Improve performance of utils.ReplaceRegexp#51872
Merged
rosstimothy merged 1 commit intomasterfrom Feb 6, 2025
Merged
Conversation
Contributor
Author
|
Flame graph from a CPU profile that inspired this change:
|
rosstimothy
force-pushed
the
tross/regex_replace_cpu
branch
from
137b14e to
4cb540f
Compare
fspmarshall
approved these changes
Feb 5, 2025
rosstimothy
force-pushed
the
tross/regex_replace_cpu
branch
from
4cb540f to
dd8a903
Compare
espadolini
reviewed
Feb 5, 2025
espadolini
approved these changes
Feb 6, 2025
Comment on lines
108
to
115
| index := expr.FindStringIndex(input) | ||
| if index == nil { | ||
| return "", trace.NotFound("no match found") | ||
| // The returned error is intentionally not wrapped to avoid | ||
| // capturing stack traces. This method is used by authorization | ||
| // logic and the additional overhead of strack trace capturing | ||
| // is a performance bottleneck. | ||
| return "", ErrReplaceRegexNotFound | ||
| } | ||
| return expr.ReplaceAllString(input, replaceWith), nil |
Contributor
There was a problem hiding this comment.
Is this applying the regexp twice on the same input? Is there a way to use the regexp API to avoid that?
Contributor
Author
There was a problem hiding this comment.
I'm not seeing an obvious way, though the ReplaceRegexpWith and RegexpWithConfig APIs seem to come from a previous iteration of improving performance of trait resolution (#19097). I've not investigated fully but I wonder if they are even needed anymore?
The main changes here are: - Using an LRU cache to store compiled regular expressions. - Removing stack traces captured by trace.NotFound/trace.Wrap when there are no matches. This was heavily inspired by #23534 which made similar changes to improve the performance of utils.MatchString. The main beneficiaries of this change are services.MapRoles and services.TraitsToRoles which rely heavily on utils.ReplaceRegexp, utils.RegexpWithConfig, or ReplaceRegexpWith. BenchmarkReplaceRegexp was added to validate the improvements in this change and prevent regressions in the future. Results from before and after this change: benchstat old.txt new.txt goos: darwin goarch: arm64 pkg: github.com/gravitational/teleport/lib/utils cpu: Apple M2 Pro │ old.txt │ new.txt │ │ sec/op │ sec/op vs base │ ReplaceRegexp/same_expression-12 29.527µ ± 12% 6.837µ ± 1% -76.84% (p=0.000 n=10) ReplaceRegexp/unique_expressions-12 22.94µ ± 1% 25.10µ ± 1% +9.41% (p=0.002 n=10) ReplaceRegexp/no_matches-12 24.692µ ± 3% 2.861µ ± 1% -88.41% (p=0.000 n=10) geomean 25.57µ 7.889µ -69.15% │ old.txt │ new.txt │ │ B/op │ B/op vs base │ ReplaceRegexp/same_expression-12 22071.5 ± 1% 164.0 ± 0% -99.26% (p=0.000 n=10) ReplaceRegexp/unique_expressions-12 12.90Ki ± 1% 12.87Ki ± 0% ~ (p=0.165 n=10) ReplaceRegexp/no_matches-12 14257.00 ± 1% 15.00 ± 0% -99.89% (p=0.000 n=10) geomean 15.70Ki 318.8 -98.02% │ old.txt │ new.txt │ │ allocs/op │ allocs/op vs base │ ReplaceRegexp/same_expression-12 58.000 ± 0% 6.000 ± 0% -89.66% (p=0.000 n=10) ReplaceRegexp/unique_expressions-12 55.00 ± 0% 55.00 ± 0% ~ (p=1.000 n=10) ¹ ReplaceRegexp/no_matches-12 58.00 ± 0% 0.00 ± 0% -100.00% (p=0.000 n=10) geomean 56.98 ? ² ³ ¹ all samples are equal ² summaries must be >0 to compute geomean ³ ratios must be >0 to compute geomean
rosstimothy
force-pushed
the
tross/regex_replace_cpu
branch
from
385973a to
338b107
Compare
|
@rosstimothy See the table below for backport results.
|
rosstimothy
added a commit
that referenced
this pull request
Feb 6, 2025
The main changes here are: - Using an LRU cache to store compiled regular expressions. - Removing stack traces captured by trace.NotFound/trace.Wrap when there are no matches. This was heavily inspired by #23534 which made similar changes to improve the performance of utils.MatchString. The main beneficiaries of this change are services.MapRoles and services.TraitsToRoles which rely heavily on utils.ReplaceRegexp, utils.RegexpWithConfig, or ReplaceRegexpWith. BenchmarkReplaceRegexp was added to validate the improvements in this change and prevent regressions in the future. Results from before and after this change: benchstat old.txt new.txt goos: darwin goarch: arm64 pkg: github.com/gravitational/teleport/lib/utils cpu: Apple M2 Pro │ old.txt │ new.txt │ │ sec/op │ sec/op vs base │ ReplaceRegexp/same_expression-12 29.527µ ± 12% 6.837µ ± 1% -76.84% (p=0.000 n=10) ReplaceRegexp/unique_expressions-12 22.94µ ± 1% 25.10µ ± 1% +9.41% (p=0.002 n=10) ReplaceRegexp/no_matches-12 24.692µ ± 3% 2.861µ ± 1% -88.41% (p=0.000 n=10) geomean 25.57µ 7.889µ -69.15% │ old.txt │ new.txt │ │ B/op │ B/op vs base │ ReplaceRegexp/same_expression-12 22071.5 ± 1% 164.0 ± 0% -99.26% (p=0.000 n=10) ReplaceRegexp/unique_expressions-12 12.90Ki ± 1% 12.87Ki ± 0% ~ (p=0.165 n=10) ReplaceRegexp/no_matches-12 14257.00 ± 1% 15.00 ± 0% -99.89% (p=0.000 n=10) geomean 15.70Ki 318.8 -98.02% │ old.txt │ new.txt │ │ allocs/op │ allocs/op vs base │ ReplaceRegexp/same_expression-12 58.000 ± 0% 6.000 ± 0% -89.66% (p=0.000 n=10) ReplaceRegexp/unique_expressions-12 55.00 ± 0% 55.00 ± 0% ~ (p=1.000 n=10) ¹ ReplaceRegexp/no_matches-12 58.00 ± 0% 0.00 ± 0% -100.00% (p=0.000 n=10) geomean 56.98 ? ² ³ ¹ all samples are equal ² summaries must be >0 to compute geomean ³ ratios must be >0 to compute geomean
rosstimothy
added a commit
that referenced
this pull request
Feb 6, 2025
The main changes here are: - Using an LRU cache to store compiled regular expressions. - Removing stack traces captured by trace.NotFound/trace.Wrap when there are no matches. This was heavily inspired by #23534 which made similar changes to improve the performance of utils.MatchString. The main beneficiaries of this change are services.MapRoles and services.TraitsToRoles which rely heavily on utils.ReplaceRegexp, utils.RegexpWithConfig, or ReplaceRegexpWith. BenchmarkReplaceRegexp was added to validate the improvements in this change and prevent regressions in the future. Results from before and after this change: benchstat old.txt new.txt goos: darwin goarch: arm64 pkg: github.com/gravitational/teleport/lib/utils cpu: Apple M2 Pro │ old.txt │ new.txt │ │ sec/op │ sec/op vs base │ ReplaceRegexp/same_expression-12 29.527µ ± 12% 6.837µ ± 1% -76.84% (p=0.000 n=10) ReplaceRegexp/unique_expressions-12 22.94µ ± 1% 25.10µ ± 1% +9.41% (p=0.002 n=10) ReplaceRegexp/no_matches-12 24.692µ ± 3% 2.861µ ± 1% -88.41% (p=0.000 n=10) geomean 25.57µ 7.889µ -69.15% │ old.txt │ new.txt │ │ B/op │ B/op vs base │ ReplaceRegexp/same_expression-12 22071.5 ± 1% 164.0 ± 0% -99.26% (p=0.000 n=10) ReplaceRegexp/unique_expressions-12 12.90Ki ± 1% 12.87Ki ± 0% ~ (p=0.165 n=10) ReplaceRegexp/no_matches-12 14257.00 ± 1% 15.00 ± 0% -99.89% (p=0.000 n=10) geomean 15.70Ki 318.8 -98.02% │ old.txt │ new.txt │ │ allocs/op │ allocs/op vs base │ ReplaceRegexp/same_expression-12 58.000 ± 0% 6.000 ± 0% -89.66% (p=0.000 n=10) ReplaceRegexp/unique_expressions-12 55.00 ± 0% 55.00 ± 0% ~ (p=1.000 n=10) ¹ ReplaceRegexp/no_matches-12 58.00 ± 0% 0.00 ± 0% -100.00% (p=0.000 n=10) geomean 56.98 ? ² ³ ¹ all samples are equal ² summaries must be >0 to compute geomean ³ ratios must be >0 to compute geomean
This was referenced Feb 6, 2025
rosstimothy
added a commit
that referenced
this pull request
Feb 7, 2025
The main changes here are: - Using an LRU cache to store compiled regular expressions. - Removing stack traces captured by trace.NotFound/trace.Wrap when there are no matches. This was heavily inspired by #23534 which made similar changes to improve the performance of utils.MatchString. The main beneficiaries of this change are services.MapRoles and services.TraitsToRoles which rely heavily on utils.ReplaceRegexp, utils.RegexpWithConfig, or ReplaceRegexpWith. BenchmarkReplaceRegexp was added to validate the improvements in this change and prevent regressions in the future. Results from before and after this change: benchstat old.txt new.txt goos: darwin goarch: arm64 pkg: github.com/gravitational/teleport/lib/utils cpu: Apple M2 Pro │ old.txt │ new.txt │ │ sec/op │ sec/op vs base │ ReplaceRegexp/same_expression-12 29.527µ ± 12% 6.837µ ± 1% -76.84% (p=0.000 n=10) ReplaceRegexp/unique_expressions-12 22.94µ ± 1% 25.10µ ± 1% +9.41% (p=0.002 n=10) ReplaceRegexp/no_matches-12 24.692µ ± 3% 2.861µ ± 1% -88.41% (p=0.000 n=10) geomean 25.57µ 7.889µ -69.15% │ old.txt │ new.txt │ │ B/op │ B/op vs base │ ReplaceRegexp/same_expression-12 22071.5 ± 1% 164.0 ± 0% -99.26% (p=0.000 n=10) ReplaceRegexp/unique_expressions-12 12.90Ki ± 1% 12.87Ki ± 0% ~ (p=0.165 n=10) ReplaceRegexp/no_matches-12 14257.00 ± 1% 15.00 ± 0% -99.89% (p=0.000 n=10) geomean 15.70Ki 318.8 -98.02% │ old.txt │ new.txt │ │ allocs/op │ allocs/op vs base │ ReplaceRegexp/same_expression-12 58.000 ± 0% 6.000 ± 0% -89.66% (p=0.000 n=10) ReplaceRegexp/unique_expressions-12 55.00 ± 0% 55.00 ± 0% ~ (p=1.000 n=10) ¹ ReplaceRegexp/no_matches-12 58.00 ± 0% 0.00 ± 0% -100.00% (p=0.000 n=10) geomean 56.98 ? ² ³ ¹ all samples are equal ² summaries must be >0 to compute geomean ³ ratios must be >0 to compute geomean
github-merge-queue Bot
pushed a commit
that referenced
this pull request
Feb 7, 2025
The main changes here are: - Using an LRU cache to store compiled regular expressions. - Removing stack traces captured by trace.NotFound/trace.Wrap when there are no matches. This was heavily inspired by #23534 which made similar changes to improve the performance of utils.MatchString. The main beneficiaries of this change are services.MapRoles and services.TraitsToRoles which rely heavily on utils.ReplaceRegexp, utils.RegexpWithConfig, or ReplaceRegexpWith. BenchmarkReplaceRegexp was added to validate the improvements in this change and prevent regressions in the future. Results from before and after this change: benchstat old.txt new.txt goos: darwin goarch: arm64 pkg: github.com/gravitational/teleport/lib/utils cpu: Apple M2 Pro │ old.txt │ new.txt │ │ sec/op │ sec/op vs base │ ReplaceRegexp/same_expression-12 29.527µ ± 12% 6.837µ ± 1% -76.84% (p=0.000 n=10) ReplaceRegexp/unique_expressions-12 22.94µ ± 1% 25.10µ ± 1% +9.41% (p=0.002 n=10) ReplaceRegexp/no_matches-12 24.692µ ± 3% 2.861µ ± 1% -88.41% (p=0.000 n=10) geomean 25.57µ 7.889µ -69.15% │ old.txt │ new.txt │ │ B/op │ B/op vs base │ ReplaceRegexp/same_expression-12 22071.5 ± 1% 164.0 ± 0% -99.26% (p=0.000 n=10) ReplaceRegexp/unique_expressions-12 12.90Ki ± 1% 12.87Ki ± 0% ~ (p=0.165 n=10) ReplaceRegexp/no_matches-12 14257.00 ± 1% 15.00 ± 0% -99.89% (p=0.000 n=10) geomean 15.70Ki 318.8 -98.02% │ old.txt │ new.txt │ │ allocs/op │ allocs/op vs base │ ReplaceRegexp/same_expression-12 58.000 ± 0% 6.000 ± 0% -89.66% (p=0.000 n=10) ReplaceRegexp/unique_expressions-12 55.00 ± 0% 55.00 ± 0% ~ (p=1.000 n=10) ¹ ReplaceRegexp/no_matches-12 58.00 ± 0% 0.00 ± 0% -100.00% (p=0.000 n=10) geomean 56.98 ? ² ³ ¹ all samples are equal ² summaries must be >0 to compute geomean ³ ratios must be >0 to compute geomean
github-merge-queue Bot
pushed a commit
that referenced
this pull request
Feb 7, 2025
The main changes here are: - Using an LRU cache to store compiled regular expressions. - Removing stack traces captured by trace.NotFound/trace.Wrap when there are no matches. This was heavily inspired by #23534 which made similar changes to improve the performance of utils.MatchString. The main beneficiaries of this change are services.MapRoles and services.TraitsToRoles which rely heavily on utils.ReplaceRegexp, utils.RegexpWithConfig, or ReplaceRegexpWith. BenchmarkReplaceRegexp was added to validate the improvements in this change and prevent regressions in the future. Results from before and after this change: benchstat old.txt new.txt goos: darwin goarch: arm64 pkg: github.com/gravitational/teleport/lib/utils cpu: Apple M2 Pro │ old.txt │ new.txt │ │ sec/op │ sec/op vs base │ ReplaceRegexp/same_expression-12 29.527µ ± 12% 6.837µ ± 1% -76.84% (p=0.000 n=10) ReplaceRegexp/unique_expressions-12 22.94µ ± 1% 25.10µ ± 1% +9.41% (p=0.002 n=10) ReplaceRegexp/no_matches-12 24.692µ ± 3% 2.861µ ± 1% -88.41% (p=0.000 n=10) geomean 25.57µ 7.889µ -69.15% │ old.txt │ new.txt │ │ B/op │ B/op vs base │ ReplaceRegexp/same_expression-12 22071.5 ± 1% 164.0 ± 0% -99.26% (p=0.000 n=10) ReplaceRegexp/unique_expressions-12 12.90Ki ± 1% 12.87Ki ± 0% ~ (p=0.165 n=10) ReplaceRegexp/no_matches-12 14257.00 ± 1% 15.00 ± 0% -99.89% (p=0.000 n=10) geomean 15.70Ki 318.8 -98.02% │ old.txt │ new.txt │ │ allocs/op │ allocs/op vs base │ ReplaceRegexp/same_expression-12 58.000 ± 0% 6.000 ± 0% -89.66% (p=0.000 n=10) ReplaceRegexp/unique_expressions-12 55.00 ± 0% 55.00 ± 0% ~ (p=1.000 n=10) ¹ ReplaceRegexp/no_matches-12 58.00 ± 0% 0.00 ± 0% -100.00% (p=0.000 n=10) geomean 56.98 ? ² ³ ¹ all samples are equal ² summaries must be >0 to compute geomean ³ ratios must be >0 to compute geomean
carloscastrojumo
pushed a commit
to carloscastrojumo/teleport
that referenced
this pull request
Feb 19, 2025
The main changes here are: - Using an LRU cache to store compiled regular expressions. - Removing stack traces captured by trace.NotFound/trace.Wrap when there are no matches. This was heavily inspired by gravitational#23534 which made similar changes to improve the performance of utils.MatchString. The main beneficiaries of this change are services.MapRoles and services.TraitsToRoles which rely heavily on utils.ReplaceRegexp, utils.RegexpWithConfig, or ReplaceRegexpWith. BenchmarkReplaceRegexp was added to validate the improvements in this change and prevent regressions in the future. Results from before and after this change: benchstat old.txt new.txt goos: darwin goarch: arm64 pkg: github.com/gravitational/teleport/lib/utils cpu: Apple M2 Pro │ old.txt │ new.txt │ │ sec/op │ sec/op vs base │ ReplaceRegexp/same_expression-12 29.527µ ± 12% 6.837µ ± 1% -76.84% (p=0.000 n=10) ReplaceRegexp/unique_expressions-12 22.94µ ± 1% 25.10µ ± 1% +9.41% (p=0.002 n=10) ReplaceRegexp/no_matches-12 24.692µ ± 3% 2.861µ ± 1% -88.41% (p=0.000 n=10) geomean 25.57µ 7.889µ -69.15% │ old.txt │ new.txt │ │ B/op │ B/op vs base │ ReplaceRegexp/same_expression-12 22071.5 ± 1% 164.0 ± 0% -99.26% (p=0.000 n=10) ReplaceRegexp/unique_expressions-12 12.90Ki ± 1% 12.87Ki ± 0% ~ (p=0.165 n=10) ReplaceRegexp/no_matches-12 14257.00 ± 1% 15.00 ± 0% -99.89% (p=0.000 n=10) geomean 15.70Ki 318.8 -98.02% │ old.txt │ new.txt │ │ allocs/op │ allocs/op vs base │ ReplaceRegexp/same_expression-12 58.000 ± 0% 6.000 ± 0% -89.66% (p=0.000 n=10) ReplaceRegexp/unique_expressions-12 55.00 ± 0% 55.00 ± 0% ~ (p=1.000 n=10) ¹ ReplaceRegexp/no_matches-12 58.00 ± 0% 0.00 ± 0% -100.00% (p=0.000 n=10) geomean 56.98 ? ² ³ ¹ all samples are equal ² summaries must be >0 to compute geomean ³ ratios must be >0 to compute geomean
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
3 participants
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.
The main changes here are:
This was heavily inspired by #23534 which made similar changes to improve the performance of utils.MatchString.
The main beneficiaries of this change are services.MapRoles and services.TraitsToRoles which rely heavily on
utils.ReplaceRegexp, utils.RegexpWithConfig, or ReplaceRegexpWith.
BenchmarkReplaceRegexp was added to validate the improvements in this change and prevent regressions in the future. Results from before and after this change:
$ benchstat old.txt new.txt goos: darwin goarch: arm64 pkg: github.com/gravitational/teleport/lib/utils cpu: Apple M2 Pro │ old.txt │ new.txt │ │ sec/op │ sec/op vs base │ ReplaceRegexp/same_expression-12 29.527µ ± 12% 6.837µ ± 1% -76.84% (p=0.000 n=10) ReplaceRegexp/unique_expressions-12 22.94µ ± 1% 25.10µ ± 1% +9.41% (p=0.002 n=10) ReplaceRegexp/no_matches-12 24.692µ ± 3% 2.861µ ± 1% -88.41% (p=0.000 n=10) geomean 25.57µ 7.889µ -69.15% │ old.txt │ new.txt │ │ B/op │ B/op vs base │ ReplaceRegexp/same_expression-12 22071.5 ± 1% 164.0 ± 0% -99.26% (p=0.000 n=10) ReplaceRegexp/unique_expressions-12 12.90Ki ± 1% 12.87Ki ± 0% ~ (p=0.165 n=10) ReplaceRegexp/no_matches-12 14257.00 ± 1% 15.00 ± 0% -99.89% (p=0.000 n=10) geomean 15.70Ki 318.8 -98.02% │ old.txt │ new.txt │ │ allocs/op │ allocs/op vs base │ ReplaceRegexp/same_expression-12 58.000 ± 0% 6.000 ± 0% -89.66% (p=0.000 n=10) ReplaceRegexp/unique_expressions-12 55.00 ± 0% 55.00 ± 0% ~ (p=1.000 n=10) ¹ ReplaceRegexp/no_matches-12 58.00 ± 0% 0.00 ± 0% -100.00% (p=0.000 n=10) geomean 56.98 ? ² ³ ¹ all samples are equal ² summaries must be >0 to compute geomean ³ ratios must be >0 to compute geomeanChangelog: Reduce CPU consumption required to map roles between clusters and perform trait to role resolution.