webSocket auto reconnect doesn't work if socket is closed

[yivlad]

Check existing issues

• [X] I checked there isn't already an issue for the bug I encountered.

Viem Version

2.13.1

Current Behavior

We use webSocket transport in our frontend app. After some period of inactivity browsers automatically close the web socket connections. When a user leaves the tab open and gets back after a while, webSocket transport is unable to recover from the socket CLOSED state and requests keep failing.

Expected Behavior

webSocket transport creates new connection in place of the closed one.

Steps To Reproduce

Minimal reproducible example using only viem:

import { createPublicClient, parseAbi, webSocket } from "viem";

const client = createPublicClient({
  transport: webSocket(
    "wss://mainnet.infura.io/ws/v3/84842078b09946638c03157f83405213"
  ),
});

const blockNumber = await client.readContract({
  abi: parseAbi(["function getBlockNumber() view returns (uint256)"]),
  address: "0xca11bde05977b3631167028862be2a173976ca11",
  functionName: "getBlockNumber",
});

console.log({
  blockNumber
});

(await client.transport.getRpcClient()).socket.close(); // simulate closing by browser

await client.readContract({
  abi: parseAbi(["function getBlockNumber() view returns (uint256)"]),
  address: "0xca11bde05977b3631167028862be2a173976ca11",
  functionName: "getBlockNumber",
});

In the provided code the second request fails because the web socket connection is closed.

Link to Minimal Reproducible Example

No response

Anything else?

No response

[badgerdf]

Same issue on backend side (nest.js framework). Moreover, viem doesn't notify about socket disconnect/reconnect by logs, just looks like that websocket is just stuck without any reason.

Here is client creation code.

createPublicClient({
      chain: Viem.utils.extractChain(CONFIG.CHAIN_ID),
      transport: webSocket(CONFIG.RPC.WS, {
        reconnect: true,
        retryCount: Infinity,
        retryDelay: 1000,
        timeout: 1000
      }),
    })

[0x33dm]

I have the same problem, moreover i didn't find a way of being "notified" ( like an eventListener ) about the connection status.

[flux0uz]

Same problem for us, impossible to index via websocket, no disconnection errors and no automatic reconnection. This is really annoying, because we're going to have to listen via http and explode our alchemy quotas! Is a fix in the works?

[0x33dm]

Same problem for us, impossible to index via websocket, no disconnection errors and no automatic reconnection. This is really annoying, because we're going to have to listen via http and explode our alchemy quotas! Do you know if a fix is in the works?

You can increase the number of reconnect attempts (https://viem.sh/docs/clients/transports/websocket.html), but I don't see a way to be notified of the connection status..

the only way i can think of indexing all blocks with viem would be to go chunk by chunk using http requests.

[izayl]

in my case, it because the backend nginx timeout setting

when I subscribe a topic with indexed topics, it may have long time without response, if backend setting a max timeout, the socket will be closed, return code 1006('Abnormal Closure')

there are two way to resolve this issue:

1. verify all the CloseEvent code, apply onError logic first to reuse current reconnect logic. Yes, it will not trigger onerror like above timeout case

2. implement keepalive

I just implement the keep-alive feature: #2516

it will keep send ping to ws server every 30s, and dev can set any value that less than your backend timeout setting, to make sure ws keep-alive

happy to receive some feedbacks

[0x33dm]

happy to receive some feedbacks

One thing I miss/couldn't find is a way to subscribe to connection events, like "connected," "disconnected," or "error," in order to make the application more self-aware.

During my initial tests with WS transport, I found that WS would disconnect and fail silently, and my application would stop.

i think either i'm missing something from the documentation or there is room for improvement.

[izayl]

happy to receive some feedbacks

One thing I miss/couldn't find is a way to subscribe to connection events, like "connected," "disconnected," or "error," in order to make the application more self-aware.

During my initial tests with WS transport, I found that WS would disconnect and fail silently, and my application would stop.

i think either i'm missing something from the documentation or there is room for improvement.

I think you can get the socket first, and subscribe by yourself for advanced usage

const WS_URL = ''
const webSocketClient = createPublicClient({
  chain: mainnet,
  transport: webSocket(WS_URL)
})

const socketClient = await getWebSocketRpcClient(WS_URL)
const socket = socketClient.socket

[0x33dm]

getWebSocketRpcClient

interesting i didn't find docs for this function on viem website, but my VS CODE finds it.

thank you.

[jxom]

Fixed via https://github.com/wevm/viem/commit/44281e85e28f893df6e722ebadfe74b2a9bf93e2

[0x33dm]

Fixed via 44281e8

^ amazing

---

Would be possible to also add documentation on how to listen for connection status on the websocket so services that rely on monitoring each block can recover quickly?

From @izayl i got the hint to look for getWebSocketRpcClient but i havent' played with it yet.

thank you

[flux0uz]

Hey @jxom

It seems that the problem persists, we updated our staging environment yesterday and this morning the events are not caught... So the websocket connection must have closed... No errors on the watchContractEvent onError callback.

[izayl]

@flux0uz u can try with following code to verify the ws close reason first, see what's the code of close event

replace the ws_rpc_url to yourselfs

if (publicClient.transport.type === 'webSocket') {
  const client = await getWebSocketRpcClient(ws_rpc_url!)
  const socket = client.socket
  socket.addEventListener('close', (e) => {
    console.error('websocket closed', e)
  })
}

[flux0uz]

Thanks @izayl, no errors were detected on the socket "close" event. I've just realised that the problem is with the fallback([]). By removing the fallback on the public provider used to listen to events, everything works normally!

@jxom The fallback option for websocket transport is breaking the auto-reconnect and not throwing any errors.

[flux0uz]

Back to this issue! Another test this morning, listening to the sockets close (with @izayl code). No close log, and we don't receive any events. PublicProvider with a single websocket connection via Alchemy and no fallback.

[izayl]

Back to this issue! Another test this morning, listening to the sockets close (with @izayl code). No close log, and we don't receive any events. PublicProvider with a single websocket connection via Alchemy and no fallback.

double check your ABI, maybe has wrong order or missed indexed

[flux0uz]

Nop works correctly with http transport, but uses a ton of our Alchemy quota! What's more, it works perfectly for a while just after deployment - we wait about 1 day and then nothing is caught!

[flux0uz]

I'm now listening to message on the dev environment, the first one is OK then I keep getting : [Symbol(kData)]: '{"id":null,"jsonrpc":"2.0","error":{"code":-32700,"message":"Parse error"}}'

[justefg]

i confirm. it doesn't work. please run a minimum reproducible for at least 2 hours before merging.

[justefg]

You might want to take a look at alchemy's ws implementation. tested it and its rock solid with no connection breaking. Looking at your fix why do use ping if it's not supported by an json rpc api? I suggest using net_version or chainId like alchemy does https://github.com/alchemyplatform/alchemy-sdk-js/blob/master/src/api/alchemy-websocket-provider.ts#L746-L757

rpc endpoints: https://docs.alchemy.com/reference/net-version https://www.quicknode.com/docs/ethereum/net_version