Add criterion crate and benchmarks for some metric types #3194
Conversation
This adds some very basic benchmarks to some of the labeled/dual-labeled metrics. You can run these locally using `cargo bench`. To run a specific set of benches: `cargo bench --bench dual_labeled_counter`
travis79
force-pushed
the
marking-some-benches
branch
from
b8ac836 to
57e4292
Compare
travis79
changed the title
There was a problem hiding this comment.
These benchmarks don't actually test what you intended.
You're using the glean-core types, but calling the dispatched methods.
Because Glean is never initialized, the dispatcher will never actually run and thus the majority of the work won't be executed.
I think it would make more sense to move those benchmarks into the RLB and use the glean::private::* types after initializing Glean.
And then we probably also need at least one call to test_get_value at the end of each benchmark to ensure the dispatcher is actually going through all tasks, otherwise all we measure is how fast we can put things on to the dispatcher queue.
| c.bench_function("dual_labeled_counter_get_and_add", |b| { | ||
| b.iter(|| { | ||
| for (k, c) in &keys { | ||
| let m = black_box(metric.get(k, c)); |
There was a problem hiding this comment.
This black_box doesn't make sense. You're using the return value, so it won't be eliminated anyway.
There was a problem hiding this comment.
Ahh, I was using it under the assumption that it would encourage the compiler to not optimize this. I'll rework this and move things to the RLB as you suggest. Thanks for the pointers here, benchmarking things like this is new territory for me :)
This adds some very basic benchmarks to some of the labeled/dual-labeled metrics.
You can run these locally using
cargo bench. To run a specific set of benches:cargo bench --bench dual_labeled_counter