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
Originally the runner took the approach of logging JSON perf events to a file, representing various events, and capturing CPU and memory usage at regular intervals.
In order to generate stat summaries and more structured data, #36 is introducing stats objects, which are saved wholesale at the end of a run.
We should merge the approaches so that:
perf events are still generated, but from the stats structure instead of explicit calls
perf events could be used to "stream" noteworthy events, raw block/cycle data, as well as intermediate summaries (e.g. per stage)
it may no longer be necessary to dump a large "stats" event at the end containing all raw and summarized data
Since summaries as a single object is useful for jq processing, maybe there should be a way to regenerate the summarized stats object easily from the raw event stream.
The text was updated successfully, but these errors were encountered:
Originally the runner took the approach of logging JSON perf events to a file, representing various events, and capturing CPU and memory usage at regular intervals.
In order to generate stat summaries and more structured data, #36 is introducing
stats
objects, which are saved wholesale at the end of a run.We should merge the approaches so that:
stats
structure instead of explicit callsSince summaries as a single object is useful for
jq
processing, maybe there should be a way to regenerate the summarized stats object easily from the raw event stream.The text was updated successfully, but these errors were encountered: