Fix uArch metrics sampling for relative-seconds timestamp formats#551
Open
excelle08 wants to merge 2 commits intofacebookresearch:v2-betafrom
Open
Fix uArch metrics sampling for relative-seconds timestamp formats#551excelle08 wants to merge 2 commits intofacebookresearch:v2-betafrom
excelle08 wants to merge 2 commits intofacebookresearch:v2-betafrom
Conversation
Summary: Upgrade DjangoBench V2 to use Python 3.14 and Cinder 3.14 to take advantage of newer language features and performance improvements. This also adds CinderX for JIT compilation support in the Cinder runtime environment. Key changes: - CPython upgraded from 3.10 to 3.14 (built from source on all platforms) - Cinder upgraded from cinder/3.10 branch to meta/3.14 at pinned commit - CinderX installed for JIT functionality with validation - Removed numpy dependency in favor of Python's built-in statistics module - cassandra-driver now builds from PyPI source for Python 3.14 compatibility Differential Revision: D91604990
Summary: PerfPub failed to correctly sample uArch time-series CSVs (amd-perf-collector, amd-zen4/zen5-perf-collector, nv-perf-collector, arm-perf-collector, topdown-intel) when using breakdown.csv for timestamp alignment. These CSVs use numeric relative-second timestamps (e.g. Timestamp_Secs starting from 0) rather than time-of-day strings (e.g. "04:33:45 PM"). The existing code only handled the time-of-day format, causing a silent fallback to last_secs=300 which sampled the wrong time window and produced incorrect metrics (e.g. 8 GB/s memory BW on BGM instead of >100 GB/s). Fix: - Add TIMESTAMP_COLUMN_MAP routing each CSV filename to its timestamp column name and format type (time_of_day vs relative_secs) - Compute relative-second offsets from bm_metrics["timestamp"] epoch to align breakdown.csv absolute datetimes with uArch CSV relative timestamps - Add relative_secs matching path in get_start_end_index() that finds the closest rows by numeric offset instead of time-of-day string parsing Differential Revision: D98835676
meta-cla
bot
added
the
CLA Signed
|
@excelle08 has exported this pull request. If you are a Meta employee, you can view the originating Diff in D98835676. |
This file contains hidden or 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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
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.
Summary:
PerfPub failed to correctly sample uArch time-series CSVs (amd-perf-collector,
amd-zen4/zen5-perf-collector, nv-perf-collector, arm-perf-collector,
topdown-intel) when using breakdown.csv for timestamp alignment. These CSVs
use numeric relative-second timestamps (e.g. Timestamp_Secs starting from 0)
rather than time-of-day strings (e.g. "04:33:45 PM"). The existing code only
handled the time-of-day format, causing a silent fallback to last_secs=300
which sampled the wrong time window and produced incorrect metrics (e.g.
8 GB/s memory BW on BGM instead of >100 GB/s).
Fix:
name and format type (time_of_day vs relative_secs)
breakdown.csv absolute datetimes with uArch CSV relative timestamps
closest rows by numeric offset instead of time-of-day string parsing
Differential Revision: D98835676