silverlock•10/12/23, 10:40 AM

CColin but it seems like there's a lot of Neon fans in here who might've tried this ins...

Matt Silverlock•10/12/23, 10:41 AM

This isn’t Neon’s fault: their driver is great in lieu of other options in many serverless environments, but Hyperdrive exists because that’s still not good enough to get latency down.

import { Client } from 'pg'; import { Hono, MiddlewareHandler } from 'hono'; type Bindings = { HYPERDRIVE: Hyperdrive; }; const app = new Hono<{ Bindings: Bindings; Variables: { DB: Client; }; }>(); const dbMiddleware: MiddlewareHandler = async (c, next) => { c.set('DB', new Client({ connectionString: c.env.HYPERDRIVE.connectionString })); await next(); }; app.get('/', dbMiddleware, async (c) => { try { // Connect to our database await c.var.DB.connect(); // Test query let result = await c.var.DB.query({ text: 'SELECT * FROM pg_tables' }); // Return result rows as JSON return c.json(result); } catch (e) { console.log(e); return Response.json({ error: JSON.stringify(e) }, { status: 500 }); } }); export default app;

Hyperdrive is a connection pooler and avoids this exact thing. A few connections are opened from Hyp

Similar Threads

Similar Threads

Similar Threads