summaryrefslogtreecommitdiff
path: root/service/watcher/watcher.go
diff options
context:
space:
mode:
Diffstat (limited to 'service/watcher/watcher.go')
-rw-r--r--service/watcher/watcher.go124
1 files changed, 84 insertions, 40 deletions
diff --git a/service/watcher/watcher.go b/service/watcher/watcher.go
index 63dce3d5..08d477fa 100644
--- a/service/watcher/watcher.go
+++ b/service/watcher/watcher.go
@@ -8,11 +8,17 @@ import (
)
const (
- // EventMaxTTL thrown when worker is removed due MaxTTL being reached. Context is roadrunner.WorkerError
- EventMaxTTL = iota + 8000
-
// EventMaxMemory caused when worker consumes more memory than allowed.
- EventMaxMemory
+ EventMaxMemory = iota + 8000
+
+ // EventMaxTTL thrown when worker is removed due TTL being reached. Context is roadrunner.WorkerError
+ EventMaxTTL
+
+ // EventMaxIdleTTL triggered when worker spends too much time at rest.
+ EventMaxIdleTTL
+
+ // EventMaxIdleTTL triggered when worker spends too much time doing the task (max_execution_time).
+ EventMaxExecTTL
)
// handles watcher events
@@ -20,76 +26,114 @@ type listener func(event int, ctx interface{})
// defines the watcher behaviour
type watcherConfig struct {
- // MaxTTL defines maximum time worker is allowed to live.
- MaxTTL time.Duration
-
// MaxMemory defines maximum amount of memory allowed for worker. In megabytes.
MaxMemory uint64
-}
-// Normalize watcher config and upscale the durations.
-func (c *watcherConfig) Normalize() error {
- // Always use second based definition for time durations
- if c.MaxTTL < time.Microsecond {
- c.MaxTTL = time.Second * time.Duration(c.MaxTTL.Nanoseconds())
- }
+ // TTL defines maximum time worker is allowed to live.
+ TTL int64
+
+ // MaxIdleTTL defines maximum duration worker can spend in idle mode.
+ MaxIdleTTL int64
- return nil
+ // MaxExecTTL defines maximum lifetime per job.
+ MaxExecTTL int64
}
type watcher struct {
- lsn listener
- interval time.Duration
- cfg *watcherConfig
- stop chan interface{}
+ lsn listener
+ tick time.Duration
+ cfg *watcherConfig
// list of workers which are currently working
- //working map[*roadrunner.Worker]time.Time
+ sw *stateWatcher
+
+ stop chan interface{}
}
// watch the pool state
-func (watch *watcher) watch(p roadrunner.Pool) {
+func (wch *watcher) watch(p roadrunner.Pool) {
now := time.Now()
+
for _, w := range p.Workers() {
- if watch.cfg.MaxTTL != 0 && now.Sub(w.Created) >= watch.cfg.MaxTTL {
- err := fmt.Errorf("max TTL reached (%s)", watch.cfg.MaxTTL)
+ if w.State().Value() == roadrunner.StateInvalid {
+ // skip duplicate assessment
+ continue
+ }
+
+ s, err := util.WorkerState(w)
+ if err != nil {
+ continue
+ }
+
+ if wch.cfg.TTL != 0 && now.Sub(w.Created).Seconds() >= float64(wch.cfg.TTL) {
+ err := fmt.Errorf("max TTL reached (%vs)", wch.cfg.TTL)
if p.Remove(w, err) {
- watch.report(EventMaxTTL, w, err)
+ wch.report(EventMaxTTL, w, err)
}
+ continue
}
- state, err := util.WorkerState(w)
- if err != nil {
+ if wch.cfg.MaxMemory != 0 && s.MemoryUsage >= wch.cfg.MaxMemory*1024*1024 {
+ err := fmt.Errorf("max allowed memory reached (%vMB)", wch.cfg.MaxMemory)
+ if p.Remove(w, err) {
+ wch.report(EventMaxMemory, w, err)
+ }
continue
}
- if watch.cfg.MaxMemory != 0 && state.MemoryUsage >= watch.cfg.MaxMemory*1024*1024 {
- err := fmt.Errorf("max allowed memory reached (%vMB)", watch.cfg.MaxMemory)
+ // watch the worker state changes
+ wch.sw.push(w)
+ }
+
+ wch.sw.sync(now)
+
+ if wch.cfg.MaxExecTTL != 0 {
+ for _, w := range wch.sw.find(
+ roadrunner.StateWorking,
+ now.Add(-time.Second*time.Duration(wch.cfg.MaxExecTTL)),
+ ) {
+ err := fmt.Errorf("max exec time reached (%vs)", wch.cfg.MaxExecTTL)
+ if p.Remove(w, err) {
+ // brutally
+ go w.Kill()
+ wch.report(EventMaxExecTTL, w, err)
+ }
+ }
+ }
+
+ // locale workers which are in idle mode for too long
+ if wch.cfg.MaxIdleTTL != 0 {
+ for _, w := range wch.sw.find(
+ roadrunner.StateReady,
+ now.Add(-time.Second*time.Duration(wch.cfg.MaxIdleTTL)),
+ ) {
+ err := fmt.Errorf("max idle time reached (%vs)", wch.cfg.MaxIdleTTL)
if p.Remove(w, err) {
- watch.report(EventMaxMemory, w, err)
+ wch.report(EventMaxIdleTTL, w, err)
}
}
}
}
// throw watcher event
-func (watch *watcher) report(event int, worker *roadrunner.Worker, caused error) {
- if watch.lsn != nil {
- watch.lsn(event, roadrunner.WorkerError{Worker: worker, Caused: caused})
+func (wch *watcher) report(event int, worker *roadrunner.Worker, caused error) {
+ if wch.lsn != nil {
+ wch.lsn(event, roadrunner.WorkerError{Worker: worker, Caused: caused})
}
}
// Attach watcher to the pool
-func (watch *watcher) Attach(pool roadrunner.Pool) roadrunner.Watcher {
+func (wch *watcher) Attach(pool roadrunner.Pool) roadrunner.Watcher {
wp := &watcher{
- interval: watch.interval,
- lsn: watch.lsn,
- cfg: watch.cfg,
- stop: make(chan interface{}),
+ tick: wch.tick,
+ lsn: wch.lsn,
+ cfg: wch.cfg,
+ sw: newStateWatcher(),
+ stop: make(chan interface{}),
}
go func(wp *watcher, pool roadrunner.Pool) {
- ticker := time.NewTicker(wp.interval)
+ ticker := time.NewTicker(wp.tick)
for {
select {
case <-ticker.C:
@@ -104,6 +148,6 @@ func (watch *watcher) Attach(pool roadrunner.Pool) roadrunner.Watcher {
}
// Detach watcher from the pool.
-func (watch *watcher) Detach() {
- close(watch.stop)
+func (wch *watcher) Detach() {
+ close(wch.stop)
}