CI fixes

test/cluster/cluster.go

@@ -75,7 +75,7 @@ func New(bc BootConfig, out io.Writer) *Cluster {
 func BuildFlynn(bc BootConfig, rootFS, commit string, merge bool, out io.Writer) (string, error) {
 	c := New(bc, out)
 	defer c.Shutdown()
-	return c.BuildFlynn(rootFS, commit, merge)
+	return c.BuildFlynn(rootFS, commit, merge, false)
 }
 
 func (c *Cluster) log(a ...interface{}) (int, error) {
@@ -86,7 +86,7 @@ func (c *Cluster) logf(f string, a ...interface{}) (int, error) {
 	return fmt.Fprintf(c.out, f, a...)
 }
 
-func (c *Cluster) BuildFlynn(rootFS, commit string, merge bool) (string, error) {
+func (c *Cluster) BuildFlynn(rootFS, commit string, merge bool, runTests bool) (string, error) {
 	c.log("Building Flynn...")
 
 	if err := c.setup(); err != nil {
@@ -122,9 +122,11 @@ func (c *Cluster) BuildFlynn(rootFS, commit string, merge bool) (string, error)
 		return build.Drive("hda").FS, fmt.Errorf("error running build script: %s", err)
 	}
 
-	if err := runUnitTests(build, c.out); err != nil {
-		build.Kill()
-		return build.Drive("hda").FS, fmt.Errorf("unit tests failed: %s", err)
+	if runTests {
+		if err := runUnitTests(build, c.out); err != nil {
+			build.Kill()
+			return build.Drive("hda").FS, fmt.Errorf("unit tests failed: %s", err)
+		}
 	}
 
 	if err := build.Shutdown(); err != nil {
@@ -500,6 +502,19 @@ func lookupUser(name string) (int, int, error) {
 }
 
 func (c *Cluster) DumpLogs(w io.Writer) {
+	done := make(chan struct{})
+	go func() {
+		c.dumpLogs(w)
+		close(done)
+	}()
+	select {
+	case <-done:
+	case <-time.After(60 * time.Second):
+		fmt.Fprintln(w, "failed to dump logs within 60 seconds, skipping")
+	}
+}
+
+func (c *Cluster) dumpLogs(w io.Writer) {
 	streams := &Streams{Stdout: w, Stderr: w}
 	run := func(inst *Instance, cmd string) error {
 		fmt.Fprint(w, "\n\n***** ***** ***** ***** ***** ***** ***** ***** ***** *****\n\n")

test/cluster/instance.go

@@ -62,7 +62,7 @@ func (v *VMManager) NewInstance(c *VMConfig) (*Instance, error) {
 	if err != nil {
 		return nil, err
 	}
-	inst.IP = inst.tap.RemoteIP.String()
+	inst.IP = inst.tap.IP.String()
 	return inst, nil
 }
 
@@ -258,7 +258,7 @@ func (i *Instance) dialSSH(stderr io.Writer) error {
 
 var sshAttempts = attempt.Strategy{
 	Min:   5,
-	Total: 5 * time.Minute,
+	Total: 30 * time.Second,
 	Delay: time.Second,
 }
 

test/cluster/net.go

@@ -176,20 +176,17 @@ func deleteTap(name string) error {
 }
 
 type Tap struct {
-	Name              string
-	LocalIP, RemoteIP net.IP
-	bridge            *Bridge
+	Name   string
+	IP     net.IP
+	bridge *Bridge
 }
 
 func (t *Tap) Close() error {
 	if err := deleteTap(t.Name); err != nil {
 		return err
 	}
-	if t.LocalIP != nil {
-		ipallocator.ReleaseIP(t.bridge.ipNet, t.LocalIP)
-	}
-	if t.RemoteIP != nil {
-		ipallocator.ReleaseIP(t.bridge.ipNet, t.RemoteIP)
+	if t.IP != nil {
+		ipallocator.ReleaseIP(t.bridge.ipNet, t.IP)
 	}
 	return nil
 }
@@ -205,7 +202,7 @@ iface eth0 inet static
 
 func (t *Tap) WriteInterfaceConfig(f io.Writer) error {
 	return ifaceConfig.Execute(f, map[string]string{
-		"Address": t.RemoteIP.String(),
+		"Address": t.IP.String(),
 		"Gateway": t.bridge.IP(),
 	})
 }
@@ -222,13 +219,7 @@ func (t *TapManager) NewTap(uid, gid int) (*Tap, error) {
 	}
 
 	var err error
-	tap.LocalIP, err = ipallocator.RequestIP(t.bridge.ipNet, nil)
-	if err != nil {
-		tap.Close()
-		return nil, err
-	}
-
-	tap.RemoteIP, err = ipallocator.RequestIP(t.bridge.ipNet, nil)
+	tap.IP, err = ipallocator.RequestIP(t.bridge.ipNet, nil)
 	if err != nil {
 		tap.Close()
 		return nil, err
@@ -239,10 +230,6 @@ func (t *TapManager) NewTap(uid, gid int) (*Tap, error) {
 		tap.Close()
 		return nil, err
 	}
-	if err := netlink.NetworkLinkAddIp(iface, tap.LocalIP, t.bridge.ipNet); err != nil {
-		tap.Close()
-		return nil, err
-	}
 	if err := netlink.NetworkLinkUp(iface); err != nil {
 		tap.Close()
 		return nil, err

test/main.go

@@ -77,7 +77,7 @@ func main() {
 	if flynnrc == "" {
 		var rootFS string
 		testCluster = cluster.New(args.BootConfig, os.Stdout)
-		rootFS, err = testCluster.BuildFlynn(args.RootFS, "origin/master", false)
+		rootFS, err = testCluster.BuildFlynn(args.RootFS, "origin/master", false, false)
 		if err != nil {
 			testCluster.Shutdown()
 			log.Println("could not build flynn: ", err)

test/runner/runner.go

@@ -259,7 +259,7 @@ func (r *Runner) build(b *Build) (err error) {
 		c.Shutdown()
 	}()
 
-	rootFS, err := c.BuildFlynn(r.rootFS, b.Commit, b.Merge)
+	rootFS, err := c.BuildFlynn(r.rootFS, b.Commit, b.Merge, true)
 	defer removeRootFS(rootFS)
 	if err != nil {
 		return fmt.Errorf("could not build flynn: %s", err)

test/scripts/debug-info.sh

@@ -1,7 +1,7 @@
 #!/bin/bash
 #
 # This script is run by the CI runner to collect debugging information
-# which will be printed if any tests fail. It currently runs no code.
+# which will be printed if any tests fail.
 
 memwatch() {
   interests=("$@")
@@ -24,3 +24,23 @@ memwatch() {
     sleep 1
   done
 }
+
+netdebug() {
+  while true; do
+    echo "===> $(date +%H:%M:%S.%3N)"
+    trace ifconfig eth0
+    trace route -n
+    trace arp -a
+    echo "==================="
+    echo
+    sleep 5
+  done
+}
+
+trace() {
+    local cmd=$@
+    echo "+ ${cmd}"
+    $cmd
+}
+
+netdebug