rose_bunch: on failure, print bunch err to stderr #2126
Conversation
|
Great, I did wonder! |
benfitzpatrick
force-pushed
the
rose_bunch_print_err_on_failure
branch
from
1653574
to
28a9b20
Compare
benfitzpatrick
force-pushed
the
rose_bunch_print_err_on_failure
branch
from
28a9b20
to
0e3f0c4
Compare
|
@benfitzpatrick - is this going to result in the entire stderr contents from the bunch-.err file getting dumped in the stderr file? |
|
Yep, just for the ones that failed |
Not massively ideal then as the sort of things that should be being run through rose bunch should be variants on a command so, for the general case, one fails: all fail. I'm not sure duplicating the entire .stderr file is the way to go - particularly if you consider something like the reconfiguration (original use case) where it chucks out hundreds of lines of warnings to stderr during even a successful run. It may be better to resurrect item 2 of #1876 instead. |
|
A system's pollution of their output and error is their own problem! Better to see too much than nothing at all - and this is mainly targeted at users of the cylc gui rather than Rose Bush. |
So you don't see creation of redundant files as a bad thing to do? No point in having the bunch.err file if you're going to shove it in the job.err. |
|
Given that neither of us are in the Rose team any more, our opinions are partly moot. The point of having the bunch.err would be in your situation marked above - drilling down to specifics, particularly if there is a lot of error. Also in your situation above, if the system produces lots of warnings you may want to look at the bunch.err even when the job does not fail. I don't think it's redundant at all. At the moment everyone does a two-step dance - look at the job.err to see what failed, then look at one of the bunch errors. If they are doing it in the |
|
Don't think we're going to agree on this so I'll leave this one to @matthewrmshin to think about. I wonder if a |
|
🤔 |
|
This is a hard option to name and describe correctly, given that I think this new behaviour should be the default (I reckon everyone wants it except you Andy :p). If we can't name it correctly, is it worth having at all? Just pass this in as-is? |
|
Are we going for the old or new behaviour as the default? |
|
new new new |
|
Discussed offline with @benfitzpatrick and suggested some alternative names for the setting. @benfitzpatrick to follow up with other major users of rose bunch and double check they'd be ok with a change in behaviour. |
|
@benfitzpatrick - does the |
|
No, although it would be more useful if it supported wildcards... |
|
Might it be worth extending the cylc functionality then instead? |
|
Now that it is spring, I think it is time for a spring clean. Please raise a new pull request when we are ready to have another crack at this. |
This is a small change to print out the individual bunch stderr to the job.err on failure so that debugging is easier.