profiler: avoid metrics profile log noise when stopping profiling#2865
Merged
Conversation
BenchmarksBenchmark execution time: 2024-09-24 15:47:06 Comparing candidate commit 9bd35fb in PR branch Found 1 performance improvements and 0 performance regressions! Performance is the same for 57 metrics, 1 unstable metrics. scenario:BenchmarkInjectW3C-24
|
The metrics profiler insisted on at least one second between collections for two reasons: - To avoid a division by zero because it was doing integer division to convert a time.Duration to seconds, which will truncate to 0, as a ratio in a subsequent computation - In case "a system clock issue causes time to run backwards" The profiler would report an error if less than one second elapsed between collections. In practice, this resulted in misleading error logs because it's entirely likely for profiling to be stopped less than a second after the last profile collection. The restriction was not really even needed. For one, we can just do floating-point division rather than integer division to avoid the truncation problem. Also, Go has had monotonic time support by default since 2017, added in Go 1.9, and time comparison operations including time.Time.Sub, work with respect to monotonic time. We shouldn't have any issues with negative periods. We can ensure the period is positive just as a defensive measure, and fail if it's negative since this may indicate a bug in the Go runtime if it's violating the monotonicity guarantees. Fixes #2863
nsrip-dd
force-pushed
the
nick.ripley/fix-metrics-profile-error
branch
from
September 19, 2024 18:28
6273290 to
f5d8369
Compare
nsrip-dd
marked this pull request as ready for review
September 19, 2024 18:39
Contributor
Author
Yeah, my wild guess is it's alignment or some other small perturbation. After pulling in main again the "regression" went away and we even have an "improvement". 🤷 |
nsrip-dd
added a commit
that referenced
this pull request
Sep 24, 2024
The Windows system timer resolution is about 15 milliseconds (see https://learn.microsoft.com/en-us/windows-hardware/drivers/kernel/high-resolution-timers#controlling-timer-accuracy) This caused the metrics profile tests from #2865 to fail because the metrics profiler will likely be stopped in less than 15 milliseconds, meaning we'll see 0 duration between profile collection and log an error. Drop TestMetricsProfileStopEarlyNoLog because it's not going to be useful if the timer resolution is that low. Bump the period in TestShortMetricsProfile from 10ms to 20ms so that the Windows timer will (hopefully) be able to actually measure the duration.
nsrip-dd
added a commit
that referenced
this pull request
Sep 24, 2024
The fix in #2865 was intended to suppress spurious metrics profile errors when the profiler is stopped. It did so by relaxing the one-second duration constraint of the metrics profiler. However, the Windows system timer resolution is about 15 milliseconds (see https://learn.microsoft.com/en-us/windows-hardware/drivers/kernel/high-resolution-timers#controlling-timer-accuracy) This caused the metrics profile tests from #2865 to fail because the metrics profiler will likely be stopped in less than 15 milliseconds, meaning we'll see 0 duration between profile collection and log an error. This commit actually suppresses the error by checking whether the profiler was stopped (meaning interruptibleSleep was interrupted). If so, and if the metrics profiler returned an error, we instead return a sentinel error indicating that profiling was stopped. If we see that error, we just drop the profile and don't log an error. We won't upload the profile anyway. This way, we should only report an error from the metrics profiler if there is _actually_ a problem with the timer.
nsrip-dd
added a commit
that referenced
this pull request
Sep 24, 2024
The fix in #2865 was intended to suppress spurious metrics profile errors when the profiler is stopped. It did so by relaxing the one-second duration constraint of the metrics profiler. However, the Windows system timer resolution is about 15 milliseconds (see https://learn.microsoft.com/en-us/windows-hardware/drivers/kernel/high-resolution-timers#controlling-timer-accuracy) This caused the metrics profile tests from #2865 to fail because the metrics profiler will likely be stopped in less than 15 milliseconds, meaning we'll see 0 duration between profile collection and log an error. This commit actually suppresses the error by checking whether the profiler was stopped (meaning interruptibleSleep was interrupted). If so, and if the metrics profiler returned an error, we instead return a sentinel error indicating that profiling was stopped. If we see that error, we just drop the profile and don't log an error. We won't upload the profile anyway. This way, we should only report an error from the metrics profiler if there is _actually_ a problem with the timer.
nsrip-dd
added a commit
that referenced
this pull request
Sep 24, 2024
The fix in #2865 was intended to suppress spurious metrics profile errors when the profiler is stopped. It did so by relaxing the one-second duration constraint of the metrics profiler. However, the Windows system timer resolution is about 15 milliseconds (see https://learn.microsoft.com/en-us/windows-hardware/drivers/kernel/high-resolution-timers#controlling-timer-accuracy) This caused the metrics profile tests from #2865 to fail because the metrics profiler will likely be stopped in less than 15 milliseconds, meaning we'll see 0 duration between profile collection and log an error. This commit actually suppresses the error by checking whether the profiler was stopped (meaning interruptibleSleep was interrupted). If so, and if the metrics profiler returned an error, we instead return a sentinel error indicating that profiling was stopped. If we see that error, we just drop the profile and don't log an error. We won't upload the profile anyway. This way, we should only report an error from the metrics profiler if there is _actually_ a problem with the timer.
nsrip-dd
added a commit
that referenced
this pull request
Sep 25, 2024
The fix in #2865 was intended to suppress spurious metrics profile errors when the profiler is stopped. It did so by relaxing the one-second duration constraint of the metrics profiler. However, the Windows system timer resolution is about 15 milliseconds (see https://learn.microsoft.com/en-us/windows-hardware/drivers/kernel/high-resolution-timers#controlling-timer-accuracy) This caused the metrics profile tests from #2865 to fail because the metrics profiler will likely be stopped in less than 15 milliseconds, meaning we'll see 0 duration between profile collection and log an error. This commit actually suppresses the error by checking whether the profiler was stopped (meaning interruptibleSleep was interrupted). If so, and if the metrics profiler returned an error, we instead return a sentinel error indicating that profiling was stopped. If we see that error, we just drop the profile and don't log an error. We won't upload the profile anyway. This way, we should only report an error from the metrics profiler if there is _actually_ a problem with the timer.
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
2 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.
What does this PR do?
The metrics profiler insisted on at least one second between collections
for two reasons:
convert a time.Duration to seconds, which will truncate to 0, as a
ratio in a subsequent computation
The profiler would report an error if less than one second elapsed
between collections. In practice, this resulted in misleading error logs
because it's entirely likely for profiling to be stopped less than a
second after the last profile collection.
The restriction was not really even needed. For one, we can just do
floating-point division rather than integer division to avoid the
truncation problem.
Also, Go has had monotonic time support by default since 2017, added in
Go 1.9, and time comparison operations including time.Time.Sub, work
with respect to monotonic time. We shouldn't have any issues with
negative periods. We can ensure the period is positive just as a
defensive measure, and fail if it's negative since this may indicate a
bug in the Go runtime if it's violating the monotonicity guarantees.
Motivation
Reduce log noise. This log has been misleading in past escalations, with
users/support assuming the error was relevant to the actual issue being
investigated.
Fixes #2863