func (s *DeployerSuite) TestOmniProcess(t *c.C) {
if testCluster == nil {
t.Skip("cannot determine test cluster size")
}
// create and scale an omni release
omniScale := 2
totalJobs := omniScale * testCluster.Size()
client := s.controllerClient(t)
app, release := s.createApp(t)
watcher, err := client.WatchJobEvents(app.Name, release.ID)
t.Assert(err, c.IsNil)
defer watcher.Close()
t.Assert(client.PutFormation(&ct.Formation{
AppID: app.ID,
ReleaseID: release.ID,
Processes: map[string]int{"omni": omniScale},
}), c.IsNil)
err = watcher.WaitFor(ct.JobEvents{"omni": {ct.JobStateUp: totalJobs}}, scaleTimeout, nil)
t.Assert(err, c.IsNil)
// deploy using all-at-once and check we get the correct events
app.Strategy = "all-at-once"
t.Assert(client.UpdateApp(app), c.IsNil)
release.ID = ""
t.Assert(client.CreateRelease(release), c.IsNil)
deployment, err := client.CreateDeployment(app.ID, release.ID)
t.Assert(err, c.IsNil)
events := make(chan *ct.DeploymentEvent)
stream, err := client.StreamDeployment(deployment, events)
t.Assert(err, c.IsNil)
defer stream.Close()
expected := make([]*ct.Job, 0, 3*totalJobs+1)
appendEvents := func(releaseID string, state ct.JobState, count int) {
for i := 0; i < count; i++ {
expected = append(expected, &ct.Job{
ReleaseID: releaseID,
Type: "omni",
State: state,
})
}
}
appendEvents(deployment.NewReleaseID, ct.JobStateUp, totalJobs)
appendEvents(deployment.OldReleaseID, ct.JobStateDown, totalJobs)
s.waitForDeploymentStatus(t, events, "complete")
// deploy using one-by-one and check we get the correct events
app.Strategy = "one-by-one"
t.Assert(client.UpdateApp(app), c.IsNil)
release.ID = ""
t.Assert(client.CreateRelease(release), c.IsNil)
// try creating the deployment multiple times to avoid getting a
// "Cannot create deploy, one is already in progress" error (there
// is no guarantee the previous deploy has finished yet)
attempts := attempt.Strategy{Total: 10 * time.Second, Delay: 100 * time.Millisecond}
err = attempts.Run(func() (err error) {
deployment, err = client.CreateDeployment(app.ID, release.ID)
return
})
t.Assert(err, c.IsNil)
events = make(chan *ct.DeploymentEvent)
stream, err = client.StreamDeployment(deployment, events)
t.Assert(err, c.IsNil)
expected = make([]*ct.Job, 0, 4*totalJobs+1)
appendEvents(deployment.NewReleaseID, ct.JobStateUp, testCluster.Size())
appendEvents(deployment.OldReleaseID, ct.JobStateDown, testCluster.Size())
appendEvents(deployment.NewReleaseID, ct.JobStateUp, testCluster.Size())
appendEvents(deployment.OldReleaseID, ct.JobStateDown, testCluster.Size())
s.waitForDeploymentStatus(t, events, "complete")
}
func (s *SchedulerSuite) TestScaleTags(t *c.C) {
// ensure we have more than 1 host to test with
hosts, err := s.clusterClient(t).Hosts()
t.Assert(err, c.IsNil)
if len(hosts) <= 1 {
t.Skip("not enough hosts to test tagged based scheduling")
}
// watch service events so we can wait for tag changes
events := make(chan *discoverd.Event)
stream, err := s.discoverdClient(t).Service("flynn-host").Watch(events)
t.Assert(err, c.IsNil)
defer stream.Close()
waitServiceEvent := func(kind discoverd.EventKind) *discoverd.Event {
for {
select {
case event, ok := <-events:
if !ok {
t.Fatalf("service event stream closed unexpectedly: %s", stream.Err())
}
if event.Kind == kind {
return event
}
case <-time.After(10 * time.Second):
t.Fatalf("timed out waiting for service %s event", kind)
}
}
}
// wait for the watch to be current before changing tags
waitServiceEvent(discoverd.EventKindCurrent)
updateTags := func(host *cluster.Host, tags map[string]string) {
debugf(t, "setting host tags: %s => %v", host.ID(), tags)
t.Assert(host.UpdateTags(tags), c.IsNil)
event := waitServiceEvent(discoverd.EventKindUpdate)
t.Assert(event.Instance.Meta["id"], c.Equals, host.ID())
for key, val := range tags {
t.Assert(event.Instance.Meta["tag:"+key], c.Equals, val)
}
}
// create an app with a tagged process and watch job events
app, release := s.createApp(t)
formation := &ct.Formation{
AppID: app.ID,
ReleaseID: release.ID,
Tags: map[string]map[string]string{"printer": {"active": "true"}},
}
client := s.controllerClient(t)
watcher, err := client.WatchJobEvents(app.ID, release.ID)
t.Assert(err, c.IsNil)
defer watcher.Close()
// add tag to host 1
host1 := hosts[0]
updateTags(host1, map[string]string{"active": "true"})
// start jobs
debug(t, "scaling printer=2")
formation.Processes = map[string]int{"printer": 2}
t.Assert(client.PutFormation(formation), c.IsNil)
t.Assert(watcher.WaitFor(ct.JobEvents{"printer": ct.JobUpEvents(2)}, scaleTimeout, nil), c.IsNil)
assertHostJobCounts := func(expected map[string]int) {
jobs, err := client.JobList(app.ID)
t.Assert(err, c.IsNil)
actual := make(map[string]int)
for _, job := range jobs {
if job.State == ct.JobStateUp {
actual[job.HostID]++
}
}
t.Assert(actual, c.DeepEquals, expected)
}
// check all jobs on host 1
assertHostJobCounts(map[string]int{host1.ID(): 2})
// add tag to host 2
host2 := hosts[1]
updateTags(host2, map[string]string{"active": "true"})
// scale up
debug(t, "scaling printer=4")
formation.Processes["printer"] = 4
t.Assert(client.PutFormation(formation), c.IsNil)
t.Assert(watcher.WaitFor(ct.JobEvents{"printer": ct.JobUpEvents(2)}, scaleTimeout, nil), c.IsNil)
// check jobs distributed across hosts 1 and 2
assertHostJobCounts(map[string]int{host1.ID(): 2, host2.ID(): 2})
// remove tag from host 2
updateTags(host2, map[string]string{"active": ""})
// check jobs are moved to host1
jobEvents := ct.JobEvents{"printer": map[ct.JobState]int{
ct.JobStateDown: 2,
ct.JobStateUp: 2,
}}
t.Assert(watcher.WaitFor(jobEvents, scaleTimeout, nil), c.IsNil)
assertHostJobCounts(map[string]int{host1.ID(): 4})
// remove tag from host 1
updateTags(host1, map[string]string{"active": ""})
assertStateCounts := func(expected map[ct.JobState]int) {
jobs, err := client.JobList(app.ID)
t.Assert(err, c.IsNil)
actual := make(map[ct.JobState]int)
for _, job := range jobs {
actual[job.State]++
}
t.Assert(actual, c.DeepEquals, expected)
}
// check 4 pending jobs, rest are stopped
t.Assert(watcher.WaitFor(ct.JobEvents{"printer": ct.JobDownEvents(4)}, scaleTimeout, nil), c.IsNil)
assertStateCounts(map[ct.JobState]int{ct.JobStatePending: 4, ct.JobStateDown: 6})
// re-add tag to host 1
updateTags(host1, map[string]string{"active": "true"})
// check pending jobs are started on host 1
t.Assert(watcher.WaitFor(ct.JobEvents{"printer": ct.JobUpEvents(4)}, scaleTimeout, nil), c.IsNil)
assertHostJobCounts(map[string]int{host1.ID(): 4})
assertStateCounts(map[ct.JobState]int{ct.JobStateUp: 4, ct.JobStateDown: 6})
// add different tag to host 2
updateTags(host2, map[string]string{"disk": "ssd"})
// update formation tags, check jobs are moved to host 2
debug(t, "updating formation tags to disk=ssd")
formation.Tags["printer"] = map[string]string{"disk": "ssd"}
t.Assert(client.PutFormation(formation), c.IsNil)
jobEvents = ct.JobEvents{"printer": map[ct.JobState]int{
ct.JobStateDown: 4,
ct.JobStateUp: 4,
}}
t.Assert(watcher.WaitFor(jobEvents, scaleTimeout, nil), c.IsNil)
assertHostJobCounts(map[string]int{host2.ID(): 4})
assertStateCounts(map[ct.JobState]int{ct.JobStateUp: 4, ct.JobStateDown: 10})
// scale down stops the jobs
debug(t, "scaling printer=0")
formation.Processes = nil
t.Assert(client.PutFormation(formation), c.IsNil)
t.Assert(watcher.WaitFor(ct.JobEvents{"printer": ct.JobDownEvents(4)}, scaleTimeout, nil), c.IsNil)
assertStateCounts(map[ct.JobState]int{ct.JobStateDown: 14})
}
func (s *PostgresSuite) testDeploy(t *c.C, d *pgDeploy) {
// create postgres app
client := s.controllerClient(t)
app := &ct.App{Name: d.name, Strategy: "postgres"}
t.Assert(client.CreateApp(app), c.IsNil)
// copy release from default postgres app
release, err := client.GetAppRelease("postgres")
t.Assert(err, c.IsNil)
release.ID = ""
proc := release.Processes["postgres"]
delete(proc.Env, "SINGLETON")
proc.Env["FLYNN_POSTGRES"] = d.name
proc.Service = d.name
release.Processes["postgres"] = proc
t.Assert(client.CreateRelease(release), c.IsNil)
t.Assert(client.SetAppRelease(app.ID, release.ID), c.IsNil)
oldRelease := release.ID
// create formation
discEvents := make(chan *discoverd.Event)
discStream, err := s.discoverdClient(t).Service(d.name).Watch(discEvents)
t.Assert(err, c.IsNil)
defer discStream.Close()
jobEvents := make(chan *ct.Job)
jobStream, err := client.StreamJobEvents(d.name, jobEvents)
t.Assert(err, c.IsNil)
defer jobStream.Close()
t.Assert(client.PutFormation(&ct.Formation{
AppID: app.ID,
ReleaseID: release.ID,
Processes: map[string]int{"postgres": d.pgJobs, "web": d.webJobs},
}), c.IsNil)
// watch cluster state changes
type stateChange struct {
state *state.State
err error
}
stateCh := make(chan stateChange)
go func() {
for event := range discEvents {
if event.Kind != discoverd.EventKindServiceMeta {
continue
}
var state state.State
if err := json.Unmarshal(event.ServiceMeta.Data, &state); err != nil {
stateCh <- stateChange{err: err}
return
}
primary := ""
if state.Primary != nil {
primary = state.Primary.Addr
}
sync := ""
if state.Sync != nil {
sync = state.Sync.Addr
}
var async []string
for _, a := range state.Async {
async = append(async, a.Addr)
}
debugf(t, "got pg cluster state: index=%d primary=%s sync=%s async=%s",
event.ServiceMeta.Index, primary, sync, strings.Join(async, ","))
stateCh <- stateChange{state: &state}
}
}()
// wait for correct cluster state and number of web processes
var pgState state.State
var webJobs int
ready := func() bool {
if webJobs != d.webJobs {
return false
}
if pgState.Primary == nil {
return false
}
if d.pgJobs > 1 && pgState.Sync == nil {
return false
}
if d.pgJobs > 2 && len(pgState.Async) != d.pgJobs-2 {
return false
}
return true
}
for {
if ready() {
break
}
select {
case s := <-stateCh:
t.Assert(s.err, c.IsNil)
pgState = *s.state
case e, ok := <-jobEvents:
if !ok {
t.Fatalf("job event stream closed: %s", jobStream.Err())
}
debugf(t, "got job event: %s %s %s", e.Type, e.ID, e.State)
if e.Type == "web" && e.State == "up" {
webJobs++
}
case <-time.After(30 * time.Second):
t.Fatal("timed out waiting for postgres formation")
}
}
// connect to the db so we can test writes
db := postgres.Wait(d.name, fmt.Sprintf("dbname=postgres user=flynn password=%s", release.Env["PGPASSWORD"]))
dbname := "deploy-test"
t.Assert(db.Exec(fmt.Sprintf(`CREATE DATABASE "%s" WITH OWNER = "flynn"`, dbname)), c.IsNil)
db.Close()
db, err = postgres.Open(d.name, fmt.Sprintf("dbname=%s user=flynn password=%s", dbname, release.Env["PGPASSWORD"]))
t.Assert(err, c.IsNil)
defer db.Close()
t.Assert(db.Exec(`CREATE TABLE deploy_test ( data text)`), c.IsNil)
assertWriteable := func() {
debug(t, "writing to postgres database")
t.Assert(db.Exec(`INSERT INTO deploy_test (data) VALUES ('data')`), c.IsNil)
}
// check currently writeable
assertWriteable()
// check a deploy completes with expected cluster state changes
release.ID = ""
t.Assert(client.CreateRelease(release), c.IsNil)
newRelease := release.ID
deployment, err := client.CreateDeployment(app.ID, newRelease)
t.Assert(err, c.IsNil)
deployEvents := make(chan *ct.DeploymentEvent)
deployStream, err := client.StreamDeployment(deployment, deployEvents)
t.Assert(err, c.IsNil)
defer deployStream.Close()
// assertNextState checks that the next state received is in the remaining states
// that were expected, so handles the fact that some states don't happen, but the
// states that do happen are expected and in-order.
assertNextState := func(remaining []expectedPgState) int {
var state state.State
loop:
for {
select {
case s := <-stateCh:
t.Assert(s.err, c.IsNil)
if len(s.state.Async) < d.expectedAsyncs() {
// we shouldn't usually receive states with less asyncs than
// expected, but they can occur as an intermediate state between
// two expected states (e.g. when a sync does a takeover at the
// same time as a new async is started) so just ignore them.
debug(t, "ignoring state with too few asyncs")
continue
}
state = *s.state
break loop
case <-time.After(60 * time.Second):
t.Fatal("timed out waiting for postgres cluster state")
}
}
if state.Primary == nil {
t.Fatal("no primary configured")
}
log := func(format string, v ...interface{}) {
debugf(t, "skipping expected state: %s", fmt.Sprintf(format, v...))
}
outer:
for i, expected := range remaining {
if state.Primary.Meta["FLYNN_RELEASE_ID"] != expected.Primary {
log("primary has incorrect release")
continue
}
if state.Sync == nil {
if expected.Sync == "" {
return i
}
log("state has no sync node")
continue
}
if state.Sync.Meta["FLYNN_RELEASE_ID"] != expected.Sync {
log("sync has incorrect release")
continue
}
if state.Async == nil {
if expected.Async == nil {
return i
}
log("state has no async nodes")
continue
}
if len(state.Async) != len(expected.Async) {
log("expected %d asyncs, got %d", len(expected.Async), len(state.Async))
continue
}
for i, release := range expected.Async {
if state.Async[i].Meta["FLYNN_RELEASE_ID"] != release {
log("async[%d] has incorrect release", i)
continue outer
}
}
return i
}
t.Fatal("unexpected pg state")
return -1
}
expected := d.expected(oldRelease, newRelease)
var expectedIndex, newWebJobs int
loop:
for {
select {
case e, ok := <-deployEvents:
if !ok {
t.Fatal("unexpected close of deployment event stream")
}
switch e.Status {
case "complete":
break loop
case "failed":
t.Fatalf("deployment failed: %s", e.Error)
}
debugf(t, "got deployment event: %s %s", e.JobType, e.JobState)
if e.JobState != "up" && e.JobState != "down" {
continue
}
switch e.JobType {
case "postgres":
// move on if we have seen all the expected events
if expectedIndex >= len(expected) {
continue
}
skipped := assertNextState(expected[expectedIndex:])
expectedIndex += 1 + skipped
case "web":
if e.JobState == "up" && e.ReleaseID == newRelease {
newWebJobs++
}
}
case <-time.After(2 * time.Minute):
t.Fatal("timed out waiting for deployment")
}
}
// check we have the correct number of new web jobs
t.Assert(newWebJobs, c.Equals, d.webJobs)
// check writeable now deploy is complete
assertWriteable()
}
func (s *DeployerSuite) TestOmniProcess(t *c.C) {
if testCluster == nil {
t.Skip("cannot determine test cluster size")
}
// create and scale an omni release
omniScale := 2
totalJobs := omniScale * testCluster.Size()
client := s.controllerClient(t)
app, release := s.createApp(t)
watcher, err := client.WatchJobEvents(app.Name, release.ID)
t.Assert(err, c.IsNil)
defer watcher.Close()
t.Assert(client.PutFormation(&ct.Formation{
AppID: app.ID,
ReleaseID: release.ID,
Processes: map[string]int{"omni": omniScale},
}), c.IsNil)
err = watcher.WaitFor(ct.JobEvents{"omni": {ct.JobStateUp: totalJobs}}, scaleTimeout, nil)
t.Assert(err, c.IsNil)
// deploy using all-at-once and check we get the correct events
app.Strategy = "all-at-once"
t.Assert(client.UpdateApp(app), c.IsNil)
release.ID = ""
t.Assert(client.CreateRelease(release), c.IsNil)
deployment, err := client.CreateDeployment(app.ID, release.ID)
t.Assert(err, c.IsNil)
events := make(chan *ct.DeploymentEvent)
stream, err := client.StreamDeployment(deployment, events)
t.Assert(err, c.IsNil)
defer stream.Close()
expected := make([]*ct.DeploymentEvent, 0, 4*totalJobs+1)
appendEvents := func(releaseID string, state ct.JobState, count int) {
for i := 0; i < count; i++ {
event := &ct.DeploymentEvent{
ReleaseID: releaseID,
JobType: "omni",
JobState: state,
Status: "running",
}
expected = append(expected, event)
}
}
expected = append(expected, &ct.DeploymentEvent{ReleaseID: deployment.NewReleaseID, Status: "pending"})
appendEvents(deployment.NewReleaseID, ct.JobStateStarting, totalJobs)
appendEvents(deployment.NewReleaseID, ct.JobStateUp, totalJobs)
appendEvents(deployment.OldReleaseID, ct.JobStateStopping, totalJobs)
appendEvents(deployment.OldReleaseID, ct.JobStateDown, totalJobs)
expected = append(expected, &ct.DeploymentEvent{ReleaseID: deployment.NewReleaseID, Status: "complete"})
waitForDeploymentEvents(t, events, expected)
// deploy using one-by-one and check we get the correct events
app.Strategy = "one-by-one"
t.Assert(client.UpdateApp(app), c.IsNil)
release.ID = ""
t.Assert(client.CreateRelease(release), c.IsNil)
deployment, err = client.CreateDeployment(app.ID, release.ID)
t.Assert(err, c.IsNil)
events = make(chan *ct.DeploymentEvent)
stream, err = client.StreamDeployment(deployment, events)
t.Assert(err, c.IsNil)
expected = make([]*ct.DeploymentEvent, 0, 4*totalJobs+1)
expected = append(expected, &ct.DeploymentEvent{ReleaseID: deployment.NewReleaseID, Status: "pending"})
appendEvents(deployment.NewReleaseID, ct.JobStateStarting, testCluster.Size())
appendEvents(deployment.NewReleaseID, ct.JobStateUp, testCluster.Size())
appendEvents(deployment.OldReleaseID, ct.JobStateStopping, testCluster.Size())
appendEvents(deployment.OldReleaseID, ct.JobStateDown, testCluster.Size())
appendEvents(deployment.NewReleaseID, ct.JobStateStarting, testCluster.Size())
appendEvents(deployment.NewReleaseID, ct.JobStateUp, testCluster.Size())
appendEvents(deployment.OldReleaseID, ct.JobStateStopping, testCluster.Size())
appendEvents(deployment.OldReleaseID, ct.JobStateDown, testCluster.Size())
expected = append(expected, &ct.DeploymentEvent{ReleaseID: deployment.NewReleaseID, Status: "complete"})
waitForDeploymentEvents(t, events, expected)
}
func (s *SchedulerSuite) TestGracefulShutdown(t *c.C) {
app, release := s.createApp(t)
client := s.controllerClient(t)
debug(t, "scaling to blocker=1")
watcher, err := client.WatchJobEvents(app.ID, release.ID)
t.Assert(err, c.IsNil)
defer watcher.Close()
t.Assert(client.PutFormation(&ct.Formation{
AppID: app.ID,
ReleaseID: release.ID,
Processes: map[string]int{"blocker": 1},
}), c.IsNil)
var jobID string
err = watcher.WaitFor(ct.JobEvents{"blocker": ct.JobUpEvents(1)}, scaleTimeout, func(job *ct.Job) error {
jobID = job.ID
return nil
})
t.Assert(err, c.IsNil)
jobs, err := s.discoverdClient(t).Instances("test-http-blocker", 10*time.Second)
t.Assert(err, c.IsNil)
t.Assert(jobs, c.HasLen, 1)
jobAddr := jobs[0].Addr
debug(t, "subscribing to backend events from all routers")
routers, err := s.discoverdClient(t).Instances("router-api", 10*time.Second)
t.Assert(err, c.IsNil)
routerEvents := make(chan *router.StreamEvent)
for _, r := range routers {
events := make(chan *router.StreamEvent)
stream, err := routerc.NewWithAddr(r.Addr).StreamEvents(&router.StreamEventsOptions{
EventTypes: []router.EventType{
router.EventTypeBackendUp,
router.EventTypeBackendDown,
router.EventTypeBackendDrained,
},
}, events)
t.Assert(err, c.IsNil)
defer stream.Close()
go func(router *discoverd.Instance) {
for event := range events {
if event.Backend != nil && event.Backend.JobID == jobID {
debugf(t, "got %s router event from %s", event.Event, router.Host())
routerEvents <- event
}
}
}(r)
}
debug(t, "adding HTTP route with backend drain enabled")
route := &router.HTTPRoute{
Domain: random.String(32) + ".com",
Service: "test-http-blocker",
DrainBackends: true,
}
t.Assert(client.CreateRoute(app.ID, route.ToRoute()), c.IsNil)
waitForRouterEvents := func(typ router.EventType) {
debugf(t, "waiting for %d router %s events", len(routers), typ)
count := 0
for {
select {
case event := <-routerEvents:
if event.Event != typ {
t.Fatal("expected %s router event, got %s", typ, event.Event)
}
count++
if count == len(routers) {
return
}
case <-time.After(30 * time.Second):
t.Fatalf("timed out waiting for router %s events", typ)
}
}
}
waitForRouterEvents(router.EventTypeBackendUp)
debug(t, "making blocked HTTP request through each router")
reqErrs := make(chan error)
for _, router := range routers {
req, err := http.NewRequest("GET", "http://"+router.Host()+"/block", nil)
t.Assert(err, c.IsNil)
req.Host = route.Domain
res, err := http.DefaultClient.Do(req)
t.Assert(err, c.IsNil)
t.Assert(res.StatusCode, c.Equals, http.StatusOK)
go func() {
defer res.Body.Close()
data, err := ioutil.ReadAll(res.Body)
if err == nil && !bytes.Equal(data, []byte("done")) {
err = fmt.Errorf("unexpected response: %q", data)
}
reqErrs <- err
}()
}
debug(t, "scaling to blocker=0")
t.Assert(client.PutFormation(&ct.Formation{
AppID: app.ID,
ReleaseID: release.ID,
Processes: map[string]int{"blocker": 0},
}), c.IsNil)
t.Assert(watcher.WaitFor(ct.JobEvents{"blocker": {ct.JobStateStopping: 1}}, scaleTimeout, nil), c.IsNil)
waitForRouterEvents(router.EventTypeBackendDown)
debug(t, "checking new HTTP requests return 503")
for _, router := range routers {
req, err := http.NewRequest("GET", "http://"+router.Host()+"/ping", nil)
t.Assert(err, c.IsNil)
req.Host = route.Domain
res, err := http.DefaultClient.Do(req)
t.Assert(err, c.IsNil)
res.Body.Close()
t.Assert(res.StatusCode, c.Equals, http.StatusServiceUnavailable)
}
debug(t, "checking blocked HTTP requests are still blocked")
select {
case err := <-reqErrs:
t.Fatal(err)
default:
}
debug(t, "unblocking HTTP requests")
res, err := http.Get("http://" + jobAddr + "/unblock")
t.Assert(err, c.IsNil)
t.Assert(res.StatusCode, c.Equals, http.StatusOK)
debug(t, "checking the blocked HTTP requests completed without error")
for range routers {
if err := <-reqErrs; err != nil {
t.Fatal(err)
}
}
waitForRouterEvents(router.EventTypeBackendDrained)
debug(t, "waiting for the job to exit")
t.Assert(watcher.WaitFor(ct.JobEvents{"blocker": ct.JobDownEvents(1)}, scaleTimeout, nil), c.IsNil)
}