Because the library requires that manifest to fetch static content from the KV

@User mentioned a possible solution in the GitHub issue I linked

> The necessary work here is to add __STATIC_CONTENT_MANIFEST as a text module to the module manifest passed into ModuleAssets -- this is the equivalent of a text blob in the service worker format. It'll be exposed as a module that can be imported, not under env however.

Once that is done my PR to kv-asset-handler can be adjusted

I’ll need to write some new tests for these scenarios

Small arch question: If we have a shared document platform like "google docs" (not really, just a example); Is supposed each document be a own DO ? To serialize the requests to each document etc? Thats will not impose too much memory pressure? If i have 1000 concurrently documents being edited at one particular time.

Some warning in advance (without guaranties) can help to save few things on memory. Some objects are small enough.

any ideas when the billing for DOs and websockets will be adjusted?

the blog post which announced them mentioned a fix being in the works but its been nearly 3 months since than and I haven't heard anything in this regard

Are there any examples of workers talking to each other via websockets? Or any gotchas related to that? Hoping to try something out around that soon

I dont think you can do that because theres no way to establish a route between workrs

I wonder if it could connect like an external client then via the public URL of the other object?

Then the endpoint handles routing it like any other request

Is there a limit on the number of

keys

to

delete

on durable object storage?

if it doesnt say I would reckon not, but im not sure

It might fail if you try it with a bunch of keys though idk

I have a quick question regarding websockets. Is there a way to have a single websocket connection and multiple associated durable objects?

any way of doing this https://github.com/s-yadav/FlakeId without using a npm package in a worker?

Ok, so answering my own question:

delete(keys)

currently seems to have a limit of 128 keys, same as

get(keys)

, just got

RangeError: Maximum number of keys is 128.

- maybe add to the docs?

Just copy the code, it's extremely basic with no deps - you'll need to modify the while loop that tries to wait for one millisecond anyway to avoid an infinite loop, since Date.now() does not advance inside a worker execution (except for io)

that’s interesting, i didn’t know date.now() doesn’t advance (it basically returns the time the worker started?). What do you mean by except for io? @john.spurlock

To mitigate spectre-style attacks. You can observe

Date.now()

changing after awaiting a fetch or a cache op, for example.

oh, interesting. i’ve seen browsers reducing precision before but this is slightly more extreme, seems like? thanks john!

its more extreme because browser spectre attacks have less security implications than someone taking over a machine that has potentially thousands of tenants. Some of which may be using cloudflare for financial transactions or housing other sensitive data

besides, browser sessions are typically short lived which would make a spectre attack quite difficult

whereas workers could potentially be run for hours at a time if they have websocket connections

Is there a way to remove the execution limit? It seems as we add more connections to the DO we get request cuts at random times

Is there a sub for Stream Connect?

thanks for this, hopefully it's fixed properly soon!