forked from influxdata/kapacitor
-
Notifications
You must be signed in to change notification settings - Fork 0
/
edge.go
238 lines (212 loc) · 5.16 KB
/
edge.go
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
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
package kapacitor
import (
"errors"
"fmt"
"log"
"sync"
"github.com/influxdata/kapacitor/expvar"
"github.com/influxdata/kapacitor/models"
"github.com/influxdata/kapacitor/pipeline"
)
const (
statCollected = "collected"
statEmitted = "emitted"
defaultEdgeBufferSize = 1000
)
var ErrAborted = errors.New("edged aborted")
type StreamCollector interface {
CollectPoint(models.Point) error
Close()
}
type BatchCollector interface {
CollectBatch(models.Batch) error
Close()
}
type Edge struct {
mu sync.Mutex
closed bool
stream chan models.Point
batch chan models.Batch
logger *log.Logger
aborted chan struct{}
statsKey string
collected *expvar.Int
emitted *expvar.Int
statMap *expvar.Map
groupMu sync.RWMutex
groupStats map[models.GroupID]*edgeStat
}
func newEdge(taskName, parentName, childName string, t pipeline.EdgeType, size int, logService LogService) *Edge {
tags := map[string]string{
"task": taskName,
"parent": parentName,
"child": childName,
"type": t.String(),
}
key, sm := NewStatistics("edges", tags)
collected := &expvar.Int{}
emitted := &expvar.Int{}
sm.Set(statCollected, collected)
sm.Set(statEmitted, emitted)
e := &Edge{
statsKey: key,
statMap: sm,
collected: collected,
emitted: emitted,
aborted: make(chan struct{}),
groupStats: make(map[models.GroupID]*edgeStat),
}
name := fmt.Sprintf("%s|%s->%s", taskName, parentName, childName)
e.logger = logService.NewLogger(fmt.Sprintf("[edge:%s] ", name), log.LstdFlags)
switch t {
case pipeline.StreamEdge:
e.stream = make(chan models.Point, size)
case pipeline.BatchEdge:
e.batch = make(chan models.Batch, size)
}
return e
}
func (e *Edge) emittedCount() int64 {
return e.emitted.IntValue()
}
func (e *Edge) collectedCount() int64 {
return e.collected.IntValue()
}
// Stats for a given group for this edge
type edgeStat struct {
collected int64
emitted int64
tags models.Tags
dims models.Dimensions
}
// Get a snapshot of the current group statistics for this edge
func (e *Edge) readGroupStats(f func(group models.GroupID, collected, emitted int64, tags models.Tags, dims models.Dimensions)) {
e.groupMu.RLock()
defer e.groupMu.RUnlock()
for group, stats := range e.groupStats {
f(
group,
stats.collected,
stats.emitted,
stats.tags,
stats.dims,
)
}
}
// Close the edge, this can only be called after all
// collect calls to the edge have finished.
// Can be called multiple times.
func (e *Edge) Close() {
e.mu.Lock()
defer e.mu.Unlock()
if e.closed {
return
}
e.closed = true
e.logger.Printf(
"D! closing c: %d e: %d\n",
e.collected.IntValue(),
e.emitted.IntValue(),
)
if e.stream != nil {
close(e.stream)
}
if e.batch != nil {
close(e.batch)
}
DeleteStatistics(e.statsKey)
}
// Abort all next and collect calls.
// Items in flight may or may not be processed.
func (e *Edge) Abort() {
close(e.aborted)
e.logger.Printf(
"I! aborting c: %d e: %d\n",
e.collected.IntValue(),
e.emitted.IntValue(),
)
}
func (e *Edge) Next() (p models.PointInterface, ok bool) {
if e.stream != nil {
return e.NextPoint()
}
return e.NextBatch()
}
func (e *Edge) NextPoint() (p models.Point, ok bool) {
select {
case <-e.aborted:
case p, ok = <-e.stream:
if ok {
e.emitted.Add(1)
e.incEmitted(p.Group, p.Tags, p.Dimensions, 1)
}
}
return
}
func (e *Edge) NextBatch() (b models.Batch, ok bool) {
select {
case <-e.aborted:
case b, ok = <-e.batch:
if ok {
e.emitted.Add(1)
e.incEmitted(b.Group, b.Tags, b.PointDimensions(), int64(len(b.Points)))
}
}
return
}
func (e *Edge) CollectPoint(p models.Point) error {
e.collected.Add(1)
e.incCollected(p.Group, p.Tags, p.Dimensions, 1)
select {
case <-e.aborted:
return ErrAborted
case e.stream <- p:
return nil
}
}
func (e *Edge) CollectBatch(b models.Batch) error {
e.collected.Add(1)
e.incCollected(b.Group, b.Tags, b.PointDimensions(), int64(len(b.Points)))
select {
case <-e.aborted:
return ErrAborted
case e.batch <- b:
return nil
}
}
// Increment the emitted count of the group for this edge.
func (e *Edge) incEmitted(group models.GroupID, tags models.Tags, dims models.Dimensions, count int64) {
// we are "manually" calling Unlock() and not using defer, because this method is called
// in hot locations (NextPoint/CollectPoint) and defer have some performance penalty
e.groupMu.Lock()
if stats, ok := e.groupStats[group]; ok {
stats.emitted += count
e.groupMu.Unlock()
} else {
stats = &edgeStat{
emitted: count,
tags: tags,
dims: dims,
}
e.groupStats[group] = stats
e.groupMu.Unlock()
}
}
// Increment the collected count of the group for this edge.
func (e *Edge) incCollected(group models.GroupID, tags models.Tags, dims models.Dimensions, count int64) {
// we are "manually" calling Unlock() and not using defer, because this method is called
// in hot locations (NextPoint/CollectPoint) and defer have some performance penalty
e.groupMu.Lock()
if stats, ok := e.groupStats[group]; ok {
stats.collected += count
e.groupMu.Unlock()
} else {
stats = &edgeStat{
collected: count,
tags: tags,
dims: dims,
}
e.groupStats[group] = stats
e.groupMu.Unlock()
}
}