The node uses the GHC lightweight threading in a very extensive way -- our model nodes spawn 20+ threads.
The eventlogs, which we already collect, do have thread switching marks.
We could dramatically improve the high-level CPU spike cost attribution in locli's machine performance analyses, if we extend it to additionally ingest the eventlogs and deduce the per-thread runtime costs.
The node uses the GHC lightweight threading in a very extensive way -- our model nodes spawn 20+ threads.
The eventlogs, which we already collect, do have thread switching marks.
We could dramatically improve the high-level CPU spike cost attribution in
locli
's machine performance analyses, if we extend it to additionally ingest the eventlogs and deduce the per-thread runtime costs.