bug(trpc/next): Regression - Subscriptions do not resume if the server closes the connection unexpectedly

[avarayr]

Provide environment information

System:
OS: macOS 13.5.2
CPU: (8) x64 Apple M1 Pro
Memory: 12.33 MB / 16.00 GB
Shell: 5.9 - /bin/zsh
Binaries:
Node: 18.16.0 - ~/.nvm/versions/node/v18.16.0/bin/node
Yarn: 1.22.19 - ~/.nvm/versions/node/v18.16.0/bin/yarn
npm: 10.1.0 - ~/.nvm/versions/node/v18.16.0/bin/npm
pnpm: 8.7.6 - ~/.nvm/versions/node/v18.16.0/bin/pnpm
Watchman: 2023.08.14.00 - /opt/homebrew/bin/watchman
Browsers:
Chrome: 117.0.5938.88
Safari: 16.6
npmPackages:
@tanstack/react-query: 5.0.0
@trpc/client: next
@trpc/next: next
@trpc/react-query: next
@trpc/server: next
next: ^13.4.8 => 13.5.1
react: ^18.2.0 => 18.2.0
typescript: ^5.1.3 => 5.2.2

Describe the bug

tRPC subscriptions do not automatically resume if the server unexpectedly closes the connection.
This can happen because of various reasons

• WAF such as Cloudflare will close websocket connections after 100 seconds
• NGINX will do the same if proxy_read_timeout is set

Looks like there's a regression in the #4959

Link to reproduction

https://github.com/avarayr/trpc-ws-subs-regression

To reproduce

1. Clone reproduction repository
2. npm install
3. npm run dev
4. Go to https://127.0.0.1:3000
5. Observe the subscription doesn't restart if the server closes the socket.

Additional information

No response

👨‍👧‍👦 Contributing

• 🙋‍♂️ Yes, I'd be down to file a PR fixing this bug!

[clonecorp]

Not sure if this has been implemented yet, but I tried "resuming" manually a closed websocket connection by either calling back the .subscribe() endpoint and also recreating a client with new wsLink() parameters but it didn't work. Is there a solution for this?

[avarayr]

Not sure if this has been implemented yet, but I tried "resuming" manually a closed websocket connection by either calling back the .subscribe() endpoint and also recreating a client with new wsLink() parameters but it didn't work. Is there a solution for this?

I believe there should never be a time when there's an active subscription and the websocket is closed. Assuming you're on the stable (v10) branch, connection should always stay open. Although if server tRPC throws an error, subscription will be stopped. subscriptions will attempt to auto-resume if an error is thrown

If you're on v11 branch, there's a bug that will prevent the client from reconnecting if subscription is interrupted by the server closing the websocket for any reason or a network issue. This issue is meant to address the bug.

[clonecorp]

Not sure if this has been implemented yet, but I tried "resuming" manually a closed websocket connection by either calling back the .subscribe() endpoint and also recreating a client with new wsLink() parameters but it didn't work. Is there a solution for this?

I believe there should never be a time when there's an active subscription and the websocket is closed. Assuming you're on the stable (v10) branch, connection should always stay open. Although if server tRPC throws an error, subscription will be stopped.

If you're on v11 branch, there's a bug that will prevent the client from reconnecting if subscription is interrupted by the server closing the websocket for any reason or a network issue. This issue is meant to address the bug.

I am on the v10 branch. But if a server crashes and restarts, how do you handle it then? When my WS closes, the client gets the onClose() event present in the createWSClient() method tRPC provides. I should technically, after the server restarts properly, be able to "reconnect" to the newly created WS server instance. But when recreating a client for the new server with createWSClient() doesn't work and I can not "open" a new WS connection after that.

Maybe I am doing something wrong, but since I track every on event, I definitely see the onOpen event doesn't trigger on the second try (after the server restarts and the new connection is attempted)

[avarayr]

I am on the v10 branch. But if a server crashes and restarts, how do you handle it then? When my WS closes, the client gets the onClose() event present in the createWSClient() method tRPC provides. I should technically, after the server restarts properly, be able to "reconnect" to the newly created WS server instance. But when recreating a client for the new server with createWSClient() doesn't work and I can not "open" a new WS connection after that.

Maybe I am doing something wrong, but since I track every on event, I definitely see the onOpen event doesn't trigger on the second try (after the server restarts and the new connection is attempted)

@clonecorp It sounds like you've made a custom way of reconnecting to the server, which might be redundant.

v10 will auto-reconnect and auto-resume subscriptions if the websocket connection is interrupted for any reason (close, error)

[github-actions]

This issue has been locked because we are very unlikely to see comments on closed issues. If you are running into a similar issue, please create a new issue. Thank you.