grafana/pkg/services/alerting/alerting.go

143 lines
2.8 KiB
Go
Raw Normal View History

package alerting
import (
2016-05-23 16:02:17 +08:00
"math/rand"
"strconv"
"time"
"github.com/grafana/grafana/pkg/log"
m "github.com/grafana/grafana/pkg/models"
"github.com/grafana/grafana/pkg/setting"
"sync"
)
func Init() {
if !setting.AlertingEnabled {
return
}
log.Info("Alerting: Initializing scheduler...")
scheduler := NewScheduler()
2016-05-23 16:59:28 +08:00
go scheduler.Dispatch(&AlertRuleReader{})
go scheduler.Executor(&DummieExecutor{})
}
type Scheduler struct {
jobs []*AlertJob
runQueue chan *AlertJob
mtx sync.RWMutex
2016-05-23 16:02:17 +08:00
2016-05-23 16:59:28 +08:00
alertRuleFetcher RuleReader
2016-05-23 16:02:17 +08:00
serverId string
serverPosition int
clusterSize int
}
func NewScheduler() *Scheduler {
return &Scheduler{
jobs: make([]*AlertJob, 0),
runQueue: make(chan *AlertJob, 1000),
2016-05-23 16:02:17 +08:00
serverId: strconv.Itoa(rand.Intn(1000)),
}
}
func (this *Scheduler) heartBeat() {
2016-05-23 16:02:17 +08:00
//write heartBeat to db.
//get the modulus position of active servers
log.Info("Heartbeat: Sending heartbeat from " + this.serverId)
this.clusterSize = 1
this.serverPosition = 1
2016-05-23 16:02:17 +08:00
}
func (this *Scheduler) Dispatch(reader RuleReader) {
reschedule := time.NewTicker(time.Second * 10)
secondTicker := time.NewTicker(time.Second)
heartbeat := time.NewTicker(time.Second * 5)
this.heartBeat()
this.updateJobs(reader.Fetch)
for {
select {
case <-secondTicker.C:
this.queueJobs()
case <-reschedule.C:
this.updateJobs(reader.Fetch)
case <-heartbeat.C:
this.heartBeat()
}
}
}
func (this *Scheduler) updateJobs(f func() []m.AlertRule) {
2016-05-23 16:59:28 +08:00
log.Debug("Scheduler: UpdateJobs()")
jobs := make([]*AlertJob, 0)
rules := f()
2016-05-23 16:02:17 +08:00
this.mtx.Lock()
defer this.mtx.Unlock()
for i := this.serverPosition - 1; i < len(rules); i += this.clusterSize {
2016-05-23 16:02:17 +08:00
rule := rules[i]
jobs = append(jobs, &AlertJob{
id: rule.Id,
2016-05-23 16:02:17 +08:00
name: rule.Title,
frequency: rule.Frequency,
rule: rule,
offset: int64(len(jobs)),
})
}
log.Debug("Scheduler: Selected %d jobs", len(jobs))
this.jobs = jobs
}
func (this *Scheduler) queueJobs() {
now := time.Now().Unix()
for _, job := range this.jobs {
if now%job.frequency == 0 && job.running == false {
log.Info("Scheduler: Putting job on to run queue: %s", job.name)
this.runQueue <- job
}
}
}
func (this *Scheduler) Executor(executor Executor) {
for job := range this.runQueue {
log.Info("Executor: queue length %d", len(this.runQueue))
log.Info("Executor: executing %s", job.name)
go Measure(executor, job)
}
}
func Measure(exec Executor, rule *AlertJob) {
now := time.Now()
rule.running = true
exec.Execute(rule.rule)
rule.running = true
elapsed := time.Since(now)
log.Info("Schedular: exeuction took %v milli seconds", elapsed.Nanoseconds()/1000000)
}
type AlertJob struct {
id int64
name string
frequency int64
offset int64
delay bool
running bool
2016-05-23 16:02:17 +08:00
rule m.AlertRule
}
2016-05-23 16:59:28 +08:00
type AlertResult struct {
id int64
state string
duration time.Time
}