fix test

Signed-off-by: lhy1024 <[email protected]>

pkg/schedule/checker/rule_checker.go

@@ -463,6 +463,7 @@ loopFits:
 			// Isdisconnected is more strictly than IsUnhealthy.
 			if c.cluster.GetStore(p.GetStoreId()).IsDisconnected() {
 				hasUnhealthyFit = true
+				pinDownPeer = p
 				break loopFits
 			}
 		}
@@ -481,6 +482,10 @@ loopFits:
 			if isUnhealthyPeer(orphanPeer.GetId()) {
 				continue
 			}
+			// make sure the orphan peer is not in disconnect store.
+			if c.cluster.GetStore(orphanPeer.GetStoreId()).IsDisconnected() {
+				continue
+			}
 			// no consider witness in this path.
 			if pinDownPeer.GetIsWitness() || orphanPeer.GetIsWitness() {
 				continue
@@ -500,6 +505,8 @@ loopFits:
 					return operator.CreatePromoteLearnerOperatorAndRemovePeer("replace-down-peer-with-orphan-peer", c.cluster, region, orphanPeer, pinDownPeer)
 				case orphanPeerRole == metapb.PeerRole_Voter && destRole == metapb.PeerRole_Learner:
 					return operator.CreateDemoteLearnerOperatorAndRemovePeer("replace-down-peer-with-orphan-peer", c.cluster, region, orphanPeer, pinDownPeer)
+				case orphanPeerRole == metapb.PeerRole_Voter && destRole == metapb.PeerRole_Voter:
+					return operator.CreateRemovePeerOperator("remove-orphan-peer", c.cluster, 0, region, pinDownPeer.GetStoreId())
 				default:
 					// destRole should not same with orphanPeerRole. if role is same, it fit with orphanPeer should be better than now.
 					// destRole never be leader, so we not consider it.
@@ -522,10 +529,7 @@ loopFits:
 				ruleCheckerRemoveOrphanPeerCounter.Inc()
 				return operator.CreateRemovePeerOperator("remove-orphan-peer", c.cluster, 0, region, orphanPeer.StoreId)
 			}
-			// If the store is disconnected, we cannot be sure whether it is healthy.
-			if !c.cluster.GetStore(orphanPeer.GetStoreId()).IsDisconnected() {
-				hasHealthPeer = true
-			}
+			hasHealthPeer = true
 		}
 	}
 	ruleCheckerSkipRemoveOrphanPeerCounter.Inc()

server/cluster/cluster_test.go

@@ -2796,7 +2796,7 @@ func TestReplica(t *testing.T) {
 	re.NoError(tc.addLeaderRegion(2, 1, 2, 3, 4))
 	region = tc.GetRegion(2)
 	re.NoError(dispatchHeartbeat(co, region, stream))
-	region = waitRemovePeer(re, stream, region, 4)
+	region = waitRemovePeer(re, stream, region, 3) // store3 is down, we should remove it firstly.
 	re.NoError(dispatchHeartbeat(co, region, stream))
 	waitNoResponse(re, stream)