Remove redundant dynamic metrics

Summary:
Remove the per-repository mononoke.app.initialization_time_millisecs FB303 metric. Repository initialization latency is already emitted with the repo name through the MononokeAppStats ODS3 metric.

Remove the mononoke.scs_server.method.<method>.completion_time_ms dynamic histogram. SCS request latency is already emitted as duration_ms through MononokeScsRequest ODS3 and in the mononoke_scs_server Scuba request record. Update the checked-in oncall guidance to direct latency investigations to those supported sources.

Reviewed By: RajivTS

Differential Revision: D115717579

fbshipit-source-id: 14887e2dc8c2d87bbc4fbd3dbe08224298d091af

Author: Clara Rull

eden/mononoke/cmdlib/mononoke_app/src/repos_manager.rs

@@ -69,11 +69,6 @@ fn repos_manager_concurrency() -> Result<usize> {
 
 define_stats! {
     prefix = "mononoke.app";
-    initialization_time_millisecs: dynamic_timeseries(
-        "initialization_time_millisecs.{}",
-        (reponame: String);
-        Average, Sum, Count
-    ),
     completion_duration_secs: timeseries(Average, Sum, Count),
     // Deep-shard repo load failures (config load or facet build) via add_repo,
     // the chokepoint every deep-shard load funnels through.
@@ -278,11 +273,6 @@ impl<Repo> MononokeReposManager<Repo> {
                         .with_context(|| format!("Failed to initialize repo '{repo_name}'"))?;
                     let n = completed.fetch_add(1, Ordering::Relaxed) + 1;
                     info!("Initialized repo: {} ({}/{})", &repo_name, n, total);
-                    STATS::initialization_time_millisecs.add_value(
-                        start.elapsed().as_millis().try_into().unwrap_or(i64::MAX),
-                        (repo_name.to_string(),),
-                    );
-
                     #[cfg(fbcode_build)]
                     let instrument = Instrument_MononokeAppStats::new();
                     #[cfg(fbcode_build)]

eden/mononoke/servers/scs/scs_methods/src/source_control_impl.rs

@@ -132,8 +132,6 @@ define_stats! {
     total_request_internal_failure_permille: timeseries(Average),
     total_request_invalid_permille: timeseries(Average),
 
-    // Duration per method
-    method_completion_time_ms: dynamic_quantile_stat("method.{}.completion_time_ms", (method: String); Average, Sum, Count; P 5, P 50, P 90; Duration::from_secs(60), Duration::from_secs(600), Duration::from_secs(3600)),
     total_method_requests:  dynamic_timeseries("method.{}.total_method_requests", (method: String); Rate, Sum),
     total_method_internal_failure:  dynamic_timeseries("method.{}.total_method_internal_failure", (method: String); Rate, Sum),
 
@@ -970,10 +968,6 @@ fn log_result<T: AddScubaResponse>(
     STATS::total_request_internal_failure_permille.add_value(internal_failure * 1000);
     STATS::total_request_invalid_permille.add_value(invalid_request * 1000);
     STATS::total_request_overloaded.add_value(overloaded);
-    STATS::method_completion_time_ms.add_value(
-        stats.completion_time.as_millis_unchecked() as i64,
-        (method.to_string(),),
-    );
 
     ctx.perf_counters().insert_perf_counters(&mut scuba);
 
@@ -1140,11 +1134,6 @@ fn log_stream_complete(
     STATS::total_request_internal_failure_permille.add_value(internal_failure * 1000);
     STATS::total_request_invalid_permille.add_value(invalid_request * 1000);
     STATS::total_request_overloaded.add_value(overloaded);
-    // Only accounts for the time to start the stream, not the overall time.
-    STATS::method_completion_time_ms.add_value(
-        initial_future_stats.completion_time.as_millis_unchecked() as i64,
-        (method.to_string(),),
-    );
 
     ctx.perf_counters().insert_perf_counters(&mut scuba);