kvserver: disable early ack for commit triggers

Early-acking splits, merges, etc, isn't useful. At best, it is
inconsequential but at worst it causes flakes because the split hasn't
actually completed by the time it is acked and so it isn't yet reflected
in the descriptor, whose in-memory copy is frequently accessed during
allocation decisions.

Release note: None

pkg/kv/kvserver/replica_application_cmd.go

@@ -153,7 +153,17 @@ func (c *replicatedCmd) CanAckBeforeApplication() bool {
 	// We don't try to ack async consensus writes before application because we
 	// know that there isn't a client waiting for the result.
 	req := c.proposal.Request
-	return req.IsIntentWrite() && !req.AsyncConsensus
+	if !req.IsIntentWrite() || req.AsyncConsensus {
+		return false
+	}
+	if et, ok := req.GetArg(roachpb.EndTxn); ok && et.(*roachpb.EndTxnRequest).InternalCommitTrigger != nil {
+		// Don't early-ack for commit triggers, just to keep things simple - the
+		// caller is reasonably expecting to be able to run another replication
+		// change right away, and some code paths pull the descriptor out of memory
+		// where it may not reflect the previous operation yet.
+		return false
+	}
+	return true
 }
 
 // AckSuccess implements the apply.CheckedCommand interface.