apollo_config_manager: add update error metric

Author: Itay-Tsabary-Starkware

crates/apollo_config_manager/src/config_manager_runner.rs

@@ -16,9 +16,11 @@ use notify::{Config as NotifyConfig, EventKind, RecommendedWatcher, RecursiveMod
 use serde_json::Value;
 use tokio::sync::mpsc;
 use tokio::time::{interval, Duration as TokioDuration, Interval};
+use tracing::{error, info};
+
+use crate::metrics::{register_metrics, CONFIG_MANAGER_UPDATE_ERRORS};
 
 const FS_EVENT_CHANNEL_CAPACITY: usize = 16;
-use tracing::{error, info};
 
 #[cfg(test)]
 #[path = "config_manager_runner_tests.rs"]
@@ -38,6 +40,8 @@ impl ComponentStarter for ConfigManagerRunner {
 
         info!("ConfigManagerRunner: starting filesystem watcher");
 
+        register_metrics();
+
         if self.config_manager_config.enable_config_updates {
             let update_interval = interval(TokioDuration::from_secs_f64(
                 self.config_manager_config.config_update_interval_secs,
@@ -102,7 +106,9 @@ impl ConfigManagerRunner {
                             }
                             _ => {}
                         },
-                        Err(e) => error!("ConfigManagerRunner: watcher error: {e}"),
+                        Err(e) => {
+                            error!("ConfigManagerRunner: watcher error: {e}");
+                        }
                     }
                 }
                 // Periodic tick
@@ -119,6 +125,7 @@ impl ConfigManagerRunner {
         &mut self,
     ) -> Result<NodeDynamicConfig, Box<dyn std::error::Error + Send + Sync>> {
         let config = load_and_validate_config(self.cli_args.clone(), false).map_err(|e| {
+            CONFIG_MANAGER_UPDATE_ERRORS.increment(1);
             error!("ConfigManagerRunner: failed to update config: {e}");
             e
         })?;

crates/apollo_config_manager/src/metrics.rs

@@ -6,6 +6,21 @@ use apollo_infra::metrics::{
     RemoteClientMetrics,
     RemoteServerMetrics,
 };
-use apollo_metrics::define_infra_metrics;
+use apollo_metrics::{define_infra_metrics, define_metrics};
 
 define_infra_metrics!(config_manager);
+
+define_metrics!(
+    ConfigManager => {
+        MetricCounter {
+            CONFIG_MANAGER_UPDATE_ERRORS,
+            "config_manager_update_errors",
+            "Number of config manager update errors (load/validate or set dynamic config)",
+            init = 0
+        },
+    },
+);
+
+pub(crate) fn register_metrics() {
+    CONFIG_MANAGER_UPDATE_ERRORS.register();
+}

crates/apollo_metrics/src/metrics.rs

@@ -26,6 +26,7 @@ pub enum MetricScope {
     Blockifier,
     ClassManager,
     Committer,
+    ConfigManager,
     Consensus,
     ConsensusManager,
     ConsensusOrchestrator,