This is network team fault. 😫 You already saw my "MVP" theory: Free: worker script is pushed to two, or three core location, and when is run for the first time on a specific colo it is copied to that colo, after that run latency becomes low. the worker can be evicted as many times needed for whatever reason, but the script is there. Bundled: worker script is pushed globally, therefore latency is low. Once again workers are spawned and evicted as per colo needs.

Yeah. I feel like it's going out of India adding more latency. The thing I am not sure about in that post is: "to test without Workers is he disabling Cloudflare altogether?". If he is then it's the POP's location at fault, especially mentioning India.

We are back at the start, using Durable Objects, and having inefficient routing for the client, creates inertia.

Yeah... I am not sure of the actual latency due to Durable Objects, but he seems to be using KV only, should be relatively fast though.

India is always the same debate, also a massive market for CF. If KV Core stores are in Europe and US, and on the two locations that CF has capacity. Mumbai will return a KV in 110ms 🇪🇺 and 180ms 🇺🇸, and from Bangalore 130ms 🇪🇺 and 220ms 🇺🇸. Now, where the workers are spawned based on user location: Mumbai Free SIN Pro SIN Business SIN Enterprise BOM Bangalore Free BOM Pro LHR Business MAA Enterprise SIN

Mumbai does make sense, it's expensive, so it's Enterprise-only. But Bangalore? BOM for Free (which isn't active for Free in Mumbai) and SIN for Enterprise?

this is what the probes are showing for the last hour.

Oh, you have probes! I trust you, just doesn't make much sense... Cloudflare's team is the only one that can shed a light on that.

It is a Network Team issue, but last I reported to Support, I got the famous canned report, so I will not waste energy reporting it again. The probes also monitor on Fastly, Akamai, Stackpath.

Maybe @User or someone here can work it internally. That probe data would be interesting to see.

Define managed. I would be using MongoDB Atlas where they host for me.

Answered with MongoDB Atlas. I never used MongoDB Atlas, but if you managed your own instance of MongoDB, you could created a Node.js REST Proxy then with Cloudflare Argo Tunnel connect your worker to your instance.

Atlas is doing this for you well it's a work in progress between them and cloudflare to support workers

I'm hoping more databases add an automatic HTTP API in the future

Custom TCP Protocols are harder to work with and it'll be a long time before Node Isolates support it

I used MongoDB until 2015 then there was a shift to more $$$ so I migrated to Redis and CouchDB what I had in MongoDB. CouchDB could give agility to you project.

Either mongo or postgres is desirable for my projects

I'm feeling optimistic about the future Mongo Atlas support being added into CF Workers although meanwhile it'd frustrating I have to write a separate proxy api that runs elsewhere

You lose the scaling benefits of workers once you have to manage scaling of an external API on your own

good morning 👋

hi!!

Going to have dinner, bye ahah 😛

curl uploading scripts with

--upload-file

works, but when I use

-F "script=@file

I get

"code": 10021,
      "message": "could not read content for part metadata"

try creating a metadata.json file with

{
  "body_part": "script"
}

then add this parameter to curl along with your script -F "script=@file

-F "metadata=@metadata.json"

that worked. thanks!

i think i'm finally code complete on this, phew https://deals.bytesized.xyz/

very big monorepo with workers, sanity.io, nuxt.js, all kinds of fun stuff https://github.com/signalnerve/deals

need to go through and actually document everything, that's the project for the rest of the week before two black friday deals go live 👍

if anyone ever wants help w/ stripe + workers stuff though.... 👀 👀 👀

That looks epic @User! What kind of Stripe and Workers stuff?