jacogr/polkadot-js-tools metadata tool - abnormal closure error

[ghzlatarev]

• I'm submitting a ...

• [x] Bug report
• [ ] Feature request
• [ ] Support request
• [ ] Other

• What is the current behavior and expected behavior?

We have a CI workflow that is supposed to use the metadata diff tool to check whether we need to bump transaction_version of our runtime. Pretty much identical to https://github.com/paritytech/polkadot/blob/master/.github/workflows/release-21_extrinsic-ordering-check-from-two.yml Yesterday I noticed our own workflow is failing with the following error:

2023-04-07 05:17:20          API-WS: disconnected from ws://localhost:9944: 1006:: connection failed
2023-04-07 05:17:20          API-WS: disconnected from ws://localhost:9955: 1006:: connection failed
2023-04-07 05:17:23          API-WS: disconnected from ws://localhost:9944: 1006:: connection failed
2023-04-07 05:17:23          API-WS: disconnected from ws://localhost:9955: 1006:: connection failed

I tested locally on my machine by running 2 nodes and the following

docker pull jacogr/polkadot-js-tools
docker run --pull always --network host jacogr/polkadot-js-tools metadata ws://localhost:9945 ws://localhost:9944

And reproduced the error

• What is the motivation for changing the behavior?

• Please tell us about your environment:

• Version:

• Environment:

  • [ ] Node.js
  • [ ] Browser
  • [ ] Other (limited support for other environments)

• Language:

  • [x] JavaScript
  • [ ] TypeScript (include tsc --version)
  • [ ] Other

[jacogr]

The 1006 is coming from the RPC node. The docker tooling is only reporting that there has been a closure. There is nothing code-wise that can be done, the above is local setup related.

[ghzlatarev]

Do you have any suggestions what to look at, I'm a little lost? I can reproduce it on fresh ubuntu AWS instance and locally on a Mac, and I can connect to this node through the browser polkadot-js front-end locally. I would've expected that to fail as well.

[wilwade]

@ghzlatarev I am testing a solution: Use host.docker.internal instead of localhost.

Testing it out now for our own upgrade. It also means you can remove --network host

docker run --pull always jacogr/polkadot-js-tools:0.55.3 metadata ws://host.docker.internal:9946 ws://host.docker.internal:9944

Should know if it works later today.

[wilwade]

@ghzlatarev Looks like that was a bust. So host.docker.internal alone is not the answer.

[wilwade]

@ghzlatarev

Found a solution:

• Simple working test version: https://github.com/wilwade/testing-docker/blob/03eb98e22d5defb8902ade72376f3ee1f8c351d9/.github/workflows/release.yml#L135-L163
• More complex final version: https://github.com/LibertyDSNP/frequency/blob/517cb2b20feafe2dad62031ea462638b33b183b8/.github/workflows/release.yml#L529

I don't know why, but the docker is just not connecting to the host network (or not the right/same one at least).

I just created a new docker network and ran the node in the docker. It is actually nice in a few ways, but it solved the issue for me.