[release][tune] [build_base] Fix tune_scalability_durable_trainable configuration to force experiment syncs less often
#31649
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Signed-off-by: Justin Yu <[email protected]>
…y_tune_scalability_test
…y_tune_scalability_test
…y_tune_scalability_test
krfricke
requested changes
Jan 13, 2023
@krfricke Yes, we could increase the training time and increase the test pass threshold. Currently, the fake training time per iteration is 6s. |
|
@justinvyu I'd prefer that option then, could you update the PR? Thanks! |
Signed-off-by: Justin Yu <[email protected]>
…y_tune_scalability_test
krfricke
approved these changes
Jan 17, 2023
andreapiso
pushed a commit
to andreapiso/ray
that referenced
this pull request
Jan 22, 2023
… configuration to force experiment syncs less often (ray-project#31649) Deflakes `tune_scalability_durable_trainable` release test by configuring it so that forced experiment checkpoint syncs happen less frequently, cutting down on the test runtime. Signed-off-by: Justin Yu <[email protected]> Signed-off-by: Andrea Pisoni <[email protected]>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Deflakes
tune_scalability_durable_trainablerelease test by configuring it so that forced experiment checkpoint syncs happen less frequently, cutting down on the test runtime.Why are these changes needed?
The
tune_scalability_durable_trainabletest had a performance regression (running for ~300s longer) and is flaky. This was due to experiment checkpoint syncing to cloud being forced more often due to a fix by #31131. Each forced sync causes the driver to wait until that sync is finished before launching a new one, which accounts for the drastically increased runtime. This PR fixes the issue by increasing the number of checkpoints to keep, which in turn reduces the frequency of forced cloud syncs.Question: How can we provide a better default for users? Can we provide a maximum amount of time spent waiting on cloud syncs? So that even if the user configures
keep_checkpoints_numin such a way that syncs are getting forced very frequently, their script runtime doesn't get affected past a certain amount.Related issue number
Closes #31506
Checks
git commit -s) in this PR.scripts/format.shto lint the changes in this PR.