Added metrics to track task/alloc start/restarts/dead events

This commit is contained in:
Diptanu Choudhury
2017-08-19 01:28:48 -07:00
parent d710ecdfe5
commit 45583d757e

View File

@@ -9,6 +9,7 @@ import (
"sync"
"time"
metrics "github.com/armon/go-metrics"
"github.com/boltdb/bolt"
"github.com/hashicorp/go-multierror"
"github.com/hashicorp/nomad/client/allocdir"
@@ -645,6 +646,7 @@ func (r *AllocRunner) setTaskState(taskName, state string, event *structs.TaskEv
taskState.Failed = true
}
if event.Type == structs.TaskRestarting {
metrics.IncrCounter([]string{"client", "allocs", r.alloc.Job.Name, r.alloc.TaskGroup, taskName, "restart"}, 1)
taskState.Restarts++
taskState.LastRestart = time.Unix(0, event.Time)
}
@@ -668,12 +670,14 @@ func (r *AllocRunner) setTaskState(taskName, state string, event *structs.TaskEv
// Capture the start time if it is just starting
if taskState.State != structs.TaskStateRunning {
taskState.StartedAt = time.Now().UTC()
metrics.IncrCounter([]string{"client", "allocs", r.alloc.Job.Name, r.alloc.TaskGroup, taskName, "running"}, 1)
}
case structs.TaskStateDead:
// Capture the finished time. If it has never started there is no finish
// time
if !taskState.StartedAt.IsZero() {
taskState.FinishedAt = time.Now().UTC()
metrics.IncrCounter([]string{"client", "allocs", r.alloc.Job.Name, r.alloc.TaskGroup, taskName, "dead"}, 1)
}
// Find all tasks that are not the one that is dead and check if the one
@@ -740,6 +744,9 @@ func (r *AllocRunner) Run() {
defer close(r.waitCh)
go r.dirtySyncState()
// Incr alloc runner start counter
metrics.IncrCounter([]string{"client", "allocs", r.alloc.Job.Name, r.alloc.TaskGroup, "start"}, 1)
// Find the task group to run in the allocation
alloc := r.Alloc()
tg := alloc.Job.LookupTaskGroup(alloc.TaskGroup)
@@ -922,6 +929,14 @@ func (r *AllocRunner) handleDestroy() {
// state as we wait for a destroy.
alloc := r.Alloc()
// Incr the alloc destroy counter
metrics.IncrCounter([]string{"client", "allocs", r.alloc.Job.Name, r.alloc.TaskGroup, "destroy"}, 1)
//TODO(schmichael) updater can cause a GC which can block on this alloc
// runner shutting down. Since handleDestroy can be called by Run() we
// can't block shutdown here as it would cause a deadlock.
go r.updater(alloc)
// Broadcast and persist state synchronously
r.sendBroadcast(alloc)
if err := r.saveAllocRunnerState(); err != nil {