Show failure reason in status

This commit is contained in:
Alex Dadgar
2016-05-25 17:06:20 -07:00
parent 38dbe768e7
commit edbbcd5deb
2 changed files with 109 additions and 19 deletions

View File

@@ -2,6 +2,7 @@ package command
import (
"fmt"
"sort"
"strings"
"github.com/hashicorp/nomad/api"
@@ -151,14 +152,18 @@ func (c *EvalStatusCommand) Run(args []string) int {
c.Ui.Output(formatKV(basic))
if failures {
c.Ui.Output("\n==> Failed Allocations")
for tg, metrics := range eval.FailedTGAllocs {
c.Ui.Output("\n==> Failed Placements")
sorted := sortedTaskGroupFromMetrics(eval.FailedTGAllocs)
for _, tg := range sorted {
metrics := eval.FailedTGAllocs[tg]
noun := "allocation"
if metrics.CoalescedFailures > 0 {
noun += "s"
}
c.Ui.Output(fmt.Sprintf("Task Group %q (failed to place %d %s):", tg, metrics.CoalescedFailures+1, noun))
dumpAllocMetrics(c.Ui, metrics, false)
c.Ui.Output("")
}
if eval.BlockedEval != "" {
@@ -170,6 +175,15 @@ func (c *EvalStatusCommand) Run(args []string) int {
return 0
}
func sortedTaskGroupFromMetrics(groups map[string]*api.AllocationMetric) []string {
tgs := make([]string, 0, len(groups))
for tg, _ := range groups {
tgs = append(tgs, tg)
}
sort.Strings(tgs)
return tgs
}
func getTriggerDetails(eval *api.Evaluation) (noun, subject string) {
switch eval.TriggeredBy {
case "job-register", "job-deregister", "periodic-job", "rolling-update":

View File

@@ -11,9 +11,16 @@ import (
"github.com/hashicorp/nomad/nomad/structs"
)
const (
// maxFailedTGs is the maximum number of task groups we show failure reasons
// for before defering to eval-status
maxFailedTGs = 5
)
type StatusCommand struct {
Meta
length int
length int
showEvals, verbose bool
}
func (c *StatusCommand) Help() string {
@@ -31,8 +38,10 @@ Status Options:
-short
Display short output. Used only when a single job is being
queried, and drops verbose information about allocations
and evaluations.
queried, and drops verbose information about allocations.
-evals
Display the evaluations associated with the job.
-verbose
Display full information.
@@ -45,12 +54,13 @@ func (c *StatusCommand) Synopsis() string {
}
func (c *StatusCommand) Run(args []string) int {
var short, verbose bool
var short bool
flags := c.Meta.FlagSet("status", FlagSetClient)
flags.Usage = func() { c.Ui.Output(c.Help()) }
flags.BoolVar(&short, "short", false, "")
flags.BoolVar(&verbose, "verbose", false, "")
flags.BoolVar(&c.showEvals, "evals", false, "")
flags.BoolVar(&c.verbose, "verbose", false, "")
if err := flags.Parse(args); err != nil {
return 1
@@ -65,7 +75,7 @@ func (c *StatusCommand) Run(args []string) int {
// Truncate the id unless full length is requested
c.length = shortId
if verbose {
if c.verbose {
c.length = fullId
}
@@ -221,27 +231,65 @@ func (c *StatusCommand) outputPeriodicInfo(client *api.Client, job *api.Job) err
func (c *StatusCommand) outputJobInfo(client *api.Client, job *api.Job) error {
var evals, allocs []string
// Query the evaluations
jobEvals, _, err := client.Jobs().Evaluations(job.ID, nil)
if err != nil {
return fmt.Errorf("Error querying job evaluations: %s", err)
}
// Query the allocations
jobAllocs, _, err := client.Jobs().Allocations(job.ID, nil)
if err != nil {
return fmt.Errorf("Error querying job allocations: %s", err)
}
// Query the evaluations
jobEvals, _, err := client.Jobs().Evaluations(job.ID, nil)
if err != nil {
return fmt.Errorf("Error querying job evaluations: %s", err)
}
// Determine latest evaluation with failures whose follow up hasn't
// completed.
evalsByID := make(map[string]*api.Evaluation, len(jobEvals))
for _, eval := range jobEvals {
evalsByID[eval.ID] = eval
}
var latestFailedPlacement *api.Evaluation
for _, eval := range evalsByID {
if len(eval.FailedTGAllocs) == 0 {
// Skip evals without failures
continue
}
// Check if created blocked eval is finished
if blocked, ok := evalsByID[eval.BlockedEval]; ok {
if blocked.Status == "complete" {
continue
}
}
if latestFailedPlacement == nil || latestFailedPlacement.CreateIndex < eval.CreateIndex {
latestFailedPlacement = eval
}
}
// Format the evals
evals = make([]string, len(jobEvals)+1)
evals[0] = "ID|Priority|Triggered By|Status"
evals[0] = "ID|Priority|Triggered By|Status|Placement Failures"
for i, eval := range jobEvals {
evals[i+1] = fmt.Sprintf("%s|%d|%s|%s",
evals[i+1] = fmt.Sprintf("%s|%d|%s|%s|%t",
limit(eval.ID, c.length),
eval.Priority,
eval.TriggeredBy,
eval.Status)
eval.Status,
len(eval.FailedTGAllocs) != 0,
)
}
if c.verbose || c.showEvals {
c.Ui.Output("\n==> Evaluations")
c.Ui.Output(formatList(evals))
}
if latestFailedPlacement != nil {
c.outputFailedPlacements(latestFailedPlacement)
}
// Format the allocs
@@ -257,13 +305,41 @@ func (c *StatusCommand) outputJobInfo(client *api.Client, job *api.Job) error {
alloc.ClientStatus)
}
c.Ui.Output("\n==> Evaluations")
c.Ui.Output(formatList(evals))
c.Ui.Output("\n==> Allocations")
c.Ui.Output(formatList(allocs))
return nil
}
func (c *StatusCommand) outputFailedPlacements(failedEval *api.Evaluation) {
if failedEval == nil || len(failedEval.FailedTGAllocs) == 0 {
return
}
c.Ui.Output("\n==> Last Placement Failure")
sorted := sortedTaskGroupFromMetrics(failedEval.FailedTGAllocs)
for i, tg := range sorted {
if i >= maxFailedTGs {
break
}
metrics := failedEval.FailedTGAllocs[tg]
noun := "allocation"
if metrics.CoalescedFailures > 0 {
noun += "s"
}
c.Ui.Output(fmt.Sprintf("Task Group %q (failed to place %d %s):", tg, metrics.CoalescedFailures+1, noun))
dumpAllocMetrics(c.Ui, metrics, false)
c.Ui.Output("")
}
if len(sorted) > maxFailedTGs {
trunc := fmt.Sprintf("Placement failures truncated. To see remainder run:\nnomad eval-status %s", failedEval.ID)
c.Ui.Output(trunc)
}
}
// convertApiJob is used to take a *api.Job and convert it to an *struct.Job.
// This function is just a hammer and probably needs to be revisited.
func convertApiJob(in *api.Job) (*structs.Job, error) {