1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
|
package worker_watcher //nolint:golint,stylecheck
import (
"context"
"sync"
"github.com/spiral/errors"
"github.com/spiral/roadrunner/v2/pkg/events"
"github.com/spiral/roadrunner/v2/pkg/worker"
)
// workerCreateFunc can be nil, but in that case, dead stack will not be replaced
func NewSyncWorkerWatcher(allocator worker.Allocator, numWorkers uint64, events events.Handler) Watcher {
ww := &workerWatcher{
stack: NewWorkersStack(numWorkers),
allocator: allocator,
events: events,
}
return ww
}
type workerWatcher struct {
mutex sync.RWMutex
stack *Stack
allocator worker.Allocator
events events.Handler
}
func (ww *workerWatcher) Watch(workers []worker.SyncWorker) error {
for i := 0; i < len(workers); i++ {
ww.stack.Push(workers[i])
go func(swc worker.SyncWorker) {
ww.wait(swc)
}(workers[i])
}
return nil
}
// Get is not a thread safe operation
func (ww *workerWatcher) Get(ctx context.Context) (worker.SyncWorker, error) {
const op = errors.Op("worker_watcher_get_free_worker")
// FAST PATH
// thread safe operation
w, stop := ww.stack.Pop()
if stop {
return nil, errors.E(op, errors.WatcherStopped)
}
// fast path, worker not nil and in the ReadyState
if w != nil && w.State().Value() == worker.StateReady {
return w, nil
}
// =========================================================
// SLOW PATH
// no free workers in the stack
// try to continuously get free one
for {
select {
default:
w, stop = ww.stack.Pop()
if stop {
return nil, errors.E(op, errors.WatcherStopped)
}
if w == nil {
continue
}
switch w.State().Value() {
case worker.StateRemove:
err := ww.Remove(w)
if err != nil {
return nil, errors.E(op, err)
}
// try to get next
continue
case
// all the possible wrong states
worker.StateInactive,
worker.StateDestroyed,
worker.StateErrored,
worker.StateStopped,
worker.StateInvalid,
worker.StateKilling,
worker.StateWorking, // ??? how
worker.StateStopping:
// worker doing no work because it in the stack
// so we can safely kill it (inconsistent state)
_ = w.Kill()
// try to get new worker
continue
// return only workers in the Ready state
case worker.StateReady:
return w, nil
}
case <-ctx.Done():
return nil, errors.E(op, errors.NoFreeWorkers, errors.Str("no free workers in the stack, timeout exceed"))
}
}
}
func (ww *workerWatcher) Allocate() error {
ww.stack.mutex.Lock()
const op = errors.Op("worker_watcher_allocate_new")
sw, err := ww.allocator()
if err != nil {
return errors.E(op, errors.WorkerAllocate, err)
}
ww.addToWatch(sw)
ww.stack.mutex.Unlock()
ww.Push(sw)
return nil
}
// Remove
func (ww *workerWatcher) Remove(wb worker.SyncWorker) error {
ww.mutex.Lock()
defer ww.mutex.Unlock()
const op = errors.Op("worker_watcher_remove_worker")
// set remove state
wb.State().Set(worker.StateRemove)
if ww.stack.FindAndRemoveByPid(wb.Pid()) {
err := wb.Kill()
if err != nil {
return errors.E(op, err)
}
return nil
}
return nil
}
// O(1) operation
func (ww *workerWatcher) Push(w worker.SyncWorker) {
ww.mutex.Lock()
defer ww.mutex.Unlock()
ww.stack.Push(w)
}
// Destroy all underlying stack (but let them to complete the task)
func (ww *workerWatcher) Destroy(ctx context.Context) {
// destroy stack, we don't use ww mutex here, since we should be able to push worker
ww.stack.Destroy(ctx)
}
// Warning, this is O(n) operation, and it will return copy of the actual workers
func (ww *workerWatcher) List() []worker.SyncWorker {
return ww.stack.Workers()
}
func (ww *workerWatcher) wait(w worker.BaseProcess) {
const op = errors.Op("worker_watcher_wait")
err := w.Wait()
if err != nil {
ww.events.Push(events.WorkerEvent{
Event: events.EventWorkerError,
Worker: w,
Payload: errors.E(op, err),
})
}
if w.State().Value() == worker.StateDestroyed {
// worker was manually destroyed, no need to replace
ww.events.Push(events.PoolEvent{Event: events.EventWorkerDestruct, Payload: w})
return
}
_ = ww.stack.FindAndRemoveByPid(w.Pid())
err = ww.Allocate()
if err != nil {
ww.events.Push(events.PoolEvent{
Event: events.EventPoolError,
Payload: errors.E(op, err),
})
}
}
func (ww *workerWatcher) addToWatch(wb worker.SyncWorker) {
go func() {
ww.wait(wb)
}()
}
|