Does anyone have a reliable way to track timings inside of a worker? I am aware that `Date.now` and

Does anyone have a reliable way to track timings inside of a worker? I am aware that

Date.now

Date.now

Date.now

Date.now and

performance.now

performance.now

performance.now

performance.now are only ever updated when I/O occurs, but how do you guys go about debugging internal latency issues?

juriOP•12/19/23, 6:52 PM

For example, if my worker is apparently doing some heavy internal work but you can't find out where, how would you go about that?

Jjuri For example, if my worker is apparently doing some heavy internal work but you c...

DaniFoldi•12/19/23, 6:57 PM

I'm not entirely sure because I've never tested this, but during local development if you open an inspector, do you not see accurate timing data there?

Jjuri Does anyone have a reliable way to track timings inside of a worker? I am aware ...

kian•12/19/23, 6:59 PM

Internal as in how does Cloudflare do it, or the community?

DDaniFoldi I'm not entirely sure because I've never tested this, but during local developme...

juriOP•12/19/23, 7:00 PM

Would you consider these numbers as "true" then?

juriOP•12/19/23, 7:01 PM

I think my issue is more about an operation inside of the worker taking longer than expected (parsing data, for example), which is not I/O so

Date.now

Date.now

is never updated.

Kkian Internal as in how does Cloudflare do it, or the community?

juriOP•12/19/23, 7:02 PM

Oh, internal as in code inside the worker taking a long time to execute — but at the same time not being able to identify what piece of code is responsible because comparing

Date.now

Date.now

inside the worker is not accurate.

kian•12/19/23, 7:04 PM

There isn't really a whole lot you can do. Timing with local development should be accurate, but not in remote development or when published.

juriOP•12/19/23, 7:06 PM

This will be really hard then, I assume. I unfortunately cannot use the worker locally as it requires being connected to a Dispatch Namespace and D1

DaniFoldi•12/19/23, 7:07 PM

yeah dispatch namespaces don't yet have support in local development, still in an early PR phase meowsad

Unsmart•12/19/23, 7:07 PM

I mean you can always pull out a test where you dont use dispatch namespaces and run it locally to test

DaniFoldi•12/19/23, 7:08 PM

of course you can perform an isolated test to check the timings there, but it's not the best developer experience

kian•12/19/23, 7:11 PM

Yeahhh. Measuring anything latency or CPU/RAM related is a dead end on Workers which sucks as eventually the only billing model is based on CPU runtime that you can only measure on a "whole Worker" basis using the analytics.

kian•12/19/23, 7:18 PM

Well, you can get accurate per-request CPU time if you feel like paying for Enterprise and using LogPush :^)

UUnsmart I mean you can always pull out a test where you dont use dispatch namespaces and...

juriOP•12/19/23, 7:21 PM

Unfortunately the whole process requires access to Dispatch Namespaces :/

Kkian Well, you can get accurate per-request CPU time if you feel like paying for Ente...

juriOP•12/19/23, 7:22 PM

We should most definitely have access to that, I have not used it before, I'll try out another solution I was proposed and will then check LogPush out!

Unsmart•12/19/23, 7:22 PM

a fetch to a dispatch namespace uses 0ms cpu time so you can figure out where your cpu usage is by taking out specific code blocks and trying to figure out what uses the most cpu there

Kkian Well, you can get accurate per-request CPU time if you feel like paying for Ente...

juriOP•12/19/23, 7:34 PM

How would you suggest to go about this? Send individual logs to LogPush at the desired spots? The idea is that LogPush then reveals the actual wall-clock times when the logs were pushed, correct?

kian•12/19/23, 7:35 PM

LogPush creates a log per request, you can't send anything to it

kian•12/19/23, 7:36 PM

That log can include WorkerCPUTimeWorkerCPUTime and WorkerWallTimeUsWorkerWallTimeUs

juriOP•12/19/23, 7:38 PM

Oh, I see, I also just read that when setting it up. But this is only the total of the request and not individual "checkpoints" I can track during code execution?

kian•12/19/23, 7:39 PM

Correct, total of the request.

juriOP•12/19/23, 8:41 PM

Is the TimestampMsTimestampMs for every item inside the LogsLogs array wall-clock time or is it based on the same time

performance.now

performance.now

Date.now

Date.now

use?

kian•12/19/23, 8:43 PM

Same as performance/date

kian•12/19/23, 8:43 PM

There is nothing that provides accurate time for individual events in Workers that is available to non-employees.

juriOP•12/19/23, 10:59 PM

Another idea: could we use Workers Analytics Engine for this? When you call writeDataPointwriteDataPoint, I would expect it to insert data with a correct timestamp.

juriOP•12/19/23, 11:00 PM

There must be some way to gather metrics inside of a worker reliably.

Jjuri There must be some way to gather metrics inside of a worker reliably.

kian•12/19/23, 11:06 PM

It's intentionally not supposed to be possible.

kian•12/19/23, 11:06 PM

https://developers.cloudflare.com/workers/learning/security-model/#step-1-disallow-timers-and-multi-threading

Fizzy Kombucha•12/19/23, 11:43 PM

i am trying to build a sveltekit SSR app that will be deployed to pages. in my app i'm using mongodb. I can't build the project because the workers runtime apparently expects node js apis to be imported with the syntax 'node:crypto'. i have no control over how the mongodb client does its imports. i have no idea how to fix this or proceed.
related issue: https://community.cloudflare.com/t/cloudflare-pages-not-compatible-with-node-modules/566178
https://developers.cloudflare.com/workers/runtime-apis/nodejs

i am building my svelte kit app with 'CF_PAGES=1 vite build'.

Cloudflare Community

Cloudflare Pages not compatible with Node modules

I am having an error when deploying a SvelteKit app that uses some Node.js packages in the server. I get this error in the logs: The package “path” wasn’t found on the file system but is built into node. Are you trying to bundle for node? You can use “platform: ‘node’” to do that, which will remove this error. So, where do I have to use “plat...

Node.js compatibility · Cloudflare Workers docs

Implemented Node.js runtime APIs and enablement instructions for your Worker project.

Fizzy Kombucha•12/19/23, 11:43 PM

getting a bunch of errors like:

> Using @sveltejs/adapter-cloudflare
Starting mongo...
✘ [ERROR] Could not resolve "dns"

    node_modules/.pnpm/mongodb@6.3.0/node_modules/mongodb/lib/sdam/srv_polling.js:4:20:
      4 │ const dns = require("dns");
        ╵                     ~~~~~

  The package "dns" wasn't found on the file system but is built into node. Are you trying to bundle
  for node? You can use "platform: 'node'" to do that, which will remove this error.

✘ [ERROR] Could not resolve "stream"

> Using @sveltejs/adapter-cloudflare
Starting mongo...
✘ [ERROR] Could not resolve "dns"

    node_modules/.pnpm/mongodb@6.3.0/node_modules/mongodb/lib/sdam/srv_polling.js:4:20:
      4 │ const dns = require("dns");
        ╵                     ~~~~~

  The package "dns" wasn't found on the file system but is built into node. Are you trying to bundle
  for node? You can use "platform: 'node'" to do that, which will remove this error.

✘ [ERROR] Could not resolve "stream"

Fizzy Kombucha•12/19/23, 11:44 PM

i've been scouring google and can't seem to find a solution. was hoping someone here would know

Fizzy Kombucha•12/19/23, 11:50 PM

welp it looks like i found my answer. mongodb simply won't work on workers

https://www.mongodb.com/community/forums/t/cloudflare-workers-integration-is-now-possible/226708/11

MongoDB Developer Community Forums

Cloudflare Workers integration is now possible

Hi @Alex_Bjorlig, We are working with Cloudflare on this still, however at the moment we are blocked as Atlas requires TLS for all connections, however Cloudflare Workers only supports mTLS at the moment (and TLS cannot be disabled in Altas).

kian•12/20/23, 1:49 AM

nodejs_compat

nodejs_compat

relies on them using the

node:

node:

prefix when importing &

node_compat

node_compat

are just best-effort polyfills.

kian•12/20/23, 1:49 AM

Puppeteer won't work in Workers unless you're using Cloudflare's fork with #browser-rendering-api-beta

Fizzy Kombucha•12/20/23, 1:51 AM

yea it really sucks. i built a small app with mongodb. was running it all locally and everything works fine. then i go to actually build and deploy it... now i have to swap out mongo for another db

kian•12/20/23, 1:55 AM

Workers can't spawn subprocesses or have a filesystem, so I don't think Puppeteer would ever work anyways.

kian•12/20/23, 1:55 AM

#browser-rendering-api-beta is a fork that talks to a remote browser hosted elsewhere on CF's infra

kian•12/20/23, 1:58 AM

You'd use CF's Puppeteer fork and pass it the binding, which then controls the remote browser.

kian•12/20/23, 1:58 AM

https://developers.cloudflare.com/browser-rendering/get-started/screenshots/#5-code

kian•12/20/23, 2:05 AM

What are you trying to do?

SSDD•12/20/23, 10:16 AM

can we turn off caching in cf workers?

Does anyone have a reliable way to track timings inside of a worker? I am aware that `Date.now` and

Similar Threads

Similar Threads

Similar Threads