/benchmark command seems not to work

[Dandandan]

Describe the bug

I tried to run bencmark for #9593 (comment)

but this resulted in a failure:
https://github.com/apache/arrow-datafusion/actions/runs/8294100431/job/22698438691

the input device is not a TTY

To Reproduce

No response

Expected behavior

No response

Additional context

No response

[Dandandan]

FYI @gruuya

[gruuya]

Thanks @Dandandan; I've encountered that problem originally and (I thought I) resolved it by removing -ti from the tpch_dbgen docker command, after which the thing worked on our fork.

Will investigate more.

[gruuya]

It's weird that the exact same workflow works on our fork, but not here; seems to suggest a difference in some settings?
Fwiw, I inspected the setup and it seems to be using the identical os/runner/image etc versions.

Either way I opened #9626 as a potential workaround, except I can't really guarantee it works now (the command fails becuase it's still using the version from the main in the action).

[Dandandan]

Thanks @gruuya that worked 🎉

The next failure is a missing file
#9593 (comment)

[gruuya]

Hmm, that's another weird one, where GHA obviously has the adequate env var but somehow the script invocation doesn't.

My best guess is that this is because the immediately preceding step (where the env vars were generated) was run in a different shell now, and so GHA somehow hasn't propagated the env vars yet to the first step (but it does to the subsequent ones).

Anyway, I've opened another potential mitigation here: #9632

Thanks for the patience!

[Dandandan]

That didn't resolve it yet :)
#9593 (comment)

[gruuya]

Ok, third time's the charm: #9636 🤞🏻

[Dandandan]

Thanks @gruuya 🥳

[gruuya]

I'm glad to see it working now, though I'm not that happy with the finicky nature of the development process, courtesy of GHA idiosyncrasies.

I've noted down a potential improvement to the trial-and-error dev approach here: #9638