Astro.js SSR failing every 2nd urql query, ECONNRESET error

[djkato]

Hello, development has been a huge pain, as at some point astro started throwing errors at me even after a page was already received in the browser. Cannot read properties of undefined (reading 'map'), for which urql is the issue, as many queries just return undefined and give -104 ECONNRESET error. I have to ctrl-r a few times to get my site back.

This also affects my deployment, and it straight up crashes my storefront in a way that not even pm2 can restart the node process

I have a nanostore for my urql client, which I initialise like so:
connect.ts

export function connect(options?: { api_url?: string, token?: Token }): Client {
  const client = new Client({
    url: options?.api_url || import.meta.env.PUBLIC_GRAPHQL_URL,
    exchanges: [cacheExchange, fetchExchange, retryExchange({ initialDelayMs: 100, maxNumberAttempts: 5 })],
    requestPolicy: "cache-and-network",
    fetchOptions: () => {
      if (options?.token) {
        return {
          headers: { authorization: options.token ? `Bearer ${options.token}` : "" }
        }
      }
      else return {}
    }
  })
  return client
}

server-client.ts

import { connect } from "#scripts/connect";
export let _gql_client = map(connect()

Then use like so in my SSR astro pages:
pages/index.astro

---
import { _gql_client } from "#stores/server-client"
let { product: product_slug } = Astro.params

const product_result = await _gql_client
    .get()
    .query<ProductDetailsQuery, ProductDetailsQueryVariables>(query_product_details, { slug: product_slug!, channel: default_channel })

const product = product_result.data?.product

const name = parse_pretty_text(product?.name)
---

<h1 set:html={name} />
.
.
.

doing console.log on the result of product_result

{
  operation: {
    key: -5439908619,
    query: { kind: 'Document', definitions: [Array], loc: [Object] },
    variables: { first: 1, sort_direction: 'DESC', sort_field: 'DATE' },
    extensions: undefined,
    kind: 'query',
    context: {
      _instance: undefined,
      url: 'https://redacted',
      fetchSubscriptions: undefined,
      fetchOptions: [Function: fetchOptions],
      fetch: undefined,
      preferGetMethod: undefined,
      requestPolicy: 'cache-and-network',
      suspense: false,
      meta: [Object]
    }
  },
  data: {
    products: {
      totalCount: 11278,
      pageInfo: [Object],
      edges: [Array],
      __typename: 'ProductCountableConnection'
    }
  },
  error: undefined,
  extensions: undefined,
  hasNext: false,
  stale: false
}
undefined
CombinedError: [Network] fetch failed
    at makeErrorResult (file:node_modules/@urql/core/dist/urql-core-
chunk.mjs:335:10)
    at fetchOperation (file:node_modules/@urql/core/dist/urql-core-c
hunk.mjs:563:11)
    at process.processTicksAndRejections (node:internal/process/task_queues:95:5)
    at async file:node_modules/wonka/dist/wonka.mjs:877:20 {
  graphQLErrors: [],
  networkError: TypeError: fetch failed
      at Object.fetch (node:internal/deps/undici/undici:11372:11)
      at process.processTicksAndRejections (node:internal/process/task_queues:95:5)
      at async fetchOperation (file:/node_modules/@urql/core/dist/urq
l-core-chunk.mjs:466:18)
      at async file:node_modules/wonka/dist/wonka.mjs:877:20 {
    cause: Error: read ECONNRESET
        at TLSWrap.onStreamRead (node:internal/stream_base_commons:217:20) {
      errno: -104,
      code: 'ECONNRESET',
      syscall: 'read'
    }
  },
  response: undefined
}

I got that one without error once, then refreshed and got two errors at the same time. Sometimes it does product and after that fail, sometimes one component on a page fails and another queries fine...

[JoviDeCroock]

The retry exchange after fetch won't work. I don't think urql can cause ECONRESET that's a failure communicating from point x to point y 😅

I wonder whether the retry after fetch is causing some weird loop but I don't think so as fetch is terminating

[djkato]

Could Astro/vite refresh somehow cause that? Like mid fetching it force reloads or something. Or is this an issue higher than urql, like at node networking or something? I don't get internet interrupts, and my remote api doesn't have any timeouts or limits I think.