Also are you using hibernation or the

Also are you using hibernation or the normal ws API?

11 Replies

conor•2mo ago

Yes I'm using the hibernation API I believe.

const websocket = this.state.getWebSockets("my-tag")
            if (websocket.length === 0) {
                console.error("No websocket")
                return new Response("No websocket available", { status: 400 })
            }

const websocket = this.state.getWebSockets("my-tag")
            if (websocket.length === 0) {
                console.error("No websocket")
                return new Response("No websocket available", { status: 400 })
            }

I do a lookup against the state of websockets based on tag when replying, but I don't get that error so it seems the websocket is always there.

Browsers periodically send protocol level pings so I'd be surprised if this was the reason for premature disconnection.

That was my understanding too, though I've read mixed accounts of what people have experienced online. It'd be cheap to test an application level ping I guess. I'm going to try another browser next. Ok I've reproduced in vanilla Chrome Version 127.0.6533.100. A more nuanced workload description is that my browser client sends about 2900 messages at once, receives immediate responses from the websocket server, then sits idle receiving further messages from the server with more data. First message from client -> server at 10:45am PDT, last message from server -> client at 10:49am PDT before the socket is closed and no more messages flow in any direction.

Milan•2mo ago

Huh, and you're saying the client receives 1006? I'm on chrome connected to a DO I have, which I think is just the hibernation example we provide in our docs. I sent a websocket message at 13:19, then waited until right now (13:54), sent another message and got a reply from the DO. My close handler never ran Can you wrangler tail the DO? Do you see any errors in your dashboard?

conor•2mo ago

Huh, and you're saying the client receives 1006?

That's right.

I'm on chrome connected to a DO I have, which I think is just the hibernation example we provide in our docs.

Maybe I can try to deploy that then iteratively adjust to add the rest of the pieces in my current workflow to debug further. The websocket server I have receives a message, checks if it has a specific field, if not it queues it for further work, if so it returns it to the browser client. The queue consumer pops off messages, does work, then sends the result as a message back to the websocket server.

Can you wrangler tail the DO? Do you see any errors in your dashboard?

I can but I notice logs stop after some time, and I see more continous logging coming from the log tab in the web UI so I haven't been trusting it much.

Milan•2mo ago

if not it queues it for further work,

With an alarm()? Or something else? I wonder if the DO itself is just crashing for some reason. You can DM me your account ID I can try to take a look (but will be a bit busy with other stuff today)

conor•2mo ago

npx wrangler tail gives me: Unknown Event - Ok @ 8/10/2024, 11:10:03 AM Then I get this warning:

Tail is currently in sampling mode due to the high volume of messages. To prevent messages from being dropped consider adding filters.

Then the last message timestamp is: Unknown Event - Ok @ 8/10/2024, 11:10:04 AM There are more responses coming through the websocket still so the logs don't seem to keep up in tail I just use request_queue_binding.send to queue the message, then the consumer does work, looks up the websocket in the state (consumer is the same DO object, I guess), and sends the resulting message. It works fine until the 1006 error.

Milan•2mo ago

🤦‍♂️ sorry, I forgot there's an issue with logs coming from our websocket hibernation methods. The runtime is emitting logs correctly but the service that interprets them isn't consuming them correctly as of late. We (Durable Objects team) don't contribute to that service but have let the codeowners know and it should be fixed soon

conor•2mo ago

Ok cool that eliminates me being the issue haha Is my account ID just my email?

Milan•2mo ago

Nope, there's some actual hex ID, let me see if I can figure out how to get it wrangler whoami I think that should work?

conor•2mo ago

Ah I have an account_id in my wrangler.toml as well

Milan•2mo ago

(Don't paste here tho!!! this is still a public channel 😛 ) Send me a DM instead

conor•2mo ago

Yeah will DM you, the whoami matches the account_id in the wrangler.toml

Gaming

Programming

Also are you using hibernation or the