Overlay fix for transient IP reuse

drivers/overlay/ov_network.go

@@ -769,7 +769,7 @@ func (n *network) watchMiss(nlSock *nl.NetlinkSocket) {
 					continue
 				}
 				n.driver.peerAdd(n.id, "dummy", ip, IPmask, mac, vtep, l2Miss, l3Miss, false)
-			} else if l3Miss && time.Since(t) > 500*time.Millisecond {
+			} else if l3Miss && time.Since(t) > time.Second {
 				// All the local peers will trigger a miss notification but this one is expected and the local container will reply
 				// autonomously to the ARP request
 				// In case the gc_thresh3 values is low kernel might reject new entries during peerAdd. This will trigger the following
@@ -778,9 +778,9 @@ func (n *network) watchMiss(nlSock *nl.NetlinkSocket) {
 				// Entries which are marked as permanent are never deleted by the garbage-collector.
 				// The time limit here is to guarantee that the dbSearch is not
 				// done too frequently causing a stall of the peerDB operations.
-				t = time.Now()
 				pKey, pEntry, err := n.driver.peerDbSearch(n.id, ip)
-				if !pEntry.isLocal {
+				if err == nil && !pEntry.isLocal {
+					t = time.Now()
 					logrus.Warnf("miss notification for peer:%+v l3Miss:%t l2Miss:%t, if the problem persist check the gc_thresh on the host pKey:%+v pEntry:%+v err:%v",
 						neigh, l3Miss, l2Miss, *pKey, *pEntry, err)
 				}

drivers/overlay/peerdb.go

@@ -27,32 +27,29 @@ type peerEntry struct {
 }
 
 func (p *peerEntry) MarshalDB() peerEntryDB {
-	ones, bits := p.peerIPMask.Size()
 	return peerEntryDB{
-		eid:            p.eid,
-		vtep:           p.vtep.String(),
-		isLocal:        p.isLocal,
-		peerIPMaskOnes: ones,
-		peerIPMaskBits: bits,
+		eid:        p.eid,
+		vtep:       p.vtep.String(),
+		peerIPMask: p.peerIPMask.String(),
+		isLocal:    p.isLocal,
 	}
 }
 
 // This the structure saved into the set (SetMatrix), due to the implementation of it
 // the value inserted in the set has to be Hashable so the []byte had to be converted into
 // strings
 type peerEntryDB struct {
-	eid            string
-	vtep           string
-	peerIPMaskOnes int
-	peerIPMaskBits int
-	isLocal        bool
+	eid        string
+	vtep       string
+	peerIPMask string
+	isLocal    bool
 }
 
 func (p *peerEntryDB) UnMarshalDB() peerEntry {
 	return peerEntry{
 		eid:        p.eid,
 		vtep:       net.ParseIP(p.vtep),
-		peerIPMask: net.CIDRMask(p.peerIPMaskOnes, p.peerIPMaskBits),
+		peerIPMask: net.IPMask(net.ParseIP(p.peerIPMask)),
 		isLocal:    p.isLocal,
 	}
 }
@@ -454,19 +451,22 @@ func (d *driver) peerDeleteOp(nid, eid string, peerIP net.IP, peerIPMask net.IPM
 		logrus.Warn(err)
 	}
 
-	// Remove fdb entry to the bridge for the peer mac
-	if err := sbox.DeleteNeighbor(vtep, peerMac, true); err != nil {
-		if _, ok := err.(osl.NeighborSearchError); ok && dbEntries > 0 {
-			// We fall in here if there is a transient state and if the neighbor that is being deleted
-			// was never been configured into the kernel (we allow only 1 configuration at the time per <ip,mac> mapping)
-			return nil
+	// Local peers do not have any local configuration to delete
+	if !localPeer {
+		// Remove fdb entry to the bridge for the peer mac
+		if err := sbox.DeleteNeighbor(vtep, peerMac, true); err != nil {
+			if _, ok := err.(osl.NeighborSearchError); ok && dbEntries > 0 {
+				// We fall in here if there is a transient state and if the neighbor that is being deleted
+				// was never been configured into the kernel (we allow only 1 configuration at the time per <ip,mac> mapping)
+				return nil
+			}
+			return fmt.Errorf("could not delete fdb entry for nid:%s eid:%s into the sandbox:%v", nid, eid, err)
 		}
-		return fmt.Errorf("could not delete fdb entry for nid:%s eid:%s into the sandbox:%v", nid, eid, err)
-	}
 
-	// Delete neighbor entry for the peer IP
-	if err := sbox.DeleteNeighbor(peerIP, peerMac, true); err != nil {
-		return fmt.Errorf("could not delete neighbor entry for nid:%s eid:%s into the sandbox:%v", nid, eid, err)
+		// Delete neighbor entry for the peer IP
+		if err := sbox.DeleteNeighbor(peerIP, peerMac, true); err != nil {
+			return fmt.Errorf("could not delete neighbor entry for nid:%s eid:%s into the sandbox:%v", nid, eid, err)
+		}
 	}
 
 	if dbEntries == 0 {

networkdb/networkdb.go

@@ -505,7 +505,10 @@ func (nDB *NetworkDB) deleteNodeNetworkEntries(nid, node string) {
 				nDB.deleteEntry(nid, tname, key)
 			}
 
-			nDB.broadcaster.Write(makeEvent(opDelete, tname, nid, key, entry.value))
+			// Notify to the upper layer only entries not already marked for deletion
+			if !oldEntry.deleting {
+				nDB.broadcaster.Write(makeEvent(opDelete, tname, nid, key, entry.value))
+			}
 			return false
 		})
 }