[CT-161] RPC server goes into infinitely compiling state when cancelling dbt deps

[davidharting]

Background

In the Cloud IDE, when you cancel a dbt deps command, the rpc server enters a compiling state and never comes out in my testing (after 5+ minutes). This appears to be a bug with dbt core or the RPC server, rather than cloud getting out of sync with the rpc server. The cloud IDe sends a “kill” command when the cancel button is clicked. As the IDE polls on status, it gets “compiling” back every time. So far, I have only been able to get this to happen with deps and no other commands.

Reproduction steps

In dbt Cloud:

1. Start an IDE session
2. Make sure you are in a Ready state
3. Use the command bar to run dbt deps
4. Hit the Cancel button while it is running
5. Notice
6. That the deps command successfully cancels
7. That the server is now in a compiling state and stays in that state

You can also watch me repro this in this Loom.

Expected behavior

I expect the server to go to compiling momentarily after cancelling, but to then finish compiling and enter either an error or ready state.

[jtcohen6]

I was able to reproduce this with dbt-rpc locally. My suspicion is:

• deps is an instance of BlocksManifestTasks (i.e. it has both flags RequiresConfigReloadBefore + RequiresManifestReloadAfter)
• For manifest-blocking tasks, compilation cannot finish (= return to "ready" status) until the task has completed. See the logic here
• When the task is killed (akin to keyboard interrupt), it never completes (either success or error), and so it never exits the set_parse_state_with context manager

(I'm sure I'm slightly off for a few of the finer details)

You can also watch me repro in this Loom :)

[davidharting]

@jtcohen6 Lol, the matching sweatshirt Loom 😆 Thanks so much for the reproduction and analysis! Learned a lot about internals from watching. I need to get my local dbt set up restored so that I can provide local repro for this sort of thing!

[iknox-fa]

We have a few issues regarding dbt not handling inturrupts well.. Spike has been made to investigate further:
#83

[github-actions]

This issue has been marked as Stale because it has been open for 180 days with no activity. If you would like the issue to remain open, please remove the stale label or comment on the issue, or it will be closed in 7 days.