I’m frequently but not always getting a `scheduler...
# development
h
I’m frequently but not always getting a
scheduler.ExecutionError
when calling
./v2 --changed-since=master fmt
that only includes the very long
Computing Select
clause and then
<None>
Has anyone else seen this? I started noticing it yesterday.
e
I hit it every time and
--changed-since
is not required.
Copy code
$ git bisect bad
6db9ab4becf5b520c12db297b10c9e6e78aa994e is the first bad commit
commit 6db9ab4becf5b520c12db297b10c9e6e78aa994e
Author: Henry Fuller <<mailto:hfuller@twitter.com|hfuller@twitter.com>>
Date:   Thu Mar 26 12:01:39 2020 -0700

    Add notify fs watcher to engine. (#9318)
...
@early-needle-54791 fyi. I'll file an issue and see if there is not a workaround so we don't need a full revert.
e
Yeah I’m gonna work on a feature gate for it today so we hopefully dont have to revert
If you are more familiar with why the fmt goal might run multiple times when invalidation happens during the goal I would be interested to learn more about that so we can fix this in the long term.
e
I'm not sure multiple times has anything to do with it unless you are. It does modify files in the build root using the engine though.
I added a note about that in the issue.
OK - I handed off the issue to you. Thanks Henry.
e
I’m not sure multiple times has anything to do with it unless you are.
As eric pointed out, and is visible in your bug report, the black tool is being run multiple times, when it should be run once. To me this indicates that the fs watcher is invalidating nodes either while they are running or right after, causing them to run again presumably. I’m not sure exactly where the code that determines why / when to run a node is but I plan to pursue that. If the build graph must have every node in a completed state for the build to succeed, that would be where the problem is.
👍 1