Merge pull request #49342 from robmry/debug_flaky_unsol_na

thaJeztah · web-flow · commit 5fd7ed2937a2 · 2025-01-28T12:14:38.000+01:00
Debug flaky unsolicited Neighbour Advertisements
diff --git a/integration/networking/bridge_linux_test.go b/integration/networking/bridge_linux_test.go
@@ -1536,6 +1536,9 @@ func TestAdvertiseAddresses(t *testing.T) {
 
 			icmps := stopICMP6Listen()
 			checkPkts("ICMP6", icmps, netip.MustParseAddr(ctr2Addr6), network.UnpackUnsolNA)
+			if t.Failed() {
+				d.TailLogsT(t, 100)
+			}
 		})
 	}
 }
diff --git a/libnetwork/osl/interface_linux.go b/libnetwork/osl/interface_linux.go
@@ -499,9 +499,37 @@ func (n *Namespace) advertiseAddrs(ctx context.Context, ifIndex int, i *Interfac
 			}
 		}
 		if naSender != nil {
-			if err := naSender.Send(); err != nil {
-				log.G(ctx).WithError(err).Warn("Failed to send unsolicited NA")
-				errs = append(errs, err)
+			// FIXME(robmry) - retry if this fails, but still return the error.
+			// In CI, the send has failed a couple of times with "write ip ::1->ff02::1: sendmsg: network is unreachable".
+			// Can't repro locally, so - try find out whether a retry helps and it's something racing, or it's a
+			// persistent problem.
+			for c := range 3 {
+				if c > 0 {
+					time.Sleep(50 * time.Millisecond)
+				}
+
+				routes, rgErr := nlh.RouteGetWithOptions(net.IPv6linklocalallnodes, &netlink.RouteGetOptions{
+					IifIndex: ifIndex,
+					SrcAddr:  net.IPv6loopback,
+				})
+				var routeStr string
+				if rgErr != nil {
+					routeStr = fmt.Sprintf("RouteGet->'%s'", rgErr.Error())
+				} else if len(routes) != 1 {
+					routeStr = fmt.Sprintf("RouteGet->%d routes", len(routes))
+				} else {
+					routeStr = fmt.Sprintf("RouteGet->'%s'", routes[0].String())
+				}
+
+				if err := naSender.Send(); err != nil {
+					log.G(ctx).WithError(err).Warn("Failed to send unsolicited NA")
+					errs = append(errs, fmt.Errorf("%s: %w", routeStr, err))
+					continue
+				}
+				if c > 0 {
+					errs = append(errs, fmt.Errorf("success"))
+				}
+				break
 			}
 		}
 		return errors.Join(errs...)

Original file line number	Diff line number	Diff line change
`@@ -1536,6 +1536,9 @@ func TestAdvertiseAddresses(t *testing.T) {`
`1536`	`1536`
`1537`	`1537`	`icmps := stopICMP6Listen()`
`1538`	`1538`	`checkPkts("ICMP6", icmps, netip.MustParseAddr(ctr2Addr6), network.UnpackUnsolNA)`
	`1539`	`+ if t.Failed() {`
	`1540`	`+ d.TailLogsT(t, 100)`
	`1541`	`+ }`
`1539`	`1542`	`})`
`1540`	`1543`	`}`
`1541`	`1544`	`}`