storage: allow backpressuring AddSSTable for splits

dt · dt · commit c02c03c9c005 · 2019-03-31T15:23:01.000Z
Nodes appear to getting swamped with AddSSTable requests when they are highly overlapping and thus very expensive.
In particular, splits are getting stuck waiting for the AddSSTable calls since they were not marked as OK to backpressure.

Release note (performance improvement): allow oversized ranges to split sooner.
diff --git a/pkg/storage/replica_backpressure.go b/pkg/storage/replica_backpressure.go
@@ -55,6 +55,7 @@ var backpressurableReqMethods = util.MakeFastIntSet(
 	int(roachpb.Increment),
 	int(roachpb.Delete),
 	int(roachpb.DeleteRange),
+	int(roachpb.AddSSTable),
 )
 
 // backpressurableSpans contains spans of keys where write backpressuring

Original file line number	Diff line number	Diff line change
`@@ -55,6 +55,7 @@ var backpressurableReqMethods = util.MakeFastIntSet(`
`55`	`55`	`int(roachpb.Increment),`
`56`	`56`	`int(roachpb.Delete),`
`57`	`57`	`int(roachpb.DeleteRange),`
	`58`	`+ int(roachpb.AddSSTable),`
`58`	`59`	`)`
`59`	`60`
`60`	`61`	`// backpressurableSpans contains spans of keys where write backpressuring`