close tikv#5753 fix unsafe recovery auto detect mode

Signed-off-by: Connor1996 <zbk602423539@gmail.com>
Connor1996 · Dec 5, 2022 · 82fa664 · 82fa664
1 parent 8158054
commit 82fa664
Show file tree

Hide file tree

Showing 2 changed files with 56 additions and 0 deletions.
diff --git a/server/cluster/unsafe_recovery_controller.go b/server/cluster/unsafe_recovery_controller.go
@@ -1000,6 +1000,21 @@ func (u *unsafeRecoveryController) generateForceLeaderPlan(newestRegionTree *reg
 					storeRecoveryPlan.ForceLeader.FailedStores = append(storeRecoveryPlan.ForceLeader.FailedStores, store)
 				}
 			}
+			if u.autoDetect {
+				// For auto detect, the failedStores is empty. So need to add the detected failed store to the list
+				for _, peer := range u.getFailedPeers(leader.Region()) {
+					found := false
+					for _, store := range storeRecoveryPlan.ForceLeader.FailedStores {
+						if store == peer.StoreId {
+							found = true
+							break
+						}
+					}
+					if !found {
+						storeRecoveryPlan.ForceLeader.FailedStores = append(storeRecoveryPlan.ForceLeader.FailedStores, peer.StoreId)
+					}
+				}
+			}
 			storeRecoveryPlan.ForceLeader.EnterForceLeaders = append(storeRecoveryPlan.ForceLeader.EnterForceLeaders, region.GetId())
 			u.recordAffectedRegion(leader.Region())
 			hasPlan = true

diff --git a/server/cluster/unsafe_recovery_controller_test.go b/server/cluster/unsafe_recovery_controller_test.go
@@ -40,6 +40,44 @@ func newStoreHeartbeat(storeID uint64, report *pdpb.StoreReport) *pdpb.StoreHear
 	}
 }
 
+func hasQuorum(region *metapb.Region, failedStores []uint64) bool {
+	hasQuorum := func(voters []*metapb.Peer) bool {
+		numFailedVoters := 0
+		numLiveVoters := 0
+
+		for _, voter := range voters {
+			found := false
+			for _, store := range failedStores {
+				if store == voter.GetStoreId() {
+					found = true
+					break
+				}
+			}
+			if found {
+				numFailedVoters += 1
+			} else {
+				numLiveVoters += 1
+			}
+		}
+		return numFailedVoters < numLiveVoters
+	}
+
+	// consider joint consensus
+	var incomingVoters []*metapb.Peer
+	var outgoingVoters []*metapb.Peer
+
+	for _, peer := range region.Peers {
+		if peer.Role == metapb.PeerRole_Voter || peer.Role == metapb.PeerRole_IncomingVoter {
+			incomingVoters = append(incomingVoters, peer)
+		}
+		if peer.Role == metapb.PeerRole_Voter || peer.Role == metapb.PeerRole_DemotingVoter {
+			outgoingVoters = append(outgoingVoters, peer)
+		}
+	}
+
+	return hasQuorum(incomingVoters) && hasQuorum(outgoingVoters)
+}
+
 func applyRecoveryPlan(re *require.Assertions, storeID uint64, storeReports map[uint64]*pdpb.StoreReport, resp *pdpb.StoreHeartbeatResponse) {
 	plan := resp.GetRecoveryPlan()
 	if plan == nil {
@@ -55,6 +93,9 @@ func applyRecoveryPlan(re *require.Assertions, storeID uint64, storeReports map[
 			for _, report := range reports.PeerReports {
 				region := report.GetRegionState().GetRegion()
 				if region.GetId() == forceLeader {
+					if hasQuorum(region, forceLeaders.GetFailedStores()) {
+						re.FailNow("should not enter force leader when quorum is still alive")
+					}
 					report.IsForceLeader = true
 					break
 				}