Merge #76144

76144: kvserver: enforce min duration for slow_replication_threshold r=erikgrinaker a=tbg We've seen recent problems (#75728) related to mis-configured cluster settings. Preemptively tighten the range of allowable values for the `kv.replica_circuit_breaker.slow_replication_threshold` setting. Touches #33007. Release note: None Co-authored-by: Tobias Grieger <[email protected]>
cockroachdb · Feb 11, 2022 · cd3458b · cd3458b
2 parents 18e4193 + 2d8cc40
commit cd3458b
Showing 1 changed file with 13 additions and 1 deletion.
diff --git a/pkg/kv/kvserver/replica_circuit_breaker.go b/pkg/kv/kvserver/replica_circuit_breaker.go
@@ -49,7 +49,19 @@ var replicaCircuitBreakerSlowReplicationThreshold = settings.RegisterPublicDurat
 	"kv.replica_circuit_breaker.slow_replication_threshold",
 	"duration after which slow proposals trip the per-Replica circuit breaker (zero duration disables breakers)",
 	defaultReplicaCircuitBreakerSlowReplicationThreshold,
-	settings.NonNegativeDuration,
+	func(d time.Duration) error {
+		// Setting the breaker duration too low could be very dangerous to cluster
+		// health (breaking things to the point where the cluster setting can't be
+		// changed), so enforce a sane minimum.
+		const min = 500 * time.Millisecond
+		if d == 0 {
+			return nil
+		}
+		if d <= min {
+			return errors.Errorf("must specify a minimum of %s", min)
+		}
+		return nil
+	},
 )
 
 // Telemetry counter to count number of trip events.