Cypress networking: EPIPE error (causing missing snapshots)

[bylertall]

Requesting additional support for ticket 546822. We're seeing occasional missing snapshots and finding error logging like Error: write EPIPE04:20. We expect to have 30 snapshots but we occasionally end up with 28 or 29 due to these failures. It's adding unnecessary noise/dev-burden having to approve when missing snapshots are added back to our baseline.

Package versions:

@percy/cypress v3.3.1
cypress v8.7.0

Please let us know if you need additional information. Thank you.

[bylertall]

It looks like the docs for proper configuration of parallel tests have changed quite a bit since we setup our implementation. I'm working on fixing that config now, so this support request may not be needed. I'll follow up here if we're successful in fixing this on our end.

[bylertall]

Looks like this is still an issue for us. See this run. Getting the same error message in Semaphore, "Error: write EPIPE".

[Robdel12]

Hey @bylertall! Can you tell me more about your Cypress test suite setup (plugins, post processors, etc)? That error is coming from Cypress's request backend (we wrap it here to make requests: https://github.com/percy/percy-cypress/blob/master/index.js#L17). Is it always on the same snapshot or others?

It's failing to POST the captured DOM snapshot to the local running Percy server -- not sure why.

[Robdel12]

I think the only way we'll be able to debug this is with a reproducible example. The debugging will probably be messy and deep in the networking internals of Cypress. 😳 What we know is the error is coming from Cypress's networking (/home/semaphore/.cache/Cypress/8.7.0/Cypress/resources/app/packages/server/node_modules/@cypress/request/request.js). It'll probably take some time to figure out what's different about the env between the three players here (the SDK, the tests, and the test runner) 🤨

[bylertall]

I dug further into the logs for the first run on Cypress v9 where we ended up with 27 snapshots. I found 3 [percy] Could not take DOM snapshot error logs, each followed by [percy] [object Object]. I have the verbose flag still on for percy. Any idea why I'm not getting the full error output? I'm assuming it's the same issue but would be nice to confirm.

[bylertall]

I also just noticed a different type of error on another run of cypress v8. "RequestError: Error: socket hang up"

[Robdel12]

I've seen the verbose flag not work if yarn or npm does a weird resolution and the logger package is out of sync from the rest of the packages. A nombom should correct that. In any case, PERCY_DEBUG=* will always get the Percy debug logs.

I'm not sure what's going on but there's going to have to be digging and debugging into the Cypress request package too. The POST is failing here: https://github.com/percy/percy-cypress/blob/master/index.js#L62 but it fails within the cypress internals since that's what we're using to make requests https://github.com/percy/percy-cypress/blob/master/index.js#L17

Due to the number of issues we've had with Cypress networking things we've been discussing dropping it and going back to making users register a task so we can make vanilla HTTP requests again. This won't be something we get to in the near future though and would be a major breaking change.

[bylertall]

Thanks for the notes about getting the logs going again and for the additional context around the problems you've had with cypress's networking tooling. Appreciate your time and links to source code to help us debug further.