-
Notifications
You must be signed in to change notification settings - Fork 140
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
[Profiler] Avoid too many syscall calls for timer-create-based CPU profiler #6067
[Profiler] Avoid too many syscall calls for timer-create-based CPU profiler #6067
Conversation
Datadog ReportBranch report: ✅ 0 Failed, 366526 Passed, 2340 Skipped, 16h 41m 51.65s Total Time |
Execution-Time Benchmarks Report ⏱️Execution-time results for samples comparing the following branches/commits: Execution-time benchmarks measure the whole time it takes to execute a program. And are intended to measure the one-off costs. Cases where the execution time results for the PR are worse than latest master results are shown in red. The following thresholds were used for comparing the execution times:
Note that these results are based on a single point-in-time result for each branch. For full results, see the dashboard. Graphs show the p99 interval based on the mean and StdDev of the test run, as well as the mean value of the run (shown as a diamond below the graph). gantt
title Execution time (ms) FakeDbCommand (.NET Framework 4.6.2)
dateFormat X
axisFormat %s
todayMarker off
section Baseline
This PR (6067) - mean (70ms) : 68, 72
. : milestone, 70,
master - mean (70ms) : 67, 73
. : milestone, 70,
section CallTarget+Inlining+NGEN
This PR (6067) - mean (1,100ms) : 1081, 1119
. : milestone, 1100,
master - mean (1,105ms) : 1067, 1143
. : milestone, 1105,
gantt
title Execution time (ms) FakeDbCommand (.NET Core 3.1)
dateFormat X
axisFormat %s
todayMarker off
section Baseline
This PR (6067) - mean (109ms) : 106, 112
. : milestone, 109,
master - mean (109ms) : 105, 112
. : milestone, 109,
section CallTarget+Inlining+NGEN
This PR (6067) - mean (768ms) : 757, 779
. : milestone, 768,
master - mean (766ms) : 751, 782
. : milestone, 766,
gantt
title Execution time (ms) FakeDbCommand (.NET 6)
dateFormat X
axisFormat %s
todayMarker off
section Baseline
This PR (6067) - mean (92ms) : 89, 94
. : milestone, 92,
master - mean (92ms) : 90, 94
. : milestone, 92,
section CallTarget+Inlining+NGEN
This PR (6067) - mean (725ms) : 714, 737
. : milestone, 725,
master - mean (727ms) : 707, 747
. : milestone, 727,
gantt
title Execution time (ms) HttpMessageHandler (.NET Framework 4.6.2)
dateFormat X
axisFormat %s
todayMarker off
section Baseline
This PR (6067) - mean (190ms) : 187, 194
. : milestone, 190,
master - mean (191ms) : 187, 195
. : milestone, 191,
section CallTarget+Inlining+NGEN
This PR (6067) - mean (1,190ms) : 1168, 1211
. : milestone, 1190,
master - mean (1,202ms) : 1165, 1240
. : milestone, 1202,
gantt
title Execution time (ms) HttpMessageHandler (.NET Core 3.1)
dateFormat X
axisFormat %s
todayMarker off
section Baseline
This PR (6067) - mean (276ms) : 271, 281
. : milestone, 276,
master - mean (276ms) : 270, 281
. : milestone, 276,
section CallTarget+Inlining+NGEN
This PR (6067) - mean (939ms) : 918, 961
. : milestone, 939,
master - mean (938ms) : 918, 958
. : milestone, 938,
gantt
title Execution time (ms) HttpMessageHandler (.NET 6)
dateFormat X
axisFormat %s
todayMarker off
section Baseline
This PR (6067) - mean (264ms) : 261, 268
. : milestone, 264,
master - mean (265ms) : 261, 269
. : milestone, 265,
section CallTarget+Inlining+NGEN
This PR (6067) - mean (925ms) : 908, 942
. : milestone, 925,
master - mean (922ms) : 906, 939
. : milestone, 922,
|
Throughput/Crank Report ⚡Throughput results for AspNetCoreSimpleController comparing the following branches/commits: Cases where throughput results for the PR are worse than latest master (5% drop or greater), results are shown in red. Note that these results are based on a single point-in-time result for each branch. For full results, see one of the many, many dashboards! gantt
title Throughput Linux x64 (Total requests)
dateFormat X
axisFormat %s
section Baseline
This PR (6067) (11.245M) : 0, 11245443
master (11.101M) : 0, 11101013
benchmarks/2.9.0 (11.081M) : 0, 11080577
section Automatic
This PR (6067) (7.430M) : 0, 7429939
master (7.322M) : 0, 7321945
benchmarks/2.9.0 (7.732M) : 0, 7732233
section Trace stats
master (7.648M) : 0, 7647979
section Manual
master (11.159M) : 0, 11158874
section Manual + Automatic
This PR (6067) (6.779M) : 0, 6778675
master (6.756M) : 0, 6755889
section DD_TRACE_ENABLED=0
master (10.104M) : 0, 10103630
gantt
title Throughput Linux arm64 (Total requests)
dateFormat X
axisFormat %s
section Baseline
This PR (6067) (9.404M) : 0, 9403805
master (9.538M) : 0, 9538166
benchmarks/2.9.0 (9.798M) : 0, 9798067
section Automatic
This PR (6067) (6.598M) : 0, 6598285
master (6.560M) : 0, 6559669
section Trace stats
master (6.899M) : 0, 6898524
section Manual
master (9.530M) : 0, 9529944
section Manual + Automatic
This PR (6067) (6.133M) : 0, 6132517
master (6.111M) : 0, 6111240
section DD_TRACE_ENABLED=0
master (8.837M) : 0, 8836832
gantt
title Throughput Windows x64 (Total requests)
dateFormat X
axisFormat %s
section Baseline
This PR (6067) (10.099M) : 0, 10098813
master (10.017M) : 0, 10017307
benchmarks/2.9.0 (10.067M) : 0, 10067315
section Automatic
This PR (6067) (6.776M) : 0, 6776096
master (6.715M) : 0, 6715040
benchmarks/2.9.0 (7.552M) : 0, 7552193
section Trace stats
master (7.357M) : 0, 7357178
section Manual
master (9.959M) : 0, 9958707
section Manual + Automatic
This PR (6067) (6.275M) : 0, 6274970
master (6.139M) : 0, 6139159
section DD_TRACE_ENABLED=0
master (9.384M) : 0, 9383899
|
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LGTM
Summary of changes
Avoid too many syscall calls
Reason for change
TL;DR; syscalls add a sensitive overhead on the application. The more you add the more the application will suffer.
In a recent PR, I added the a way to disarm/rearm the
timer_create
-based CPU profiler. This was to prevent associating CPU to unwind stacks (by all profilers) to managed threads.This adds a sensitive number of syscalls calls (ex: Walltime => 5 * 2 (threads with no trace context) + 10 * 2 (threads with trace context : Code Hotspot) every 10 ms).
Implementation details
Remote
CpuProfilerDisableScope
class.Test coverage
Other details