Merge pull request #9356 from hashicorp/b-bridgefp-upgrade-path

scheduler: enable upgrade path for bridge network finger print

scheduler/feasible.go

@@ -28,6 +28,21 @@ const (
 	FilterConstraintDevices                     = "missing devices"
 )
 
+var (
+	// predatesBridgeFingerprint returns true if the constraint matches a version
+	// of nomad that predates the addition of the bridge network finger-printer,
+	// which was added in Nomad v0.12
+	predatesBridgeFingerprint = mustBridgeConstraint()
+)
+
+func mustBridgeConstraint() version.Constraints {
+	versionC, err := version.NewConstraint("< 0.12")
+	if err != nil {
+		panic(err)
+	}
+	return versionC
+}
+
 // FeasibleIterator is used to iteratively yield nodes that
 // match feasibility constraints. The iterators may manage
 // some state for performance optimizations.
@@ -343,6 +358,18 @@ func (c *NetworkChecker) SetNetwork(network *structs.NetworkResource) {
 
 func (c *NetworkChecker) Feasible(option *structs.Node) bool {
 	if !c.hasNetwork(option) {
+
+		// special case - if the client is running a version older than 0.12 but
+		// the server is 0.12 or newer, we need to maintain an upgrade path for
+		// jobs looking for a bridge network that will not have been fingerprinted
+		// on the client (which was added in 0.12)
+		if c.networkMode == "bridge" {
+			sv, err := version.NewSemver(option.Attributes["nomad.version"])
+			if err == nil && predatesBridgeFingerprint.Check(sv) {
+				return true
+			}
+		}
+
 		c.ctx.Metrics().FilterNode(option, "missing network")
 		return false
 	}

scheduler/feasible_test.go

@@ -399,14 +399,19 @@ func TestCSIVolumeChecker(t *testing.T) {
 
 func TestNetworkChecker(t *testing.T) {
 	_, ctx := testContext(t)
-	nodes := []*structs.Node{
-		mock.Node(),
-		mock.Node(),
-		mock.Node(),
+
+	node := func(mode string) *structs.Node {
+		n := mock.Node()
+		n.NodeResources.Networks = append(n.NodeResources.Networks, &structs.NetworkResource{Mode: mode})
+		n.Attributes["nomad.version"] = "0.12.0" // mock version is 0.5.0
+		return n
+	}
+
+	nodes := []*structs.Node{
+		node("bridge"),
+		node("bridge"),
+		node("cni/mynet"),
 	}
-	nodes[0].NodeResources.Networks = append(nodes[0].NodeResources.Networks, &structs.NetworkResource{Mode: "bridge"})
-	nodes[1].NodeResources.Networks = append(nodes[1].NodeResources.Networks, &structs.NetworkResource{Mode: "bridge"})
-	nodes[2].NodeResources.Networks = append(nodes[2].NodeResources.Networks, &structs.NetworkResource{Mode: "cni/mynet"})
 
 	checker := NewNetworkChecker(ctx)
 	cases := []struct {
@@ -439,6 +444,36 @@ func TestNetworkChecker(t *testing.T) {
 	}
 }
 
+func TestNetworkChecker_bridge_upgrade_path(t *testing.T) {
+	_, ctx := testContext(t)
+
+	t.Run("older client", func(t *testing.T) {
+		// Create a client that is still on v0.11, which does not have the bridge
+		// network finger-printer (and thus no bridge network resource)
+		oldClient := mock.Node()
+		oldClient.Attributes["nomad.version"] = "0.11.0"
+
+		checker := NewNetworkChecker(ctx)
+		checker.SetNetwork(&structs.NetworkResource{Mode: "bridge"})
+
+		ok := checker.Feasible(oldClient)
+		require.True(t, ok)
+	})
+
+	t.Run("updated client", func(t *testing.T) {
+		// Create a client that is updated to 0.12, but did not detect a bridge
+		// network resource.
+		oldClient := mock.Node()
+		oldClient.Attributes["nomad.version"] = "0.12.0"
+
+		checker := NewNetworkChecker(ctx)
+		checker.SetNetwork(&structs.NetworkResource{Mode: "bridge"})
+
+		ok := checker.Feasible(oldClient)
+		require.False(t, ok)
+	})
+}
+
 func TestDriverChecker_DriverInfo(t *testing.T) {
 	_, ctx := testContext(t)
 	nodes := []*structs.Node{