func (c *ValidateCommand) Run(args []string) int {
flags := c.Meta.FlagSet("validate", FlagSetNone)
flags.Usage = func() { c.Ui.Output(c.Help()) }
if err := flags.Parse(args); err != nil {
return 1
}
// Check that we got exactly one node
args = flags.Args()
if len(args) != 1 {
c.Ui.Error(c.Help())
return 1
}
file := args[0]
// Parse the job file
job, err := jobspec.ParseFile(file)
if err != nil {
c.Ui.Error(fmt.Sprintf("Error parsing job file %s: %s", file, err))
return 1
}
// Initialize any fields that need to be.
job.InitFields()
// Check that the job is valid
if err := job.Validate(); err != nil {
c.Ui.Error(fmt.Sprintf("Error validating job: %s", err))
return 1
}
// Done!
c.Ui.Output("Job validation successful")
return 0
}
func (c *RunCommand) Run(args []string) int {
var detach, verbose, output bool
var checkIndexStr string
flags := c.Meta.FlagSet("run", FlagSetClient)
flags.Usage = func() { c.Ui.Output(c.Help()) }
flags.BoolVar(&detach, "detach", false, "")
flags.BoolVar(&verbose, "verbose", false, "")
flags.BoolVar(&output, "output", false, "")
flags.StringVar(&checkIndexStr, "check-index", "", "")
if err := flags.Parse(args); err != nil {
return 1
}
// Truncate the id unless full length is requested
length := shortId
if verbose {
length = fullId
}
// Check that we got exactly one node
args = flags.Args()
if len(args) != 1 {
c.Ui.Error(c.Help())
return 1
}
file := args[0]
// Parse the job file
job, err := jobspec.ParseFile(file)
if err != nil {
c.Ui.Error(fmt.Sprintf("Error parsing job file %s: %s", file, err))
return 1
}
// Initialize any fields that need to be.
job.InitFields()
// Check that the job is valid
if err := job.Validate(); err != nil {
c.Ui.Error(fmt.Sprintf("Error validating job: %s", err))
return 1
}
// Check if the job is periodic.
periodic := job.IsPeriodic()
// Convert it to something we can use
apiJob, err := convertStructJob(job)
if err != nil {
c.Ui.Error(fmt.Sprintf("Error converting job: %s", err))
return 1
}
if output {
req := api.RegisterJobRequest{Job: apiJob}
buf, err := json.MarshalIndent(req, "", " ")
if err != nil {
c.Ui.Error(fmt.Sprintf("Error converting job: %s", err))
return 1
}
c.Ui.Output(string(buf))
return 0
}
// Get the HTTP client
client, err := c.Meta.Client()
if err != nil {
c.Ui.Error(fmt.Sprintf("Error initializing client: %s", err))
return 1
}
// Force the region to be that of the job.
if r := job.Region; r != "" {
client.SetRegion(r)
}
// Parse the check-index
checkIndex, enforce, err := parseCheckIndex(checkIndexStr)
if err != nil {
c.Ui.Error(fmt.Sprintf("Error parsing check-index value %q: %v", checkIndexStr, err))
return 1
}
// Submit the job
var evalID string
if enforce {
evalID, _, err = client.Jobs().EnforceRegister(apiJob, checkIndex, nil)
} else {
evalID, _, err = client.Jobs().Register(apiJob, nil)
}
if err != nil {
if strings.Contains(err.Error(), api.RegisterEnforceIndexErrPrefix) {
// Format the error specially if the error is due to index
// enforcement
matches := enforceIndexRegex.FindStringSubmatch(err.Error())
if len(matches) == 2 {
c.Ui.Error(matches[1]) // The matched group
c.Ui.Error("Job not updated")
return 1
}
}
c.Ui.Error(fmt.Sprintf("Error submitting job: %s", err))
return 1
}
// Check if we should enter monitor mode
if detach || periodic {
c.Ui.Output("Job registration successful")
if periodic {
now := time.Now().UTC()
next := job.Periodic.Next(now)
c.Ui.Output(fmt.Sprintf("Approximate next launch time: %s (%s from now)",
formatTime(next), formatTimeDifference(now, next, time.Second)))
} else {
c.Ui.Output("Evaluation ID: " + evalID)
}
return 0
}
// Detach was not specified, so start monitoring
mon := newMonitor(c.Ui, client, length)
return mon.monitor(evalID, false)
}
func (c *PlanCommand) Run(args []string) int {
var diff, verbose bool
flags := c.Meta.FlagSet("plan", FlagSetClient)
flags.Usage = func() { c.Ui.Output(c.Help()) }
flags.BoolVar(&diff, "diff", true, "")
flags.BoolVar(&verbose, "verbose", false, "")
if err := flags.Parse(args); err != nil {
return 1
}
// Check that we got exactly one job
args = flags.Args()
if len(args) != 1 {
c.Ui.Error(c.Help())
return 1
}
file := args[0]
// Parse the job file
job, err := jobspec.ParseFile(file)
if err != nil {
c.Ui.Error(fmt.Sprintf("Error parsing job file %s: %s", file, err))
return 1
}
// Initialize any fields that need to be.
job.InitFields()
// Check that the job is valid
if err := job.Validate(); err != nil {
c.Ui.Error(fmt.Sprintf("Error validating job: %s", err))
return 1
}
// Convert it to something we can use
apiJob, err := convertStructJob(job)
if err != nil {
c.Ui.Error(fmt.Sprintf("Error converting job: %s", err))
return 1
}
// Get the HTTP client
client, err := c.Meta.Client()
if err != nil {
c.Ui.Error(fmt.Sprintf("Error initializing client: %s", err))
return 1
}
// Force the region to be that of the job.
if r := job.Region; r != "" {
client.SetRegion(r)
}
// Submit the job
resp, _, err := client.Jobs().Plan(apiJob, diff, nil)
if err != nil {
c.Ui.Error(fmt.Sprintf("Error during plan: %s", err))
return 1
}
// Print the diff if not disabled
if diff {
c.Ui.Output(fmt.Sprintf("%s\n",
c.Colorize().Color(strings.TrimSpace(formatJobDiff(resp.Diff, verbose)))))
}
// Print the scheduler dry-run output
c.Ui.Output(c.Colorize().Color("[bold]Scheduler dry-run:[reset]"))
c.Ui.Output(c.Colorize().Color(formatDryRun(resp.FailedTGAllocs, resp.CreatedEvals)))
c.Ui.Output("")
// Print the job index info
c.Ui.Output(c.Colorize().Color(formatJobModifyIndex(resp.JobModifyIndex, file)))
return 0
}
func (c *RunCommand) Run(args []string) int {
var detach bool
flags := c.Meta.FlagSet("run", FlagSetClient)
flags.Usage = func() { c.Ui.Output(c.Help()) }
flags.BoolVar(&detach, "detach", false, "")
if err := flags.Parse(args); err != nil {
return 1
}
// Check that we got exactly one node
args = flags.Args()
if len(args) != 1 {
c.Ui.Error(c.Help())
return 1
}
file := args[0]
// Parse the job file
job, err := jobspec.ParseFile(file)
if err != nil {
c.Ui.Error(fmt.Sprintf("Error parsing job file %s: %s", file, err))
return 1
}
// Check that the job is valid
if err := job.Validate(); err != nil {
c.Ui.Error(fmt.Sprintf("Error validating job: %s", err))
return 1
}
// Convert it to something we can use
apiJob, err := convertJob(job)
if err != nil {
c.Ui.Error(fmt.Sprintf("Error converting job: %s", err))
return 1
}
// Get the HTTP client
client, err := c.Meta.Client()
if err != nil {
c.Ui.Error(fmt.Sprintf("Error initializing client: %s", err))
return 1
}
// Submit the job
evalID, _, err := client.Jobs().Register(apiJob, nil)
if err != nil {
c.Ui.Error(fmt.Sprintf("Error submitting job: %s", err))
return 1
}
// Check if we should enter monitor mode
if detach {
c.Ui.Output("Job registration successful")
c.Ui.Output("Evaluation ID: " + evalID)
return 0
}
// Detach was not specified, so start monitoring
mon := newMonitor(c.Ui, client)
return mon.monitor(evalID)
}
func (c *RunCommand) Run(args []string) int {
var detach, verbose, output bool
flags := c.Meta.FlagSet("run", FlagSetClient)
flags.Usage = func() { c.Ui.Output(c.Help()) }
flags.BoolVar(&detach, "detach", false, "")
flags.BoolVar(&verbose, "verbose", false, "")
flags.BoolVar(&output, "output", false, "")
if err := flags.Parse(args); err != nil {
return 1
}
// Truncate the id unless full length is requested
length := shortId
if verbose {
length = fullId
}
// Check that we got exactly one node
args = flags.Args()
if len(args) != 1 {
c.Ui.Error(c.Help())
return 1
}
file := args[0]
// Parse the job file
job, err := jobspec.ParseFile(file)
if err != nil {
c.Ui.Error(fmt.Sprintf("Error parsing job file %s: %s", file, err))
return 1
}
// Initialize any fields that need to be.
job.InitFields()
// Check that the job is valid
if err := job.Validate(); err != nil {
c.Ui.Error(fmt.Sprintf("Error validating job: %s", err))
return 1
}
// Check if the job is periodic.
periodic := job.IsPeriodic()
// Convert it to something we can use
apiJob, err := convertStructJob(job)
if err != nil {
c.Ui.Error(fmt.Sprintf("Error converting job: %s", err))
return 1
}
if output {
req := api.RegisterJobRequest{apiJob}
buf, err := json.MarshalIndent(req, "", " ")
if err != nil {
c.Ui.Error(fmt.Sprintf("Error converting job: %s", err))
return 1
}
c.Ui.Output(string(buf))
return 0
}
// Get the HTTP client
client, err := c.Meta.Client()
if err != nil {
c.Ui.Error(fmt.Sprintf("Error initializing client: %s", err))
return 1
}
// Submit the job
evalID, _, err := client.Jobs().Register(apiJob, nil)
if err != nil {
c.Ui.Error(fmt.Sprintf("Error submitting job: %s", err))
return 1
}
// Check if we should enter monitor mode
if detach || periodic {
c.Ui.Output("Job registration successful")
if periodic {
c.Ui.Output(fmt.Sprintf("Approximate next launch time: %v", job.Periodic.Next(time.Now().UTC())))
} else {
c.Ui.Output("Evaluation ID: " + evalID)
}
return 0
}
// Detach was not specified, so start monitoring
mon := newMonitor(c.Ui, client, length)
return mon.monitor(evalID, false)
}
func handleRun() int {
// Parse the job file
job, err := jobspec.ParseFile(jobFile)
if err != nil {
log.Fatalf("[ERR] nomad: failed parsing job file: %v", err)
}
// Convert to an API struct for submission
apiJob, err := convertStructJob(job)
if err != nil {
log.Fatalf("[ERR] nomad: failed converting job: %v", err)
}
jobID := apiJob.ID
// Get the API client
client, err := api.NewClient(api.DefaultConfig())
if err != nil {
log.Fatalf("[ERR] nomad: failed creating nomad client: %v", err)
}
jobs := client.Jobs()
jobSubmitters := 64
if numJobs < jobSubmitters {
jobSubmitters = numJobs
}
log.Printf("[DEBUG] nomad: using %d parallel job submitters", jobSubmitters)
// Submit the job the requested number of times
errCh := make(chan error, numJobs)
stopCh := make(chan struct{})
jobsCh := make(chan *api.Job, jobSubmitters)
defer close(stopCh)
for i := 0; i < jobSubmitters; i++ {
go submitJobs(jobs, jobsCh, stopCh, errCh)
}
log.Printf("[DEBUG] nomad: submitting %d jobs", numJobs)
submitting := make(map[string]*api.Job, numJobs)
for i := 0; i < numJobs; i++ {
copy, err := copystructure.Copy(apiJob)
if err != nil {
log.Fatalf("[ERR] nomad: failed to copy api job: %v", err)
}
// Increment the job ID
jobCopy := copy.(*api.Job)
jobCopy.ID = fmt.Sprintf("%s-%d", jobID, i)
submitting[jobCopy.ID] = jobCopy
jobsCh <- jobCopy
}
// Collect errors if any
for i := 0; i < numJobs; i++ {
select {
case err := <-errCh:
if err != nil {
log.Fatalf("[ERR] nomad: failed submitting job: %v", err)
}
case <-stopCh:
return 0
}
}
// Get the jobs were submitted.
submitted, _, err := jobs.List(nil)
if err != nil {
log.Fatalf("[ERR] nomad: failed listing jobs: %v", err)
}
// See if anything didn't get registered
for _, job := range submitted {
delete(submitting, job.ID)
}
// Resubmitting anything missed
for id, missed := range submitting {
log.Printf("[DEBUG] nomad: failed submitting job %q; retrying", id)
_, _, err := jobs.Register(missed, nil)
if err != nil {
log.Printf("[ERR] nomad: failed submitting job: %v", err)
}
}
return 0
}
func handleStatus() int {
// Parse the job file to get the total expected allocs
job, err := jobspec.ParseFile(jobFile)
if err != nil {
log.Fatalf("[ERR] nomad: failed parsing job file: %v", err)
}
var totalAllocs int
for _, group := range job.TaskGroups {
totalAllocs += (group.Count * len(group.Tasks))
}
totalAllocs *= numJobs
minEvals := numJobs
log.Printf("[DEBUG] nomad: expecting %d allocs (%d evals minimum)", totalAllocs, minEvals)
// Determine the set of jobs we should track.
jobs := make(map[string]struct{})
for i := 0; i < numJobs; i++ {
// Increment the job ID
jobs[fmt.Sprintf("%s-%d", job.ID, i)] = struct{}{}
}
// Get the API client
client, err := api.NewClient(api.DefaultConfig())
if err != nil {
log.Fatalf("[ERR] nomad: failed creating nomad client: %v", err)
}
evalEndpoint := client.Evaluations()
// Set up the args
args := &api.QueryOptions{
AllowStale: true,
}
// Wait for all the evals to be complete.
cutoff := time.Now().Add(maxWait)
evals := make(map[string]*api.Evaluation, minEvals)
failedEvals := make(map[string]struct{})
blockedEvals := make(map[string]int)
EVAL_POLL:
for {
waitTime, exceeded := getSleepTime(cutoff)
if !exceeded {
log.Printf("[DEBUG] nomad: next eval poll in %s", waitTime)
time.Sleep(waitTime)
}
// Start the query
resp, _, err := evalEndpoint.List(args)
if err != nil {
// Only log and continue to skip minor errors
log.Printf("[ERR] nomad: failed querying evals: %v", err)
continue
}
// Filter out evaluations that aren't for the jobs we are tracking.
var filter []*api.Evaluation
for _, eval := range resp {
if _, ok := jobs[eval.JobID]; ok {
filter = append(filter, eval)
}
}
// Wait til all evals have gone through the scheduler.
if n := len(filter); n < minEvals {
log.Printf("[DEBUG] nomad: expect %d evals, have %d, polling again",
minEvals, n)
continue
}
// Ensure that all the evals are terminal, otherwise new allocations
// could be made.
needPoll := false
for _, eval := range filter {
switch eval.Status {
case "failed":
failedEvals[eval.ID] = struct{}{}
case "complete":
evals[eval.ID] = eval
case "canceled":
// Do nothing since it was a redundant eval.
case "blocked":
blockedEvals[eval.ID]++
tries := blockedEvals[eval.ID]
if tries < blockedEvalTries {
needPoll = true
} else if tries == blockedEvalTries {
log.Printf("[DEBUG] nomad: abandoning blocked eval %q", eval.ID)
}
case "pending":
needPoll = true
}
}
if needPoll && !exceeded {
continue EVAL_POLL
}
break
}
// We now have all the evals, gather the allocations and placement times.
// scheduleTime is a map of alloc ID to map of desired status and time.
scheduleTimes := make(map[string]map[string]int64, totalAllocs)
startTimes := make(map[string]int64, totalAllocs) // When a task was started
receivedTimes := make(map[string]int64, totalAllocs) // When a task was received by the client
failedAllocs := make(map[string]int64) // Time an alloc failed
failedReason := make(map[string]string) // Reason an alloc failed
pendingAllocs := make(map[string]int) // Counts how many time the alloc was in pending state
first := true
ALLOC_POLL:
for {
waitTime, exceeded := getSleepTime(cutoff)
if !exceeded && !first {
log.Printf("[DEBUG] nomad: next eval poll in %s", waitTime)
time.Sleep(waitTime)
}
first = false
needPoll := false
for evalID := range evals {
// Start the query
resp, _, err := evalEndpoint.Allocations(evalID, args)
if err != nil {
// Only log and continue to skip minor errors
log.Printf("[ERR] nomad: failed querying allocations: %v", err)
continue
}
for _, alloc := range resp {
// Capture the schedule time.
allocTimes, ok := scheduleTimes[alloc.ID]
if !ok {
allocTimes = make(map[string]int64, 3)
scheduleTimes[alloc.ID] = allocTimes
}
allocTimes[alloc.DesiredStatus] = alloc.CreateTime
// Ensure that they have started or have failed.
switch alloc.ClientStatus {
case "failed":
failedAllocs[alloc.ID] = alloc.CreateTime
var failures []string
for _, state := range alloc.TaskStates {
if state.State == "failed" {
failures = append(failures, state.Events[0].DriverError)
}
}
failedReason[alloc.ID] = strings.Join(failures, ",")
continue
case "pending":
pendingAllocs[alloc.ID]++
tries := pendingAllocs[alloc.ID]
if tries < pendingAllocTries {
needPoll = true
} else if tries == pendingAllocTries {
log.Printf("[DEBUG] nomad: abandoning alloc %q", alloc.ID)
}
continue
}
// Detect the start time.
for _, state := range alloc.TaskStates {
if len(state.Events) == 0 {
needPoll = true
}
for _, event := range state.Events {
time := event.Time
switch event.Type {
case "Started":
startTimes[alloc.ID] = time
case "Received":
receivedTimes[alloc.ID] = time
}
}
}
}
}
if needPoll && !exceeded {
continue ALLOC_POLL
}
break
}
// Print the failure reasons for client allocs.
for id, reason := range failedReason {
log.Printf("[DEBUG] nomad: alloc id %q failed on client: %v", id, reason)
}
// Print the results.
if l := len(failedEvals); l != 0 {
fmt.Fprintf(os.Stdout, "failed_evals|%f\n", float64(l))
}
for time, count := range accumTimes(failedAllocs) {
fmt.Fprintf(os.Stdout, "failed_allocs|%f|%d\n", float64(count), time)
}
for time, count := range accumTimes(startTimes) {
fmt.Fprintf(os.Stdout, "running|%f|%d\n", float64(count), time)
}
for time, count := range accumTimes(receivedTimes) {
fmt.Fprintf(os.Stdout, "received|%f|%d\n", float64(count), time)
}
for time, count := range accumTimesOn("run", scheduleTimes) {
fmt.Fprintf(os.Stdout, "placed_run|%f|%d\n", float64(count), time)
}
for time, count := range accumTimesOn("failed", scheduleTimes) {
fmt.Fprintf(os.Stdout, "placed_failed|%f|%d\n", float64(count), time)
}
for time, count := range accumTimesOn("stop", scheduleTimes) {
fmt.Fprintf(os.Stdout, "placed_stop|%f|%d\n", float64(count), time)
}
// Aggregate eval triggerbys.
triggers := make(map[string]int, len(evals))
for _, eval := range evals {
triggers[eval.TriggeredBy]++
}
for trigger, count := range triggers {
fmt.Fprintf(os.Stdout, "trigger:%s|%f\n", trigger, float64(count))
}
// Print if the scheduler changed scheduling decisions
flips := make(map[string]map[string]int64) // alloc id -> map[flipType]time
flipTypes := make(map[string]struct{})
for id, decisions := range scheduleTimes {
if len(decisions) < 2 {
continue
}
// Have decision -> time
// 1) time -> decision
// 2) sort times
// 3) print transitions
flips[id] = make(map[string]int64)
inverted := make(map[int64]string, len(decisions))
times := make([]int, 0, len(decisions))
for k, v := range decisions {
inverted[v] = k
times = append(times, int(v))
}
sort.Ints(times)
for i := 1; i < len(times); i++ {
from := decisions[inverted[int64(times[i-1])]]
to := decisions[inverted[int64(times[i])]]
flipType := fmt.Sprintf("%s-to-%s", from, to)
flips[id][flipType] = int64(times[i])
flipTypes[flipType] = struct{}{}
}
}
for flipType, _ := range flips {
for time, count := range accumTimesOn(flipType, flips) {
fmt.Fprintf(os.Stdout, "%v|%f|%d\n", flipType, float64(count), time)
}
}
return 0
}
