New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
ENH: runtime HEATMAP/DXT cross-check log #22
ENH: runtime HEATMAP/DXT cross-check log #22
Conversation
If reviewers want to look in to the CI issues, that may be helpful as well (if it isn't obvious, you can start by diffing the current logs with previous successful logs for the repo, etc.) |
6cd8882
to
679597d
Compare
* add a carefully crafted darshan log using latest `darshan` `main` branch that contains "runtime" `HEATMAP` module and DXT tracing, to facilitate the writing of carefully designed tests for internal consistency of data structures (and to facilitate rapid visual inspection as well) * see the `README` for a very detailed description of how the data was produced (this is perhaps a bit too detailed, though I'm mostly doing this so that more folks can generate carefully-crafted testing data using pure Python without waiting for me for example)
679597d
to
9b3624c
Compare
At a glance the CI errors don't seem directly related to this PR, it's acting like
I'm guessing this is because the CI operates a little differently in this repo vs. the main one (i.e. I was expecting a failure somewhere in
And on Jakob's branch: |
I tried reproducing the CI steps locally on my machine and hit the same errors. As @nawtrey points out, I can work around the issue just by dropping the I noticed this comparing these CI failures to the last success: Last success:
This job:
So maybe it's related to a newer pytest version? Anything we might expect to change with |
Was also able to quickly confirm that |
I looked through the Like I noted above, locally In the mean time I tried making some changes to the |
Sounds good, feel free to do that and maybe open an issue reminding us to try unpinning the future. We could also just use |
I'll submit a PR with pytest pinned and open an issue so that we revisit this at some point. There are additional issues on #23 I need to look into, but we should be good on this PR. Should I merge despite pydarshan failures related to heatmap bindings? Or hold off until darshan-hpc/darshan#615 is in? |
I'd probably vote for merging, then using the log in this PR as a test case for flushing the code paths in Jakob's PR (I already pasted a draft "sanity check" style test over there, so worst case scenario we do something like that initially instead of the deeper testing I had in mind for the future with the diagonals..). I may be biased though. Once thing I do want to make sure of is that gh-23 confusion isn't an indication of a deeper problem with the runtime heatmap parsing infra. |
I would also vote for merging this, we are probably better off making sure darshan-hpc/darshan#615 is well tested than worrying too much about a CI failure in this repo. |
add a carefully crafted darshan log using latest
darshan
main
branch that contains "runtime"HEATMAP
moduleand DXT tracing, to facilitate the writing of carefully
designed tests for internal consistency of data structures
(and to facilitate rapid visual inspection as well)
see the
README
for a very detailed description of howthe data was produced (this is perhaps a bit too detailed,
though I'm mostly doing this so that more folks can generate
carefully-crafted testing data using pure Python without
waiting for me for example)
Here is a sample comparison of the DXT vs. runtime heatmaps in a hacked version of Jakob's branch from gh-615, with a few debug prints as well, to show data structure similarity on top of the obvious visual similarities. I hope it is clear that the write activity, which I intentionally made diagonal for visual inspection, matches.
Debug prints:
If developers would like to modify the application to, for example, produce read events as well, I think I've provided sufficient detail for folks to make that adjustment (feel free to do that and push in here/adjust README, etc.). If not, this may provide sufficient detail for initial flushing of runtime
HEATMAP
code paths on the Python side in gh-615, and for subsequent more detailed testing/cross-comparison with the DXT heatmap results later on.