#56 exclude precommit and prevote for missing consecutive blocks alert

example-config.yml

@@ -76,6 +76,8 @@ default_alert_config:
   consecutive_enabled: yes
   # How many missed blocks should trigger a notification?
   consecutive_missed: 5
+  # The following option can be used to exclude cases where a validator submits prevote and precommit, but the proposer does not include it in the block
+  consecutive_missed_exclude_preactions: false
   # Consecutive Missed alert Pagerduty Severity
   consecutive_priority: critical
 

td2/types.go

@@ -257,6 +257,8 @@ type AlertConfig struct {
 	ConsecutivePriority string `yaml:"consecutive_priority"`
 	// Whether to alert on consecutive missed blocks
 	ConsecutiveAlerts *bool `yaml:"consecutive_enabled"`
+	// Whether to exclude prevote/precommit misses from consecutive miss counting
+	ConsecutiveMissExcludePreactions *bool `yaml:"consecutive_missed_exclude_preactions"`
 
 	// Window is how many blocks missed as a percentage of the slashing window to trigger an alert
 	Window *int `yaml:"percentage_missed"`

td2/types_test.go

@@ -13,6 +13,7 @@ func alertConfigsEqual(a, b AlertConfig) bool {
 		intPtrEqual(a.ConsecutiveMissed, b.ConsecutiveMissed) &&
 		a.ConsecutivePriority == b.ConsecutivePriority &&
 		boolPtrEqual(a.ConsecutiveAlerts, b.ConsecutiveAlerts) &&
+		boolPtrEqual(a.ConsecutiveMissExcludePreactions, b.ConsecutiveMissExcludePreactions) &&
 		intPtrEqual(a.Window, b.Window) &&
 		a.PercentagePriority == b.PercentagePriority &&
 		boolPtrEqual(a.PercentageAlerts, b.PercentageAlerts) &&
@@ -499,4 +500,3 @@ func TestFloatVal(t *testing.T) {
 		})
 	}
 }
-

td2/ws.go

@@ -127,8 +127,10 @@ func (cc *ChainConfig) WsRun() {
 					cc.lastBlockTime = time.Now()
 					info := getAlarms(cc.name)
 					cc.blocksResults = append([]int{int(signState)}, cc.blocksResults[:len(cc.blocksResults)-1]...)
+					excludePreactions := boolVal(cc.Alerts.ConsecutiveMissExcludePreactions)
 					if signState < 3 && cc.valInfo.Bonded {
-						warn := fmt.Sprintf("❌ warning      %s missed block %d on %s", cc.valInfo.Moniker, update.Height, cc.ChainId)
+						missTypes := []string{"block", "precommit", "prevote"}
+						warn := fmt.Sprintf("❌ warning      %s missed %s %d on %s", cc.valInfo.Moniker, missTypes[signState], update.Height, cc.ChainId)
 						info += warn + "\n"
 						cc.lastError = time.Now().UTC().String() + " " + info
 						l(warn)
@@ -141,11 +143,15 @@ func (cc *ChainConfig) WsRun() {
 					case StatusPrecommit:
 						cc.statPrecommitMiss += 1
 						cc.statTotalMiss += 1
-						cc.statConsecutiveMiss += 1
+						if !excludePreactions {
+							cc.statConsecutiveMiss += 1
+						}
 					case StatusPrevote:
 						cc.statPrevoteMiss += 1
 						cc.statTotalMiss += 1
-						cc.statConsecutiveMiss += 1
+						if !excludePreactions {
+							cc.statConsecutiveMiss += 1
+						}
 					case StatusSigned:
 						cc.statTotalSigns += 1
 						cc.statConsecutiveMiss = 0