feat(python,rust): Support engine callback for LazyFrame.profile
#21534
+220
−92
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
wence-
requested review from
ritchie46,
c-peters,
alexander-beedie,
MarcoGorelli,
reswqa and
orlp
as code owners
|
This is an alternative to #20693 that tries to avoid exposing too many internals. |
wence-
changed the title
LazyFrame.profile
wence-
force-pushed
the
wence/fea/profile-gpu
branch
from
c9a8794 to
bb518af
Compare
wence-
commented
Feb 28, 2025
wence-
commented
Feb 28, 2025
This was referenced Feb 28, 2025
So that we can support the GPU engine in profiled collection of a lazyframe, plumb through a mechanism for recording raw timings for nodes that were executed through the PythonScan node. This necessitates some small changes to the internals of the NodeTimer, since `Instant`s are opaque. We instead directly store durations (as nanoseconds since the query start) and when calling into the IR post-optimization callback, provide a duration that is the number of nanoseconds since the query was started. On the Python side we can then keep track and record durations independently, offsetted by this optimisation duration. As a side-effect, `profile` now correctly measures the optimisation time in the logical plan, rather than as previously just the time to produce the physical plan from the optimised logical plan.
wence-
force-pushed
the
wence/fea/profile-gpu
branch
from
bb518af to
6f893e8
Compare
wence-
changed the title
LazyFrame.profileLazyFrame.profile
github-actions
bot
added
enhancement
wence-
commented
Feb 28, 2025
| with_columns.map(|x| x.into_iter().map(|x| x.to_string()).collect::<Vec<_>>()), | ||
| predicate, | ||
| n_rows, | ||
| // Is this boolean is true, callback should return |
There was a problem hiding this comment.
Suggested change
| // Is this boolean is true, callback should return | |
| // If this boolean is true, callback should return |
ritchie46
reviewed
Mar 1, 2025
| @@ -29,7 +29,19 @@ impl NodeTimer { | |||
| let nodes = &mut data.0; | |||
| nodes.push(name); | |||
| let ticks = &mut data.1; | |||
| ticks.push((start, end)) | |||
| ticks.push(( | |||
There was a problem hiding this comment.
Can we replace this whole inner with:
store_raw(
start.duration_since(self.query_start),
end.duration_since(self.query_start),
name)
| # Don't run on GPU in _eager mode (but don't warn) | ||
| is_gpu = False | ||
|
|
||
| callback = _gpu_engine_callback( |
| Ok(()) | ||
| }) | ||
| ldf._collect_post_opt(|root, lp_arena, expr_arena, _| { | ||
| post_opt_callback(&lambda, root, lp_arena, expr_arena, None) |
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.
So that we can support the GPU engine in profiled collection of a lazyframe, plumb through a mechanism for recording raw timings for nodes that were executed through the PythonScan node.
This necessitates some small changes to the internals of the NodeTimer, since
Instants are opaque. We instead directly store durations (as nanoseconds since the query start) and when calling into the IR post-optimization callback, provide a duration that is the number of nanoseconds since the query was started. On the Python side we can then keep track and record durations independently, offsetted by this optimisation duration.As a side-effect,
profilenow correctly measures the optimisation time in the logical plan, rather than as previously just the time to produce the physical plan from the optimised logical plan.