Hello everyone,👋 Can someone explain me how I can have a median cpu time of 78ms while the limit is

A

anurag•3/1/24, 11:51 AM

Cloudflare Worker is waiting for the whole response before streaming the output to the client, which defeats the entire point of streaming. I don't know what I am doing wrong.

router.post("/v1/streaming", async (request, { OPENAI_API }) => {
    let { readable, writable } = new TransformStream();
    const client = new ChatOpenAI({
      openAIApiKey: OPENAI_API,
    });

    const stream = await client.stream(
      "Write a 2 sentence poem about a sunset."
    );
    let writer = writable.getWriter();
    const textEncoder = new TextEncoder();

    for await (const part of stream) {
      if (part.content.length) {
        // Assuming part.content is a string containing the text
        const words = part.content.split(" "); // Split the text into words
        for (const word of words) {
          const sseFormattedData = `data: ${word}\n\n`; // Format each word as SSE
          await writer.write(textEncoder.encode(sseFormattedData));
          console.log("🚀 - forawait - sseFormattedData:", sseFormattedData);
          await new Promise((r) => setTimeout(r, 1000)); // Wait for 1 second before sending the next word
        }
      }
    }

    writer.close();

    return new Response(readable, {
      // headers: {
      //   "Content-Type": "text/event-stream",
      //   "Cache-Control": "no-cache",
      //   Connection: "keep-alive",
      // },
    });
  })

router.post("/v1/streaming", async (request, { OPENAI_API }) => {
    let { readable, writable } = new TransformStream();
    const client = new ChatOpenAI({
      openAIApiKey: OPENAI_API,
    });

    const stream = await client.stream(
      "Write a 2 sentence poem about a sunset."
    );
    let writer = writable.getWriter();
    const textEncoder = new TextEncoder();

    for await (const part of stream) {
      if (part.content.length) {
        // Assuming part.content is a string containing the text
        const words = part.content.split(" "); // Split the text into words
        for (const word of words) {
          const sseFormattedData = `data: ${word}\n\n`; // Format each word as SSE
          await writer.write(textEncoder.encode(sseFormattedData));
          console.log("🚀 - forawait - sseFormattedData:", sseFormattedData);
          await new Promise((r) => setTimeout(r, 1000)); // Wait for 1 second before sending the next word
        }
      }
    }

    writer.close();

    return new Response(readable, {
      // headers: {
      //   "Content-Type": "text/event-stream",
      //   "Cache-Control": "no-cache",
      //   Connection: "keep-alive",
      // },
    });
  })

router.post("/v1/streaming", async (request, { OPENAI_API }) => {
    let { readable, writable } = new TransformStream();
    const client = new ChatOpenAI({
      openAIApiKey: OPENAI_API,
    });

    const stream = await client.stream(
      "Write a 2 sentence poem about a sunset."
    );
    let writer = writable.getWriter();
    const textEncoder = new TextEncoder();

    for await (const part of stream) {
      if (part.content.length) {
        // Assuming part.content is a string containing the text
        const words = part.content.split(" "); // Split the text into words
        for (const word of words) {
          const sseFormattedData = `data: ${word}\n\n`; // Format each word as SSE
          await writer.write(textEncoder.encode(sseFormattedData));
          console.log("🚀 - forawait - sseFormattedData:", sseFormattedData);
          await new Promise((r) => setTimeout(r, 1000)); // Wait for 1 second before sending the next word
        }
      }
    }

    writer.close();

    return new Response(readable, {
      // headers: {
      //   "Content-Type": "text/event-stream",
      //   "Cache-Control": "no-cache",
      //   Connection: "keep-alive",
      // },
    });
  })

router.post("/v1/streaming", async (request, { OPENAI_API }) => {
    let { readable, writable } = new TransformStream();
    const client = new ChatOpenAI({
      openAIApiKey: OPENAI_API,
    });

    const stream = await client.stream(
      "Write a 2 sentence poem about a sunset."
    );
    let writer = writable.getWriter();
    const textEncoder = new TextEncoder();

    for await (const part of stream) {
      if (part.content.length) {
        // Assuming part.content is a string containing the text
        const words = part.content.split(" "); // Split the text into words
        for (const word of words) {
          const sseFormattedData = `data: ${word}\n\n`; // Format each word as SSE
          await writer.write(textEncoder.encode(sseFormattedData));
          console.log("🚀 - forawait - sseFormattedData:", sseFormattedData);
          await new Promise((r) => setTimeout(r, 1000)); // Wait for 1 second before sending the next word
        }
      }
    }

    writer.close();

    return new Response(readable, {
      // headers: {
      //   "Content-Type": "text/event-stream",
      //   "Cache-Control": "no-cache",
      //   Connection: "keep-alive",
      // },
    });
  })

K

kian•3/1/24, 2:22 PM

Not currently.

DDragon Hi @Rasel Rana Rocky - Do you have a sample project to get started with a simple...

R

Rasel Rana Rocky•3/1/24, 3:57 PM

Not yet. I'm cooking a terraform solution. Once I'm done, I'm gonna make a boilerplate around the whole thing to kick start projects.

Nnwdles Hi there! I have two questions. 1. Is it possible to create a token with permis...

C

Cyb3r-Jak3•3/1/24, 4:02 PM

1. No, api tokens get all workers. 2. I believe the /aaa/ will only fire on domain.com/aaa

S

Sean Kerr•3/1/24, 4:08 PM

Is it possible to have multiple workers run on the same custom domain, or do I need to have one worker that handles routes for the entire domain? It seems when I try to deploy a second worker, it wants to link the new worker the the already existing custom domain and remove it from the old worker. If this is the case, what is the point of the Workers Routes section under the domain?

Nnwdles thank you, and also, sorry, it's formatting, I asked about `* / bbb/ *` and `* /...

C

Cyb3r-Jak3•3/1/24, 4:09 PM

I know the most specific route wins but not sure how it would be calculated. I can play around with it later today and see

K

kian•3/1/24, 4:13 PM

I don't think you can even have infix wildcards can you?

C

Chaika•3/1/24, 4:14 PM

There's a ton of docs on routes here: https://developers.cloudflare.com/workers/configuration/routing/routes/#validity

C

Chaika•3/1/24, 4:15 PM

Route patterns may not contain infix wildcards or query parameters. For example, neither
example.com/*.jpg
example.com/*.jpg
example.com/*.jpg
example.com/*.jpg nor
example.com/?foo=*
example.com/?foo=*
example.com/?foo=*
example.com/?foo=* are valid route patterns.
When more than one route pattern could match a request URL, the most specific route pattern wins. For example, the pattern
www.example.com/*
www.example.com/*
www.example.com/*
www.example.com/* would take precedence over
*.example.com/*
*.example.com/*
*.example.com/*
*.example.com/* when matching a request for
https://www.example.com/
https://www.example.com/
https://www.example.com/
https://www.example.com/. The pattern
example.com/hello/*
example.com/hello/*
example.com/hello/*
example.com/hello/* would take precedence over
example.com/*
example.com/*
example.com/*
example.com/* when matching a request for example.com/hello/world.

RRayane Hello everyone,👋 Can someone explain me how I can have a median cpu time of 78m...

R

RayaneOP•3/1/24, 5:08 PM

Up

RRayane Up ⬆️

C

Chaika•3/1/24, 5:14 PM

I would read this: How are CPU limits enforced in the free tier?

C

Chaika•3/1/24, 5:14 PM

tldr being you don't have enough requests to trigger it/the limit is loose and you get an initial quota

C

Chaika•3/1/24, 5:15 PM

if you get enough requests, and those requests all exceed the time, you'd start seeing errors

C

Chaika•3/1/24, 5:20 PM

(isolates in that is referring to how Workers run. When you send a request to a Worker, it hits a CF machine/metal, and creates an Isolate for your Worker if there isn't one already on that machine. You stay connected to the same machine thanks to keepalive, but otherwise routing is random/doesn't prefer metals where you already have isolates afaik, so with little requests over a decent period of time, you could exceed the limit regularly until you get enough requests/popular enough)
Not really abusable because if your service becomes popular/gets a consistent amount of requests you'll burn through the initial quota and have to stay within cpu time

HHello, I’m Allie!30 seconds of CPU time, and theoretically indefinite runtime, though it may be e...

J

Joy•3/1/24, 7:06 PM

Hey @HardlyWorkin'
is it 10ms or 30s of CPU time for free plan. Because in a blog I found it to be 30s for standard plan
Blog Link : https://blog.cloudflare.com/workers-pricing-scale-to-zero/

The Cloudflare Blog

New Workers pricing — never pay to wait on I/O again

Announcing new pricing for Cloudflare Workers, where you are billed based on CPU time, and never for the idle time that your Worker spends waiting on network requests and other I/O.

K

kian•3/1/24, 7:08 PM

10ms

K

kian•3/1/24, 7:08 PM

Standard pricing is part of the Workers Paid plan

J

jacob•3/1/24, 7:11 PM

Hey everyone, I'm building a basic Discord bot that needs to make use of some minor assets for message embeds, i.e images and I'm experimenting with doing some light image processing. Is it possible to keep a static file like an image and have the worker be aware of it when executed or is it run in complete isolation? Essentially right now I'm just trying to figure out how I can get the path of it and whether it is even worth trying

J

jacob•3/1/24, 7:12 PM

And if it is completely isolated is there any way I could possibly "import" the asset so that it becomes part of the function?

Jjacob Hey everyone, I'm building a basic Discord bot that needs to make use of some mi...

I

Isaac McFadyen•3/1/24, 7:13 PM

Yes you can embed it, but there's no filesystem so there's no "get the path" of it.

I

Isaac McFadyen•3/1/24, 7:13 PM

But you can embed it as a static asset. One second.

I

Isaac McFadyen•3/1/24, 7:14 PM

I believe you can just import it as an asset, i.e.

import Image from "./image.png"

import Image from "./image.png"

import Image from "./image.png"

import Image from "./image.png" and then Wrangler will bundle it for you.

I

Isaac McFadyen•3/1/24, 7:14 PM

Keep in mind that will give you the raw image data, not a path to it.

J

jacob•3/1/24, 7:15 PM

Yeah that's fine, looks like I might need a loader to support it. But I will look into it thanks!

Kkian Standard pricing is part of the Workers Paid plan

J

Joy•3/1/24, 7:17 PM

I am only using a Cloudflare Worker for one of my tasks, and I need it to have at least 15 seconds of CPU time. So, opting for a standard plan is the only solution, correct?

JJoy I am only using a Cloudflare Worker for one of my tasks, and I need it to have a...

I

Isaac McFadyen•3/1/24, 7:19 PM

So CPU Time != Wall Time. Wall time is the actual time it takes (example: fetching a slow API is wall time) whereas CPU time is the actual work done by the CPU (i.e. hashing an asset is CPU intensive).

I

Isaac McFadyen•3/1/24, 7:19 PM

You get 10ms of CPU Time on the free plan. Even if you use more wall time, that's not CPU time and might still be allowed.

I

Isaac McFadyen•3/1/24, 7:20 PM

For example, if I fetch an external API that takes 45 seconds (on the free plan!) then it will work, because that's only maybe like 2ms of CPU time

Jjacob Yeah that's fine, looks like I might need a loader to support it. But I will loo...

K

kian•3/1/24, 7:20 PM

https://developers.cloudflare.com/workers/wrangler/configuration/#bundling

Kkian <https://developers.cloudflare.com/workers/wrangler/configuration/#bundling>

I

Isaac McFadyen•3/1/24, 7:20 PM

TY, this is what I was looking for

Kkian <https://developers.cloudflare.com/workers/wrangler/configuration/#bundling>

J

jacob•3/1/24, 7:21 PM

Thanks!

IIsaac McFadyen For example, if I fetch an external API that takes 45 seconds (on the free plan!...

J

Joy•3/1/24, 7:22 PM

My use case is a little bit different.
I need to make 10-15 API calls, but there's a rate limit of 1 API call per second.
What are all the possible solutions for this use case. Can you please suggest.

JJoy My use case is a little bit different. I need to make 10-15 API calls, but there...

I

Isaac McFadyen•3/1/24, 7:23 PM

So this is allowed on the default Workers Free plan. Each API call might take a second, but you can easily make 10 or 15 of them on the Free plan because each call might only take 1ms of CPU time. It's good to test, you might need the paid plan, but likely not.

IIsaac McFadyen So this is allowed on the default Workers Free plan. Each API call might take a ...

J

Joy•3/1/24, 7:25 PM

The API will get resolve in few ms. But then I have to wait for 1s before I make another one. So my question is will this 1s be treated as CPU time or wall time?

JJoy The API will get resolve in few ms. But then I have to wait for 1s before I make...

I

Isaac McFadyen•3/1/24, 7:25 PM

Wall time. CPU time is only the work that the actual CPU does (processing power, basically) so waiting for a second or waiting on an outgoing request will not consume CPU time.

IIsaac McFadyen Wall time. CPU time is only the work that the actual CPU does (processing power,...

J

Joy•3/1/24, 7:43 PM

Alright. Let me try it then.
Also thank you so much for helping me out.

B

bun•3/1/24, 8:09 PM

when doing local worker development and using R2. I see that

.wrangler/state/v3/r2

.wrangler/state/v3/r2

is created containing my files, but are they only accessible through the binding? does wrangler cli run some alternative local server to serve files ?

C

Cyb3r-Jak3•3/1/24, 8:12 PM

Wrangler local commands should also use it or you can use miniflare directly

B

bun•3/1/24, 8:12 PM

to launch a r2 server ?

C

Cyb3r-Jak3•3/1/24, 8:17 PM

What do you mean by launch a server? Wrangler doesn’t have a way to serve them over http if that’s what you’re looking for

W

Walshy•3/1/24, 9:53 PM

you should set

compatibility_flags = ["nodejs_compat"]

compatibility_flags = ["nodejs_compat"]

rather than node_compat

(compat flags uses runtime node APIs,

node_compat

node_compat

uses polyfills)

then usage would just be like

import { createHash } from 'node:crypto';

function sha256(text) {
  return createHash('sha256').update(text).digest('hex')
}

import { createHash } from 'node:crypto';

function sha256(text) {
  return createHash('sha256').update(text).digest('hex')
}

import { createHash } from 'node:crypto';

function sha256(text) {
  return createHash('sha256').update(text).digest('hex')
}

import { createHash } from 'node:crypto';

function sha256(text) {
  return createHash('sha256').update(text).digest('hex')
}

P

piffie•3/1/24, 10:16 PM

Is there a reason why github wrangler-action per default uses an old version of wrangler for deployment?

B

bun•3/2/24, 12:14 AM

when doing

fetch

fetch

, one of the enterprise features is to cache the fetch by a specific cacheKey, although for non-enterprise customers, what is the cache key instead? the URL ? or the request object as a whole?

Hello everyone,👋 Can someone explain me how I can have a median cpu time of 78ms while the limit is

Similar Threads

Similar Threads

Similar Threads