EXC-1516: Add new metric for advance install code executions

dfinity · Feb 5, 2024 · 5634346 · 5634346
1 parent 2211428
commit 5634346
Show file tree

Hide file tree

Showing 3 changed files with 74 additions and 7 deletions.
diff --git a/rs/execution_environment/src/scheduler.rs b/rs/execution_environment/src/scheduler.rs
@@ -1606,8 +1606,10 @@ impl Scheduler for SchedulerImpl {
 
         // Subnet queues: execute long running install code call if present.
         {
-            let measurement_scope =
-                MeasurementScope::nested(&self.metrics.round_subnet_queue, &root_measurement_scope);
+            let measurement_scope = MeasurementScope::nested(
+                &self.metrics.round_advance_long_install_code,
+                &root_measurement_scope,
+            );
 
             let mut subnet_round_limits = scheduler_round_limits.subnet_round_limits();
             state = self.advance_long_running_install_code(

diff --git a/rs/execution_environment/src/scheduler/scheduler_metrics.rs b/rs/execution_environment/src/scheduler/scheduler_metrics.rs
@@ -70,6 +70,7 @@ pub(super) struct SchedulerMetrics {
     pub(super) round_consensus_queue: ScopedMetrics,
     pub(super) round_postponed_raw_rand_queue: ScopedMetrics,
     pub(super) round_subnet_queue: ScopedMetrics,
+    pub(super) round_advance_long_install_code: ScopedMetrics,
     pub(super) round_scheduling_duration: Histogram,
     pub(super) round_update_sign_with_ecdsa_contexts_duration: Histogram,
     pub(super) round_inner: ScopedMetrics,
@@ -404,6 +405,32 @@ impl SchedulerMetrics {
                     metrics_registry,
                 ),
             },
+            round_advance_long_install_code: ScopedMetrics {
+                duration: duration_histogram(
+                    "execution_round_advance_long_install_code_duration_seconds",
+                    "The duration of advancing an in progress long install code in \
+                          an execution round",
+                    metrics_registry,
+                ),
+                instructions: instructions_histogram(
+                    "execution_round_advance_long_install_code_instructions",
+                    "The number of instructions executed during advancing \
+                        an in progress install code in an execution round",
+                    metrics_registry,
+                ),
+                slices: slices_histogram(
+                    "execution_round_advance_long_install_code_slices",
+                    "The number of slices executed executed during advancing \
+                        an in progress install code in an execution round",
+                    metrics_registry,
+                ),
+                messages: messages_histogram(
+                    "execution_round_advance_long_install_code_messages",
+                    "The number of messages executed during advancing \
+                        an in progress install code in an execution round",
+                    metrics_registry,
+                ),
+            },
             round_scheduling_duration: duration_histogram(
                 "execution_round_scheduling_duration_seconds",
                 "The duration of execution round scheduling in seconds.",

diff --git a/rs/execution_environment/src/scheduler/tests.rs b/rs/execution_environment/src/scheduler/tests.rs
@@ -2502,16 +2502,30 @@ fn execution_round_metrics_are_recorded() {
         13
     );
     assert_eq!(test.state().metadata.subnet_metrics.num_canisters, 3);
-    assert_eq!(3, metrics.round_subnet_queue.duration.get_sample_count());
     assert_eq!(
-        3,
+        1,
+        metrics
+            .round_advance_long_install_code
+            .duration
+            .get_sample_count()
+    );
+    assert_eq!(
+        1,
+        metrics
+            .round_advance_long_install_code
+            .messages
+            .get_sample_count()
+    );
+    assert_eq!(2, metrics.round_subnet_queue.duration.get_sample_count());
+    assert_eq!(
+        2,
         metrics.round_subnet_queue.instructions.get_sample_count()
     );
     assert_eq!(
         30,
         metrics.round_subnet_queue.instructions.get_sample_sum() as u64,
     );
-    assert_eq!(3, metrics.round_subnet_queue.messages.get_sample_count());
+    assert_eq!(2, metrics.round_subnet_queue.messages.get_sample_count());
     assert_eq!(
         3,
         metrics.round_subnet_queue.messages.get_sample_sum() as u64,
@@ -4317,7 +4331,15 @@ fn dts_allow_only_one_long_install_code_execution_at_any_time() {
             .round_subnet_queue
             .slices
             .get_sample_sum(),
-        2.0
+        1.0
+    );
+    assert_eq!(
+        test.scheduler()
+            .metrics
+            .round_advance_long_install_code
+            .slices
+            .get_sample_sum(),
+        1.0
     );
     assert_eq!(
         test.scheduler()
@@ -4369,16 +4391,32 @@ fn dts_allow_only_one_long_install_code_execution_at_any_time() {
             .round_subnet_queue
             .slices
             .get_sample_sum(),
-        4.0
+        2.0
     );
     assert_eq!(
         test.scheduler()
             .metrics
             .round_subnet_queue
             .messages
             .get_sample_sum(),
+        1.0
+    );
+    assert_eq!(
+        test.scheduler()
+            .metrics
+            .round_advance_long_install_code
+            .slices
+            .get_sample_sum(),
         2.0
     );
+    assert_eq!(
+        test.scheduler()
+            .metrics
+            .round_advance_long_install_code
+            .messages
+            .get_sample_sum(),
+        1.0
+    );
 
     // Execute another round to refresh the metrics
     test.execute_round(ExecutionRoundType::OrdinaryRound);