Slack 19.0 backport 16997 (#568)

* PRS and ERS don't promote replicas taking backups (vitessio#16997)

Signed-off-by: Eduardo J. Ortega U <[email protected]>

go/vt/vtctl/grpcvtctldserver/testutil/test_tmclient.go

@@ -184,6 +184,8 @@ type TabletManagerClient struct {
 		EventJitter   time.Duration
 		ErrorAfter    time.Duration
 	}
+	// Backing Up - keyed by tablet alias.
+	TabletsBackupState map[string]bool
 	// keyed by tablet alias.
 	ChangeTabletTypeResult map[string]error
 	ChangeTabletTypeDelays map[string]time.Duration
@@ -918,6 +920,9 @@ func (fake *TabletManagerClient) ReplicationStatus(ctx context.Context, tablet *
 	}
 
 	if result, ok := fake.ReplicationStatusResults[key]; ok {
+		if _, ok = fake.TabletsBackupState[key]; ok {
+			result.Position.BackupRunning = fake.TabletsBackupState[key]
+		}
 		return result.Position, result.Error
 	}
 

go/vt/vtctl/reparentutil/emergency_reparenter.go

@@ -259,7 +259,7 @@ func (erp *EmergencyReparenter) reparentShardLocked(ctx context.Context, ev *eve
 	// 2. Remove the tablets with the Must_not promote rule
 	// 3. Remove cross-cell tablets if PreventCrossCellPromotion is specified
 	// Our final primary candidate MUST belong to this list of valid candidates
-	validCandidateTablets, err = erp.filterValidCandidates(validCandidateTablets, stoppedReplicationSnapshot.reachableTablets, prevPrimary, opts)
+	validCandidateTablets, err = erp.filterValidCandidates(validCandidateTablets, stoppedReplicationSnapshot.reachableTablets, stoppedReplicationSnapshot.tabletsBackupState, prevPrimary, opts)
 	if err != nil {
 		return err
 	}
@@ -740,9 +740,12 @@ func (erp *EmergencyReparenter) promoteNewPrimary(
 	return nil
 }
 
-// filterValidCandidates filters valid tablets, keeping only the ones which can successfully be promoted without any constraint failures and can make forward progress on being promoted
-func (erp *EmergencyReparenter) filterValidCandidates(validTablets []*topodatapb.Tablet, tabletsReachable []*topodatapb.Tablet, prevPrimary *topodatapb.Tablet, opts EmergencyReparentOptions) ([]*topodatapb.Tablet, error) {
+// filterValidCandidates filters valid tablets, keeping only the ones which can successfully be promoted without any
+// constraint failures and can make forward progress on being promoted. It will filter out candidates taking backups
+// if possible.
+func (erp *EmergencyReparenter) filterValidCandidates(validTablets []*topodatapb.Tablet, tabletsReachable []*topodatapb.Tablet, tabletsBackupState map[string]bool, prevPrimary *topodatapb.Tablet, opts EmergencyReparentOptions) ([]*topodatapb.Tablet, error) {
 	var restrictedValidTablets []*topodatapb.Tablet
+	var notPreferredValidTablets []*topodatapb.Tablet
 	for _, tablet := range validTablets {
 		tabletAliasStr := topoproto.TabletAliasString(tablet.Alias)
 		// Remove tablets which have MustNot promote rule since they must never be promoted
@@ -769,7 +772,18 @@ func (erp *EmergencyReparenter) filterValidCandidates(validTablets []*topodatapb
 			}
 			continue
 		}
-		restrictedValidTablets = append(restrictedValidTablets, tablet)
+		// Put candidates that are running a backup in a separate list
+		backingUp, ok := tabletsBackupState[tabletAliasStr]
+		if ok && backingUp {
+			erp.logger.Infof("Setting %s in list of valid candidates taking a backup", tabletAliasStr)
+			notPreferredValidTablets = append(notPreferredValidTablets, tablet)
+		} else {
+			restrictedValidTablets = append(restrictedValidTablets, tablet)
+		}
+	}
+	if len(restrictedValidTablets) > 0 {
+		return restrictedValidTablets, nil
 	}
-	return restrictedValidTablets, nil
+
+	return notPreferredValidTablets, nil
 }