func startJob(s *State, hc *cluster.Host, job *host.Job) error {
jobStatus := make(chan error)
events := make(chan *host.Event)
stream, err := hc.StreamEvents(job.ID, events)
if err != nil {
return err
}
go func() {
defer stream.Close()
loop:
for {
select {
case e, ok := <-events:
if !ok {
break loop
}
switch e.Event {
case "start", "stop":
jobStatus <- nil
return
case "error":
job, err := hc.GetJob(job.ID)
if err != nil {
jobStatus <- err
return
}
if job.Error == nil {
jobStatus <- fmt.Errorf("bootstrap: unknown error from host")
return
}
jobStatus <- fmt.Errorf("bootstrap: host error while launching job: %q", *job.Error)
return
default:
}
case <-time.After(30 * time.Second):
jobStatus <- errors.New("bootstrap: timed out waiting for job event")
return
}
}
jobStatus <- fmt.Errorf("bootstrap: host job stream disconnected unexpectedly: %q", stream.Err())
}()
if err := hc.AddJob(job); err != nil {
return err
}
return <-jobStatus
}
func startJob(s *State, hc *cluster.Host, job *host.Job) (*Job, error) {
data := &Job{HostID: hc.ID(), JobID: job.ID}
jobStatus := make(chan error)
events := make(chan *host.Event)
stream, err := hc.StreamEvents(data.JobID, events)
if err != nil {
return nil, err
}
go func() {
defer stream.Close()
for e := range events {
switch e.Event {
case "start", "stop":
jobStatus <- nil
return
case "error":
job, err := hc.GetJob(data.JobID)
if err != nil {
jobStatus <- err
return
}
if job.Error == nil {
jobStatus <- fmt.Errorf("bootstrap: unknown error from host")
return
}
jobStatus <- fmt.Errorf("bootstrap: host error while launching job: %q", *job.Error)
return
default:
}
}
jobStatus <- fmt.Errorf("bootstrap: host job stream disconnected unexpectedly: %q", stream.Err())
}()
if err := hc.AddJob(job); err != nil {
return nil, err
}
return data, <-jobStatus
}
func (f *ClusterFixer) FixPostgres() error {
f.l.Info("checking postgres")
service := discoverd.NewService("postgres")
leader, _ := service.Leader()
if leader == nil || leader.Addr == "" {
f.l.Info("no running postgres leader")
leader = nil
} else {
f.l.Info("found running postgres leader")
}
instances, _ := service.Instances()
f.l.Info(fmt.Sprintf("found %d running postgres instances", len(instances)))
f.l.Info("getting postgres status")
var status *pgmanager.Status
if leader != nil && leader.Addr != "" {
client := pgmanager.NewClient(leader.Addr)
var err error
status, err = client.Status()
if err != nil {
f.l.Error("error getting status from postgres leader", "error", err)
}
}
if status != nil && status.Postgres.ReadWrite {
f.l.Info("postgres claims to be read-write")
return nil
}
f.l.Info("getting postgres service metadata")
meta, err := discoverd.NewService("postgres").GetMeta()
if err != nil {
return fmt.Errorf("error getting postgres state from discoverd: %s", err)
}
var state pgstate.State
if err := json.Unmarshal(meta.Data, &state); err != nil {
return fmt.Errorf("error decoding postgres state: %s", err)
}
if state.Primary == nil {
return fmt.Errorf("no primary in postgres state")
}
f.l.Info("getting postgres primary job info", "job.id", state.Primary.Meta["FLYNN_JOB_ID"])
job, host, err := f.GetJob(state.Primary.Meta["FLYNN_JOB_ID"])
if err != nil {
if state.Sync != nil {
f.l.Error("unable to get primary job info", "error", err)
f.l.Info("getting postgres sync job info", "job.id", state.Sync.Meta["FLYNN_JOB_ID"])
job, host, err = f.GetJob(state.Sync.Meta["FLYNN_JOB_ID"])
if err != nil {
return fmt.Errorf("unable to get postgres primary or sync job details: %s", err)
}
} else {
return fmt.Errorf("unable to get postgres primary job details: %s", err)
}
}
if leader != nil && state.Singleton {
return fmt.Errorf("postgres leader is running in singleton mode, unable to fix")
}
waitForInstance := func(jobID string) (func() (string, error), error) {
watchCh := make(chan *discoverd.Event)
upCh := make(chan string)
stream, err := service.Watch(watchCh)
if err != nil {
return nil, fmt.Errorf("error watching discoverd service: %s", err)
}
go func() {
var current bool
for event := range watchCh {
if event.Kind == discoverd.EventKindCurrent {
current = true
continue
}
if !current || event.Kind != discoverd.EventKindUp {
continue
}
if event.Instance.Meta["FLYNN_JOB_ID"] == jobID {
upCh <- event.Instance.Addr
}
}
}()
return func() (string, error) {
f.l.Info("waiting for postgres instance to start", "job.id", jobID)
defer stream.Close()
select {
case addr := <-upCh:
return addr, nil
case <-time.After(time.Minute):
return "", fmt.Errorf("timed out waiting for postgres instance to come up")
}
}, nil
}
var wait func() (string, error)
have := len(instances)
want := 2
if state.Singleton {
want = 1
}
if have >= want {
return fmt.Errorf("already have enough postgres instances, unable to fix")
}
f.l.Info("attempting to start missing postgres jobs", "want", want, "have", have)
if leader == nil {
// if no postgres, attempt to start
job.ID = cluster.GenerateJobID(host.ID(), "")
f.FixJobEnv(job)
f.l.Info("starting postgres primary job", "job.id", job.ID)
wait, err = waitForInstance(job.ID)
if err != nil {
return err
}
if err := host.AddJob(job); err != nil {
return fmt.Errorf("error starting postgres primary job on %s: %s", host.ID(), err)
}
have++
}
if want > have {
// if not enough postgres instances, start another
var secondHost *cluster.Host
for _, h := range f.hosts {
if h.ID() != host.ID() {
secondHost = h
break
}
}
if secondHost == nil {
// if there are no other hosts, use the same one we put the primary on
secondHost = host
}
job.ID = cluster.GenerateJobID(secondHost.ID(), "")
f.FixJobEnv(job)
f.l.Info("starting second postgres job", "job.id", job.ID)
if wait == nil {
wait, err = waitForInstance(job.ID)
if err != nil {
return err
}
}
if err := utils.ProvisionVolume(secondHost, job); err != nil {
return fmt.Errorf("error creating postgres volume on %s: %s", secondHost.ID(), err)
}
if err := secondHost.AddJob(job); err != nil {
return fmt.Errorf("error starting additional postgres job on %s: %s", secondHost.ID(), err)
}
}
if wait != nil {
addr, err := wait()
if err != nil {
return err
}
if leader != nil {
addr = leader.Addr
}
f.l.Info("waiting for postgres to come up read-write")
return pgmanager.NewClient(addr).WaitForReadWrite(5 * time.Minute)
}
return nil
}
func (f *ClusterFixer) FixSirenia(svc string) error {
log := f.l.New("fn", "FixSirenia", "service", svc)
service := discoverd.NewService(svc)
instances, _ := service.Instances()
leader, _ := service.Leader()
log.Info("getting service metadata")
meta, err := service.GetMeta()
if err != nil {
return fmt.Errorf("error getting sirenia state from discoverd: %s", err)
}
var state state.State
if err := json.Unmarshal(meta.Data, &state); err != nil {
return fmt.Errorf("error decoding state: %s", err)
}
if state.Primary == nil {
return fmt.Errorf("no primary in sirenia state")
}
log.Info("getting primary job info", "job.id", state.Primary.Meta["FLYNN_JOB_ID"])
primaryJob, primaryHost, err := f.GetJob(state.Primary.Meta["FLYNN_JOB_ID"])
if err != nil {
log.Error("unable to get primary job info")
}
var syncJob *host.Job
var syncHost *cluster.Host
if state.Sync != nil {
log.Info("getting sync job info", "job.id", state.Sync.Meta["FLYNN_JOB_ID"])
syncJob, syncHost, err = f.GetJob(state.Sync.Meta["FLYNN_JOB_ID"])
if err != nil {
log.Error("unable to get sync job info")
}
}
waitForInstance := func(jobID string) (func() (string, error), error) {
watchCh := make(chan *discoverd.Event)
upCh := make(chan string)
stream, err := service.Watch(watchCh)
if err != nil {
return nil, fmt.Errorf("error watching discoverd service: %s", err)
}
go func() {
var current bool
for event := range watchCh {
if event.Kind == discoverd.EventKindCurrent {
current = true
continue
}
if !current || event.Kind != discoverd.EventKindUp {
continue
}
if event.Instance.Meta["FLYNN_JOB_ID"] == jobID {
upCh <- event.Instance.Addr
}
}
}()
return func() (string, error) {
log.Info("waiting for instance to start", "job.id", jobID)
defer stream.Close()
select {
case addr := <-upCh:
return addr, nil
case <-time.After(time.Minute):
return "", fmt.Errorf("timed out waiting for sirenia instance to come up")
}
}, nil
}
log.Info("terminating unassigned sirenia instances")
outer:
for _, i := range instances {
if i.Addr == state.Primary.Addr || (state.Sync != nil && i.Addr == state.Sync.Addr) {
continue
}
for _, a := range state.Async {
if i.Addr == a.Addr {
continue outer
}
}
// job not assigned in state, attempt to terminate it
if jobID, ok := i.Meta["FLYNN_JOB_ID"]; ok {
hostID, err := cluster.ExtractHostID(jobID)
if err != nil {
log.Error("error extracting host id from jobID", "jobID", jobID, "err", err)
}
h := f.Host(hostID)
if h != nil {
if err := h.StopJob(jobID); err != nil {
log.Error("error stopping unassigned sirenia job", "jobID", jobID)
}
} else {
log.Error("host not found", "hostID", hostID)
}
}
}
isRunning := func(addr string) bool {
for _, i := range instances {
if i.Addr == addr {
return true
}
}
return false
}
// if the leader isn't currently running then start it using primaryJob/primaryHost
var wait func() (string, error)
if !isRunning(state.Primary.Addr) {
// if we don't have info about the primary job attempt to promote the sync
if primaryJob == nil {
if syncJob != nil {
// set primary job to sync
primaryJob = syncJob
primaryHost = syncHost
// nil out sync job now so we can re-allocate it.
syncJob = nil
syncHost = nil
} else {
return fmt.Errorf("neither primary or sync job info available")
}
}
primaryJob.ID = cluster.GenerateJobID(primaryHost.ID(), "")
f.FixJobEnv(primaryJob)
log.Info("starting primary job", "job.id", primaryJob.ID)
wait, err = waitForInstance(primaryJob.ID)
if err != nil {
return err
}
if err := primaryHost.AddJob(primaryJob); err != nil {
return fmt.Errorf("error starting primary job on %s: %s", primaryHost.ID(), err)
}
}
if !state.Singleton && !isRunning(state.Sync.Addr) {
if syncHost == nil {
for _, h := range f.hosts {
if h.ID() != primaryHost.ID() {
syncHost = h
break
}
}
if syncHost == nil {
// if there are no other hosts, use the same one we put the primary on
syncHost = primaryHost
}
}
// if we don't have a sync job then copy the primary job
// and provision a new volume
if syncJob == nil {
syncJob = primaryJob
vol := &ct.VolumeReq{Path: "/data"}
if _, err := utils.ProvisionVolume(vol, syncHost, syncJob); err != nil {
return fmt.Errorf("error creating volume on %s: %s", syncHost.ID(), err)
}
}
syncJob.ID = cluster.GenerateJobID(syncHost.ID(), "")
f.FixJobEnv(syncJob)
log.Info("starting sync job", "job.id", syncJob.ID)
if wait == nil {
wait, err = waitForInstance(syncJob.ID)
if err != nil {
return err
}
}
if err := syncHost.AddJob(syncJob); err != nil {
return fmt.Errorf("error starting additional job on %s: %s", syncHost.ID(), err)
}
}
if wait != nil {
addr, err := wait()
if err != nil {
return err
}
if leader != nil && leader.Addr != "" {
addr = leader.Addr
}
log.Info("waiting for cluster to come up read-write", "addr", addr)
return sirenia.NewClient(addr).WaitForReadWrite(5 * time.Minute)
}
return nil
}