Prevent PITR supervisor from restarting slicer during restore

The PITR supervisor loop could restart the oplog slicer during a restore
if it read the config before the restore set pitr.enabled=false. Add a
suspendPitr/resumePitr mechanism so the restore handler can prevent the
supervisor from restarting the slicer for the duration of the restore.

Author: jcechace

cmd/pbm-agent/agent.go

@@ -25,12 +25,13 @@ import (
 )
 
 type Agent struct {
-	leadConn connect.Client
-	nodeConn *mongo.Client
-	bcp      *currentBackup
-	pitrjob  *currentPitr
-	slicerMx sync.Mutex
-	bcpMx    sync.Mutex
+	leadConn      connect.Client
+	nodeConn      *mongo.Client
+	bcp           *currentBackup
+	pitrjob       *currentPitr
+	pitrSuspended atomic.Bool
+	slicerMx      sync.Mutex
+	bcpMx         sync.Mutex
 
 	brief topo.NodeBrief
 

cmd/pbm-agent/pitr.go

@@ -57,6 +57,19 @@ func (a *Agent) removePitr() {
 	a.setPitr(nil)
 }
 
+// suspendPitr stops the current slicer and prevents the PITR supervisor
+// from restarting it. Use resumePitr to allow restarts again.
+func (a *Agent) suspendPitr() {
+	a.pitrSuspended.Store(true)
+	a.removePitr()
+}
+
+// resumePitr allows the PITR supervisor to start the slicer again
+// after a previous suspendPitr call.
+func (a *Agent) resumePitr() {
+	a.pitrSuspended.Store(false)
+}
+
 func (a *Agent) getPitr() *currentPitr {
 	a.slicerMx.Lock()
 	defer a.slicerMx.Unlock()
@@ -185,6 +198,10 @@ func (a *Agent) pitr(ctx context.Context) error {
 	l := log.FromContext(ctx).NewEvent(string(ctrl.CmdPITR), "", "", ep.TS())
 	ctx = log.SetLogEventToContext(ctx, l)
 
+	if a.pitrSuspended.Load() {
+		return nil
+	}
+
 	if !cfg.PITR.Enabled {
 		a.removePitr()
 		a.stopMon()
@@ -358,6 +375,11 @@ func (a *Agent) pitr(ctx context.Context) error {
 						stopSlicing()
 						return
 					}
+					if a.pitrSuspended.Load() {
+						l.Debug("stop slicing: pitr suspended")
+						stopSlicing()
+						return
+					}
 
 				case <-stopSlicingCtx.Done():
 					return

cmd/pbm-agent/restore.go

@@ -85,7 +85,8 @@ func (a *Agent) Restore(ctx context.Context, r *ctrl.RestoreCmd, opid ctrl.OPID,
 		} else {
 			l.Info("oplog slicer disabled")
 		}
-		a.removePitr()
+		a.suspendPitr()
+		defer a.resumePitr()
 	}
 
 	// stop balancer during the restore