Merge cockroachdb#131038

131038: crosscluster/physical: use correct lag replan cluster setting r=[azhu-crl](https://github.com/azhu-crl) a=msbutler Previously the `stream_replication.replan_flow_frequency` setting determined the frequency that the frontier processor checked for lagging nodes, not the `stream_replication.lag_check_frequency` setting, as intended. The latter didn't control anything. This patch fixes this. Epic: none Release note: none Co-authored-by: Michael Butler <[email protected]>
cthumuluru-crdb · Sep 21, 2024 · 83589fb · 83589fb
2 parents 3d2f17c + 94730d7
commit 83589fb
Showing 1 changed file with 1 addition and 5 deletions.
diff --git a/pkg/ccl/crosscluster/physical/stream_ingestion_frontier_processor.go b/pkg/ccl/crosscluster/physical/stream_ingestion_frontier_processor.go
@@ -435,11 +435,7 @@ func (sf *streamIngestionFrontier) maybePersistFrontierEntries() error {
 
 func (sf *streamIngestionFrontier) maybeCheckForLaggingNodes() error {
 	ctx := sf.Ctx()
-
-	// We halve the frequency relative to the ReplanFrequency setting (i.e.
-	// check twice as often), because the node lag checker will only restart the
-	// distSQL plan if a node is lagging for 2 checks in a row.
-	checkFreq := crosscluster.ReplanFrequency.Get(&sf.FlowCtx.Cfg.Settings.SV) / 2
+	checkFreq := crosscluster.LagCheckFrequency.Get(&sf.FlowCtx.Cfg.Settings.SV)
 	maxLag := crosscluster.InterNodeLag.Get(&sf.FlowCtx.Cfg.Settings.SV)
 	if sf.persistedReplicatedTime.IsEmpty() {
 		log.VEvent(ctx, 2, "skipping lag replanning check: no persisted replicated time")