tree c04d476c4e20c8d02e9f2aaffdb7d631cdc6ce6f
parent 79350c98f2cee15490cfc8afb1ac35e640f29396
author Ryan Savitski <rsavitski@google.com> 1658842910 +0100
committer Ryan Savitski <rsavitski@google.com> 1658842910 +0100

traced_perf: process sharding (choosing random processes for callstacks)

The config will specify N shards for one or more data sources, we need
to choose a random value [0,N) to use as the shard for all those data
sources. We know which sources share a session (i.e. a config) via
tracing_session_id as supplied by traced.

Instead of introducing session-scoped data and linking that to the
lifetime of data sources, I've opted to simply replicate this state in
each DataSourceState. The extra plumbing through EventConfig::Create is
a bit clumsy, but ultimately TargetFilter is a logical place to host
this state.

Doc: go/tperf-sharding
Bug: 223773242
Change-Id: I84982c819ef541735bf3c6821c30efed2122ae94