You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
If you would like to contribute a new async-profiler feature, I suggest you first describe the problem, provide real-life examples on how the proposed feature addresses this problem, and outline the implementation approach.
The current wall profile sends SIGVTALRM signals to all threads. Even if the minimum interval is limited, the CPU usage increases too high.
I see that Datadog uses a reservoir sampling algorithm to signal only some threads. However, such data is incomplete. To ensure that the data conforms to the actual situation as much as possible, the random rate can be reported at the same time to facilitate the user to calculate the expected value.
If the problem you're concerned about is the high overhead of wall-clock sampling with many threads, I have a better idea on how to deal with that. It'll be likely implemented in one of the next versions. Otherwise, I'm open to proposals, bug reports, contributions and pull requests.
Can a random sampling capacity (such as the datadog) be added to wall collector and an interface be added to obtain the sampling frequency?
The text was updated successfully, but these errors were encountered: