web shell: service unavailable

[obnoxxx]

In the Rook project https://github.com/rook/rook, we use this tmate action for debugging CI issues with PRs.

It was recently noticed that this does not seem to work reliably any more:

Most of the times, the ssh terminal access works, while the web shell URL yields a 503 - service unavailable error.

corresponding issue:

rook/rook#16131

A test PR to trigger the issue has also been created: rook/rook#16245

Any help/hints for troubleshooting this problem will be highly appreciated!

[Vampire]

It seems we need to search for a replacement. :-(
tmate which this action is based on is dead, unless the author of this action (or anyone else) will effectively step-up and provide replacement servers. :-(
tmate-io/tmate#322 (comment)
For me not even SSH access works.

[obnoxxx]

@Vampire wrote:

It seems we need to search for a replacement. :-(

too bad. @parth-gr mentioned upterm here: rook/rook#16131 (comment)

It also has an action.

tmate which this action is based on is dead, unless the author of this action (or anyone else) will effectively step-up and provide replacement servers. :-( tmate-io/tmate#322 (comment) For me not even SSH access works.

Thanks for the pointer, I wasn't aware. Its a pity!

For completeness, In my direct tests with tmate (not through the action), ssh still works but web shell yields 503 errors. So maybe the servers are not entirely down yet...

[Vampire]

Ignore what I said, SSH access works again for me, it probably just was a hiccup.

mentioned upterm here
It also has an action.

That's (for me) not a sufficient replacement currently.

• the original action repo is archived, but there is a new fork that is maintained now (just for those reading here)
• but the fork like the original does not support Windows which was the original reason I switched to this action

While I was hit by the SSH hiccup I also evaluated alternatives with the main requirement that they are working on all three major OS and are currently working.
So just in case we finally do need to use an alternative, the best options I currently found were:

• indeed the action-upterm if Windows is not needed or maybe gets added
• debug-via-ssh but this has some restrictions, because you need an ngrok account for it and now they also require you to give a credit card to identify as non-abuser without it being charged, but some people might not have a credit card or don't want to provide it. And afair you are also restricted to one SSH session at a time in the free ngrok account which can also become cumbersome if you debug a matrix build, especially as there by default the session stays alive when you continue and there was no easy way to kill the ngrok session an on macOS could not even continue the job. There are alternatives using ngrok but this was the best I found from testing that also worked halfway.
• most probably the best replacement unless maybe action-upterm gets Windows support was actiondebugger-by-warpbuild, as warpbuild is a tmate fork iirc and their action is a fork of this action. Maybe it could even work to specify the warpbuild servers for this action if it stayed compatible, I did not try that.