MM-27199 - Add metrics to MM server for total enabled user count (#15116)

* Add metrics to MM server for total enabled user count and include installation ID from a new env var

* diagnostics context

Co-authored-by: Mattermod <mattermod@users.noreply.github.com>

jobs/active_users/scheduler.go

@@ -0,0 +1,51 @@
+// Copyright (c) 2015-present Mattermost, Inc. All Rights Reserved.
+// See LICENSE.txt for license information.
+
+package active_users
+
+import (
+	"time"
+
+	"github.com/mattermost/mattermost-server/v5/app"
+	"github.com/mattermost/mattermost-server/v5/model"
+)
+
+const (
+	SchedFreqMinutes = 10
+)
+
+type Scheduler struct {
+	App *app.App
+}
+
+func (m *ActiveUsersJobInterfaceImpl) MakeScheduler() model.Scheduler {
+	return &Scheduler{m.App}
+}
+
+func (scheduler *Scheduler) Name() string {
+	return JobName + "Scheduler"
+}
+
+func (scheduler *Scheduler) JobType() string {
+	return model.JOB_TYPE_ACTIVE_USERS
+}
+
+func (scheduler *Scheduler) Enabled(cfg *model.Config) bool {
+	// Only enabled when Metrics are enabled.
+	return *cfg.MetricsSettings.Enable
+}
+
+func (scheduler *Scheduler) NextScheduleTime(cfg *model.Config, now time.Time, pendingJobs bool, lastSuccessfulJob *model.Job) *time.Time {
+	nextTime := time.Now().Add(SchedFreqMinutes * time.Minute)
+	return &nextTime
+}
+
+func (scheduler *Scheduler) ScheduleJob(cfg *model.Config, pendingJobs bool, lastSuccessfulJob *model.Job) (*model.Job, *model.AppError) {
+	data := map[string]string{}
+
+	if job, err := scheduler.App.Srv().Jobs.CreateJob(model.JOB_TYPE_ACTIVE_USERS, data); err != nil {
+		return nil, err
+	} else {
+		return job, nil
+	}
+}

jobs/active_users/worker.go

@@ -0,0 +1,117 @@
+// Copyright (c) 2015-present Mattermost, Inc. All Rights Reserved.
+// See LICENSE.txt for license information.
+
+package active_users
+
+import (
+	"github.com/mattermost/mattermost-server/v5/app"
+	"github.com/mattermost/mattermost-server/v5/jobs"
+	tjobs "github.com/mattermost/mattermost-server/v5/jobs/interfaces"
+	"github.com/mattermost/mattermost-server/v5/mlog"
+	"github.com/mattermost/mattermost-server/v5/model"
+)
+
+const (
+	JobName = "ActiveUsers"
+)
+
+type Worker struct {
+	name      string
+	stop      chan bool
+	stopped   chan bool
+	jobs      chan model.Job
+	jobServer *jobs.JobServer
+	app       *app.App
+}
+
+func init() {
+	app.RegisterJobsActiveUsersInterface(func(a *app.App) tjobs.ActiveUsersJobInterface {
+		return &ActiveUsersJobInterfaceImpl{a}
+	})
+}
+
+type ActiveUsersJobInterfaceImpl struct {
+	App *app.App
+}
+
+func (m *ActiveUsersJobInterfaceImpl) MakeWorker() model.Worker {
+	worker := Worker{
+		name:      JobName,
+		stop:      make(chan bool, 1),
+		stopped:   make(chan bool, 1),
+		jobs:      make(chan model.Job),
+		jobServer: m.App.Srv().Jobs,
+		app:       m.App,
+	}
+	return &worker
+}
+
+func (worker *Worker) Run() {
+	mlog.Debug("Worker started", mlog.String("worker", worker.name))
+
+	defer func() {
+		mlog.Debug("Worker finished", mlog.String("worker", worker.name))
+		worker.stopped <- true
+	}()
+
+	for {
+		select {
+		case <-worker.stop:
+			mlog.Debug("Worker received stop signal", mlog.String("worker", worker.name))
+			return
+		case job := <-worker.jobs:
+			mlog.Debug("Worker received a new candidate job.", mlog.String("worker", worker.name))
+			worker.DoJob(&job)
+		}
+	}
+}
+
+func (worker *Worker) Stop() {
+	mlog.Debug("Worker stopping", mlog.String("worker", worker.name))
+	worker.stop <- true
+	<-worker.stopped
+}
+
+func (worker *Worker) JobChannel() chan<- model.Job {
+	return worker.jobs
+}
+
+func (worker *Worker) DoJob(job *model.Job) {
+	if claimed, err := worker.jobServer.ClaimJob(job); err != nil {
+		mlog.Warn("Worker experienced an error while trying to claim job",
+			mlog.String("worker", worker.name),
+			mlog.String("job_id", job.Id),
+			mlog.String("error", err.Error()))
+		return
+	} else if !claimed {
+		return
+	}
+
+	count, err := worker.app.Srv().Store.User().Count(model.UserCountOptions{IncludeDeleted: false})
+
+	if err != nil {
+		mlog.Error("Worker: Failed to get active user count", mlog.String("worker", worker.name), mlog.String("job_id", job.Id), mlog.String("error", err.Error()))
+		worker.setJobError(job, err)
+		return
+	}
+
+	if worker.app.Metrics() != nil {
+		worker.app.Metrics().ObserveEnabledUsers(count)
+	}
+
+	mlog.Info("Worker: Job is complete", mlog.String("worker", worker.name), mlog.String("job_id", job.Id))
+	worker.setJobSuccess(job)
+}
+
+func (worker *Worker) setJobSuccess(job *model.Job) {
+	if err := worker.app.Srv().Jobs.SetJobSuccess(job); err != nil {
+		mlog.Error("Worker: Failed to set success for job", mlog.String("worker", worker.name), mlog.String("job_id", job.Id), mlog.String("error", err.Error()))
+		worker.setJobError(job, err)
+	}
+}
+
+func (worker *Worker) setJobError(job *model.Job, appError *model.AppError) {
+	if err := worker.app.Srv().Jobs.SetJobError(job, appError); err != nil {
+		mlog.Error("Worker: Failed to set job error", mlog.String("worker", worker.name), mlog.String("job_id", job.Id), mlog.String("error", err.Error()))
+	}
+}