[State Sync v2] Update alerts and metrics. (aptos-labs#676)

* [Alerts] Remove SSv1 alerts. * [Metrics] Keep consistency with existing metric names. * [Dashboards] Update SSv2 dashboard. Co-authored-by: Josh Lind <[email protected]>
pelyped · Apr 27, 2022 · 5de662d · 5de662d
1 parent 95290bd
commit 5de662d
Show file tree

Hide file tree

Showing 9 changed files with 1,523 additions and 1,548 deletions.
diff --git a/state-sync/state-sync-v2/state-sync-driver/src/driver.rs b/state-sync/state-sync-v2/state-sync-driver/src/driver.rs
@@ -290,7 +290,7 @@ impl<
 
         let operations = [
             metrics::StorageSynchronizerOperations::ExecutedTransactions,
-            metrics::StorageSynchronizerOperations::SyncedTransactions,
+            metrics::StorageSynchronizerOperations::Synced,
         ];
         for operation in operations {
             metrics::increment_gauge(

diff --git a/state-sync/state-sync-v2/state-sync-driver/src/metrics.rs b/state-sync/state-sync-v2/state-sync-driver/src/metrics.rs
@@ -14,8 +14,8 @@ pub const STORAGE_SYNCHRONIZER_PENDING_DATA: &str = "storage_synchronizer_pendin
 pub enum StorageSynchronizerOperations {
     AppliedTransactionOutputs, // Applied a chunk of transactions outputs.
     ExecutedTransactions,      // Executed a chunk of transactions.
+    Synced,                    // Wrote a chunk of transactions and outputs to storage.
     SyncedAccounts,            // Wrote a chunk of accounts to storage.
-    SyncedTransactions,        // Wrote a chunk of transactions and outputs to storage.
 }
 
 impl StorageSynchronizerOperations {
@@ -25,8 +25,8 @@ impl StorageSynchronizerOperations {
                 "applied_transaction_outputs"
             }
             StorageSynchronizerOperations::ExecutedTransactions => "executed_transactions",
+            StorageSynchronizerOperations::Synced => "synced",
             StorageSynchronizerOperations::SyncedAccounts => "synced_accounts",
-            StorageSynchronizerOperations::SyncedTransactions => "synced_transactions",
         }
     }
 }

diff --git a/state-sync/state-sync-v2/state-sync-driver/src/storage_synchronizer.rs b/state-sync/state-sync-v2/state-sync-driver/src/storage_synchronizer.rs
@@ -443,7 +443,7 @@ fn spawn_committer<
                             );
                             metrics::increment_gauge(
                                 &metrics::STORAGE_SYNCHRONIZER_OPERATIONS,
-                                metrics::StorageSynchronizerOperations::SyncedTransactions
+                                metrics::StorageSynchronizerOperations::Synced
                                     .get_label(),
                                 transactions.len() as u64,
                             );

diff --git a/state-sync/state-sync-v2/state-sync-driver/src/utils.rs b/state-sync/state-sync-v2/state-sync-driver/src/utils.rs
@@ -222,7 +222,7 @@ pub fn initialize_sync_version_gauges(storage: Arc<dyn DbReader>) -> Result<(),
     let metrics = [
         metrics::StorageSynchronizerOperations::AppliedTransactionOutputs,
         metrics::StorageSynchronizerOperations::ExecutedTransactions,
-        metrics::StorageSynchronizerOperations::SyncedTransactions,
+        metrics::StorageSynchronizerOperations::Synced,
     ];
 
     for metric in metrics {

diff --git a/terraform/fullnode/aws/fullnode/files/rules/fullnode_cluster_alerts.yml b/terraform/fullnode/aws/fullnode/files/rules/fullnode_cluster_alerts.yml
@@ -35,12 +35,6 @@ groups:
       severity: error
       summary: "State sync is not making progress (i.e., it is not keeping up with the head of the blockchain)"
     annotations:
-  - alert: State sync has no active upstream peers
-    expr: (sum by (owner, kubernetes_pod_name) (aptos_state_sync_active_upstream_peers)) == 0
-    for: 3m
-    labels:
-      severity: error
-      summary: "State sync has no active upstream peers (i.e., it has no peers to synchronize from!)"
   # lower severity alert for monitoring
   - alert: Monitoring pod crashing
     expr: count(time() - container_start_time_seconds{container!="", container!="POD", pod=~".*monitoring.*"} < 60)