// TestTaskResurrectionInt ensures that a Claim won't recreate a task that had
// been deleted (marked as done). taskmgr has a non-integration version of this
// test.
func TestTaskResurrectionInt(t *testing.T) {
etcdc, hosts := testutil.NewEtcdClient(t)
t.Parallel()
etcdc.Delete("test-resurrect", recursive)
task := m_etcd.DefaultTaskFunc("xyz", "")
conf := m_etcd.NewConfig("testclient", "test-resurrect", hosts)
coord, err := m_etcd.NewEtcdCoordinator(conf)
if err != nil {
t.Fatalf("Error creating coordinator: %v", err)
}
if err := coord.Init(nil); err != nil {
t.Fatalf("Error initializing coordinator: %v", err)
}
// Try to claim a nonexistent
if claimed := coord.Claim(task); claimed {
t.Fatal("Claiming a nonexistent task should not work but did!")
}
// Create a task, mark it as done, and try to claim it again
client := m_etcd.NewClient("test-resurrect", hosts)
if err := client.SubmitTask(m_etcd.DefaultTaskFunc("xyz", "")); err != nil {
t.Fatalf("Error submitting task xyz: %v", err)
}
if claimed := coord.Claim(task); !claimed {
t.Fatal("Failed to claim task xyz")
}
coord.Done(task)
if claimed := coord.Claim(task); claimed {
t.Fatal("Reclaimed task that was marked as done.")
}
}
func main() {
mlvl := metafora.LogLevelInfo
hostname, _ := os.Hostname()
peers := flag.String("etcd", "http://127.0.0.1:2379", "comma delimited etcd peer list")
namespace := flag.String("namespace", "koalemos", "metafora namespace")
name := flag.String("name", hostname, "node name or empty for automatic")
loglvl := flag.String("log", mlvl.String(), "set log level: [debug], info, warn, error")
flag.Parse()
hosts := strings.Split(*peers, ",")
etcdc := etcd.NewClient(hosts)
switch strings.ToLower(*loglvl) {
case "debug":
mlvl = metafora.LogLevelDebug
case "info":
mlvl = metafora.LogLevelInfo
case "warn":
mlvl = metafora.LogLevelWarn
case "error":
mlvl = metafora.LogLevelError
default:
metafora.Warnf("Invalid log level %q - using %s", *loglvl, mlvl)
}
metafora.SetLogLevel(mlvl)
conf := m_etcd.NewConfig(*name, *namespace, hosts)
// Replace NewTask func with one that returns a *koalemos.Task
conf.NewTaskFunc = func(id, value string) metafora.Task {
t := koalemos.NewTask(id)
if value == "" {
return t
}
if err := json.Unmarshal([]byte(value), t); err != nil {
metafora.Errorf("Unable to unmarshal task %s: %v", t.ID(), err)
return nil
}
return t
}
hfunc := makeHandlerFunc(etcdc)
ec, err := m_etcd.NewEtcdCoordinator(conf)
if err != nil {
metafora.Errorf("Error creating etcd coordinator: %v", err)
}
bal := m_etcd.NewFairBalancer(conf)
c, err := metafora.NewConsumer(ec, hfunc, bal)
if err != nil {
metafora.Errorf("Error creating consumer: %v", err)
os.Exit(2)
}
metafora.Infof(
"Starting koalsmosd with etcd=%s; namespace=%s; name=%s; loglvl=%s",
*peers, conf.Namespace, conf.Name, mlvl)
consumerRunning := make(chan struct{})
go func() {
defer close(consumerRunning)
c.Run()
}()
sigC := make(chan os.Signal, 1)
signal.Notify(sigC, os.Interrupt, os.Kill, syscall.SIGTERM)
select {
case s := <-sigC:
metafora.Infof("Received signal %s, shutting down", s)
case <-consumerRunning:
metafora.Warn("Consumer exited. Shutting down.")
}
c.Shutdown()
metafora.Info("Shutdown")
}
// Start the grid. Actors that were stopped from a previous exit
// of the grid but returned a "done" status of false will start
// to be scheduled. New actors can be scheduled with StartActor.
func (g *grid) Start() (<-chan bool, error) {
g.mu.Lock()
defer g.mu.Unlock()
// Start only once.
if g.started {
return g.exit, nil
}
// Use the hostname as the node identifier.
hostname, err := os.Hostname()
if err != nil {
return nil, err
}
nodeid := fmt.Sprintf("%s-%s", hostname, g.name)
// Define the metafora new task function and config.
conf := m_etcd.NewConfig(nodeid, g.name, g.etcdservers)
conf.NewTaskFunc = func(id, value string) metafora.Task {
def := NewActorDef(id)
err := json.Unmarshal([]byte(value), def)
if err != nil {
log.Printf("error: failed to schedule actor: %v, error: %v", id, err)
return nil
}
a, err := g.maker.MakeActor(def)
if err != nil {
log.Printf("error: failed to schedule actor: %v, error: %v", id, err)
return nil
}
return newHandler(g.fork(), a)
}
// Create the metafora etcd coordinator.
ec, err := m_etcd.NewEtcdCoordinator(conf)
if err != nil {
return nil, err
}
// Create the metafora consumer.
c, err := metafora.NewConsumer(ec, handler(etcd.NewClient(g.etcdservers)), m_etcd.NewFairBalancer(conf))
if err != nil {
return nil, err
}
g.metaconsumer = c
g.metaclient = m_etcd.NewClient(g.name, g.etcdservers)
g.stopped = false
g.started = true
// Close the exit channel when metafora thinks
// an exit is needed.
go func() {
defer close(g.exit)
g.metaconsumer.Run()
}()
for i := 0; i < 2*runtime.NumCPU(); i++ {
natsconn, err := g.newNatsConn()
if err != nil {
return nil, err
}
g.natsconnpool[i] = natsconn
if i == 0 {
g.natsconn = g.natsconnpool[0]
}
}
return g.exit, nil
}
// TestSleepTest is an integration test for all of m_etcd's components.
//
func TestSleepTest(t *testing.T) {
etcdc, hosts := testutil.NewEtcdClient(t)
t.Parallel()
const namespace = "sleeptest-metafora"
const sleepingtasks = "sleeping-task1"
etcdc.Delete(namespace, recursive)
holdtask := make(chan bool)
h := func(task metafora.Task, cmds <-chan *statemachine.Message) *statemachine.Message {
if task.ID() == sleepingtasks {
sleeptil := 5 * time.Second
nextstarttime := (time.Now().Add(sleeptil))
t.Logf("sleeping task:%v sleepfor:%v", task, nextstarttime)
<-holdtask
return statemachine.SleepMessage(nextstarttime)
}
cmd := <-cmds
t.Logf("non sleeping task:%v", task)
return cmd
}
newC := func(name, ns string) *metafora.Consumer {
conf := m_etcd.NewConfig(name, ns, hosts)
coord, hf, bal, err := m_etcd.New(conf, h)
if err != nil {
t.Fatalf("Error creating new etcd stack: %v", err)
}
cons, err := metafora.NewConsumer(coord, hf, bal)
if err != nil {
t.Fatalf("Error creating consumer %s:%s: %v", ns, name, err)
}
go func() {
cons.Run()
t.Logf("Consumer:%s exited.", name)
}()
return cons
}
assertRunning := func(tid string, cons ...*metafora.Consumer) {
found := false
for _, c := range cons {
tasks := c.Tasks()
if len(tasks) > 0 && found {
t.Fatal("Task already found running but another task is running on a different consumer")
}
if len(tasks) > 1 {
t.Fatalf("Expected at most 1 task, but found: %d", len(tasks))
}
if len(tasks) == 1 && tasks[0].Task().ID() == tid {
found = true
}
}
if !found {
t.Fatalf("Could not find task=%q", tid)
}
}
// Start 2 consumers
cons1 := newC("node1", namespace)
cons2 := newC("node2", namespace)
// Create clients and start some tests
cliA := m_etcd.NewClient(namespace, hosts)
if err := cliA.SubmitTask(m_etcd.DefaultTaskFunc(sleepingtasks, "")); err != nil {
t.Fatalf("Error submitting task1 to a: %v", err)
}
// Give consumers a bit to pick up tasks
time.Sleep(500 * time.Millisecond)
assertRunning(sleepingtasks, cons1, cons2)
holdtask <- true
// Give consumers a bit to pick up tasks
time.Sleep(500 * time.Millisecond)
assertRunning(sleepingtasks, cons1, cons2) // not sure if this should be true or false.
wait1 := make(chan bool)
go func() {
defer close(wait1)
// Shutdown
cons1.Shutdown()
cons2.Shutdown()
}()
timeout := time.NewTimer(1 * time.Second)
select {
case <-wait1:
case <-timeout.C:
t.Fatalf("failed waiting for shutdown")
}
// make sure all tasks are released
for _, c := range []*metafora.Consumer{cons1, cons2} {
tasks := c.Tasks()
for _, work := range tasks {
t.Fatalf("work id %v is still running", work)
}
}
}
// TestAll is an integration test for all of m_etcd's components.
//
// While huge integration tests like this are rarely desirable as they can be
// overly fragile and complex, I found myself manually repeating the tests I've
// automated here over and over. This is far more reliable than expecting
// developers to do adhoc testing of all of the m_etcd package's features.
func TestAll(t *testing.T) {
etcdc, hosts := testutil.NewEtcdClient(t)
t.Parallel()
etcdc.Delete("test-a", recursive)
etcdc.Delete("test-b", recursive)
h := func(task metafora.Task, cmds <-chan *statemachine.Message) *statemachine.Message {
cmd := <-cmds
if task.ID() == "error-test" {
return statemachine.ErrorMessage(errors.New("error-test"))
}
return cmd
}
newC := func(name, ns string) *metafora.Consumer {
conf := m_etcd.NewConfig(name, ns, hosts)
conf.Name = name
coord, hf, bal, err := m_etcd.New(conf, h)
if err != nil {
t.Fatalf("Error creating new etcd stack: %v", err)
}
cons, err := metafora.NewConsumer(coord, hf, bal)
if err != nil {
t.Fatalf("Error creating consumer %s:%s: %v", ns, name, err)
}
go cons.Run()
return cons
}
// Start 4 consumers, 2 per namespace
cons1a := newC("node1", "test-a")
cons2a := newC("node2", "test-a")
cons1b := newC("node1", "test-b")
cons2b := newC("node2", "test-b")
// Create clients and start some tests
cliA := m_etcd.NewClient("test-a", hosts)
cliB := m_etcd.NewClient("test-b", hosts)
if err := cliA.SubmitTask(m_etcd.DefaultTaskFunc("task1", "")); err != nil {
t.Fatalf("Error submitting task1 to a: %v", err)
}
if err := cliB.SubmitTask(m_etcd.DefaultTaskFunc("task1", "")); err != nil {
t.Fatalf("Error submitting task1 to b: %v", err)
}
// Give consumers a bit to pick up tasks
time.Sleep(250 * time.Millisecond)
assertRunning := func(tid string, cons ...*metafora.Consumer) {
found := false
for _, c := range cons {
tasks := c.Tasks()
if len(tasks) > 0 && found {
t.Fatal("Task already found running but another task is running on a different consumer")
}
if len(tasks) > 1 {
t.Fatalf("Expected at most 1 task, but found: %d", len(tasks))
}
if len(tasks) == 1 && tasks[0].Task().ID() == tid {
found = true
}
}
if !found {
t.Fatalf("Could not find task=%q", tid)
}
}
assertRunning("task1", cons1a, cons2a)
assertRunning("task1", cons1b, cons2b)
// Kill task1 in A
{
cmdr := m_etcd.NewCommander("test-a", etcdc)
if err := cmdr.Send("task1", statemachine.KillMessage()); err != nil {
t.Fatalf("Error sending kill to task1: %v", err)
}
time.Sleep(250 * time.Millisecond)
for _, c := range []*metafora.Consumer{cons1a, cons2a} {
tasks := c.Tasks()
if len(tasks) != 0 {
t.Fatalf("Expected no tasks but found: %d", len(tasks))
}
}
}
// Submit a bunch of tasks to A
{
tasks := []string{"task2", "task3", "task4", "task5", "task6", "task7"}
for _, tid := range tasks {
if err := cliA.SubmitTask(m_etcd.DefaultTaskFunc(tid, "")); err != nil {
t.Fatalf("Error submitting task=%q to A: %v", tid, err)
}
}
// Give them time to start
time.Sleep(800 * time.Millisecond)
// Ensure they're balanced
if err := cliA.SubmitCommand("node1", metafora.CommandBalance()); err != nil {
t.Fatalf("Error submitting balance command to cons1a: %v", err)
}
time.Sleep(800 * time.Millisecond)
if err := cliA.SubmitCommand("node2", metafora.CommandBalance()); err != nil {
t.Fatalf("Error submitting balance command to cons1a: %v", err)
}
a1tasks := cons1a.Tasks()
a2tasks := cons2a.Tasks()
for _, task := range a1tasks {
metafora.Debug("A1: ", task.Task(), " - ", task.Stopped().IsZero())
}
for _, task := range a2tasks {
metafora.Debug("A2: ", task.Task(), " - ", task.Stopped().IsZero())
}
time.Sleep(800 * time.Millisecond)
a1tasks = cons1a.Tasks()
a2tasks = cons2a.Tasks()
if len(a1tasks) < 2 || len(a1tasks) > 4 || len(a2tasks) < 2 || len(a2tasks) > 4 {
t.Fatalf("Namespace A isn't fairly balanced: node1: %d; node2: %d", len(a1tasks), len(a2tasks))
}
// Shutting down a consumer should migrate all tasks to the other
cons1a.Shutdown()
time.Sleep(800 * time.Millisecond)
a2tasks = cons2a.Tasks()
if len(a2tasks) != len(tasks) {
t.Fatalf("Consumer 2a should have received all %d tasks but only has %d.", len(tasks), len(a2tasks))
}
}
// Use Namespace B to check Error state handling
{
tasks := []string{"task8", "error-test"}
for _, tid := range tasks {
if err := cliB.SubmitTask(m_etcd.DefaultTaskFunc(tid, "")); err != nil {
t.Fatalf("Error submitting task=%q to B: %v", tid, err)
}
}
// Give them time to start
time.Sleep(time.Second)
n := len(cons1b.Tasks()) + len(cons2b.Tasks())
if n != 3 {
t.Fatalf("Expected B to be running 3 tasks but found %d", n)
}
// Resuming error-test 8*2 times should cause it to be failed
cmdr := m_etcd.NewCommander("test-b", etcdc)
for i := 0; i < statemachine.DefaultErrMax*2; i++ {
if err := cmdr.Send("error-test", statemachine.RunMessage()); err != nil {
t.Fatalf("Unexpected error resuming error-test in B: %v", err)
}
time.Sleep(500 * time.Millisecond)
}
n = len(cons1b.Tasks()) + len(cons2b.Tasks())
if n != 2 {
t.Fatalf("Expected B to be running 2 tasks but found %d", n)
}
// Resubmitting a failed task shouldn't error but also shouldn't run.
if err := cliB.SubmitTask(m_etcd.DefaultTaskFunc("error-test", "")); err != nil {
t.Fatalf("Error resubmitting error-test task to B: %v", err)
}
// Give the statemachine a moment to load the initial state and exit
time.Sleep(time.Second)
n = len(cons1b.Tasks()) + len(cons2b.Tasks())
if n != 2 {
t.Fatalf("Expected B to be running 2 tasks but found %d", n)
}
}
// Shutdown
cons2a.Shutdown()
cons1b.Shutdown()
cons2b.Shutdown()
// Make sure everything is cleaned up
respA, err := etcdc.Get("/test-a/tasks", true, true)
if err != nil {
t.Fatalf("Error getting tasks from etcd: %v", err)
}
respB, err := etcdc.Get("/test-b/tasks", true, true)
if err != nil {
t.Fatalf("Error getting tasks from etcd: %v", err)
}
nodes := []*etcd.Node{}
nodes = append(nodes, respA.Node.Nodes...)
nodes = append(nodes, respB.Node.Nodes...)
for _, node := range nodes {
if len(node.Nodes) > 1 {
t.Fatalf("%s has %d (>1) nodes. First 2: %s, %s", node.Key, len(node.Nodes), node.Nodes[0].Key, node.Nodes[1].Key)
}
}
}