-
Notifications
You must be signed in to change notification settings - Fork 45
/
workerpool.go
234 lines (197 loc) · 6.53 KB
/
workerpool.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
package orchestrator
import (
"context"
"fmt"
"io"
"time"
"github.com/streamingfast/substreams"
"github.com/streamingfast/substreams/block"
pbsubstreams "github.com/streamingfast/substreams/pb/sf/substreams/v1"
"go.uber.org/zap"
"go.uber.org/zap/zapcore"
"google.golang.org/grpc/metadata"
)
type WorkerPool struct {
workers chan *Worker
JobStats map[*Job]*JobStat
}
type JobStat struct {
ModuleName string
StartAt time.Time
RequestRange *block.Range
CurrentBlock uint64
RemainingBlock uint64
BlockCount uint64
BlockSec float64
RemoteHost string
}
func (j *JobStat) update(currentBlock uint64) {
j.CurrentBlock = currentBlock
j.RemainingBlock = j.RequestRange.ExclusiveEndBlock - j.CurrentBlock
j.BlockSec = float64(j.BlockCount) / time.Since(j.StartAt).Seconds()
j.BlockCount = j.CurrentBlock - j.RequestRange.StartBlock
}
func (j *JobStat) MarshalLogObject(enc zapcore.ObjectEncoder) error {
enc.AddString("remote_host", j.RemoteHost)
enc.AddString("module_name", j.ModuleName)
err := enc.AddObject("request_range", j.RequestRange)
if err != nil {
return err
}
enc.AddTime("start_at", j.StartAt)
enc.AddUint64("current_block", j.CurrentBlock)
enc.AddUint64("block_count", j.BlockCount)
enc.AddUint64("remaining_blocks", j.RemainingBlock)
enc.AddFloat64("block_secs", float64(j.BlockCount)/time.Since(j.StartAt).Seconds())
return nil
}
func NewWorkerPool(workerCount int, grpcClientFactory substreams.GrpcClientFactory) *WorkerPool {
zlog.Info("initiating worker pool", zap.Int("worker_count", workerCount))
workers := make(chan *Worker, workerCount)
for i := 0; i < workerCount; i++ {
workers <- &Worker{
grpcClientFactory: grpcClientFactory,
}
}
workerPool := &WorkerPool{
workers: workers,
JobStats: map[*Job]*JobStat{},
}
// FIXME: Not tied to any lifecycle of the owning element (`Service`), this is not the
// end of the world because `WorkerPool` is expected to live forever. But it would still
// be great to have it refactored (the `Service`) to be tied to the running application
// and have the `Service` close the `WorkerPool` which in turn would close the periodic
// stats logger.
workerPool.StartPeriodicLogger()
return workerPool
}
func (p *WorkerPool) StartPeriodicLogger() {
go func() {
ticker := time.NewTicker(60 * time.Second)
for {
select {
case <-ticker.C:
jobStats := make([]*JobStat, 0, len(p.JobStats))
countPerModule := map[string]uint64{}
for _, value := range p.JobStats {
jobStats = append(jobStats, value)
countPerModule[value.ModuleName] = countPerModule[value.ModuleName] + 1
}
zlog.Debug("worker pool statistics",
zap.Int("job_count", len(p.JobStats)),
zap.Reflect("job_stats", jobStats),
zap.Reflect("count_by_module", countPerModule),
)
}
}
}()
}
func (p *WorkerPool) Borrow() *Worker {
w := <-p.workers
return w
}
func (p *WorkerPool) ReturnWorker(worker *Worker) {
p.workers <- worker
}
type Worker struct {
grpcClientFactory substreams.GrpcClientFactory
}
type RetryableErr struct {
cause error
}
func (r *RetryableErr) Error() string {
return r.cause.Error()
}
func (w *Worker) Run(ctx context.Context, job *Job, jobStats map[*Job]*JobStat, requestModules *pbsubstreams.Modules, respFunc substreams.ResponseFunc) ([]*block.Range, error) {
start := time.Now()
jobLogger := zlog.With(zap.Object("job", job))
grpcClient, connClose, grpcCallOpts, err := w.grpcClientFactory(ctx)
if err != nil {
jobLogger.Error("getting grpc client", zap.Error(err))
return nil, &RetryableErr{cause: fmt.Errorf("grpc client factory: %w", err)}
}
defer connClose()
ctx = metadata.NewOutgoingContext(ctx, metadata.New(map[string]string{"substreams-partial-mode": "true"}))
request := job.createRequest(requestModules)
stream, err := grpcClient.Blocks(ctx, request, grpcCallOpts...)
if err != nil {
return nil, &RetryableErr{cause: fmt.Errorf("getting block stream: %w", err)}
}
defer func() {
stream.CloseSend()
}()
meta, err := stream.Header()
if err != nil {
jobLogger.Warn("error getting stream header", zap.Error(err))
}
remoteHostname := "unknown"
if hosts := meta.Get("host"); len(hosts) != 0 {
remoteHostname = hosts[0]
jobLogger = jobLogger.With(zap.String("remote_hostname", remoteHostname))
}
jobStat := &JobStat{
ModuleName: job.ModuleName,
RequestRange: job.requestRange,
StartAt: time.Now(),
CurrentBlock: job.requestRange.StartBlock,
RemoteHost: remoteHostname,
}
jobStats[job] = jobStat
jobLogger.Info("running job", zap.Object("job", job))
defer func() {
jobLogger.Info("job completed", zap.Object("job", job), zap.Duration("in", time.Since(start)), zap.Object("job_stat", jobStat))
delete(jobStats, job)
}()
for {
select {
case <-ctx.Done():
jobLogger.Warn("context cancel will waiting for stream data, worker is terminating")
return nil, ctx.Err()
default:
}
resp, err := stream.Recv()
if resp != nil {
switch r := resp.Message.(type) {
case *pbsubstreams.Response_Progress:
err := respFunc(resp)
if err != nil {
jobLogger.Warn("worker done on respFunc error", zap.Error(err))
return nil, &RetryableErr{cause: fmt.Errorf("sending progress: %w", err)}
}
for _, progress := range resp.GetProgress().Modules {
if f := progress.GetFailed(); f != nil {
zlog.Debug("failed execution of substreams", zap.String("reason", f.Reason))
return nil, fmt.Errorf("module %s failed on host: %s", progress.Name, f.Reason)
}
}
if len(resp.GetProgress().Modules) > 0 {
module := resp.GetProgress().Modules[0]
if rangeCount := len(module.GetProcessedRanges().ProcessedRanges); rangeCount > 0 {
endBlock := module.GetProcessedRanges().ProcessedRanges[rangeCount-1].EndBlock
jobStat.update(endBlock)
}
}
case *pbsubstreams.Response_SnapshotData:
_ = r.SnapshotData
case *pbsubstreams.Response_SnapshotComplete:
_ = r.SnapshotComplete
case *pbsubstreams.Response_Data:
// These are not returned by virtue of `returnOutputs`
}
}
if err != nil {
if err == io.EOF {
jobLogger.Info("worker done", zap.Object("job", job))
trailers := stream.Trailer().Get("substreams-partials-written")
var partialsWritten []*block.Range
if len(trailers) != 0 {
jobLogger.Info("partial written", zap.String("trailer", trailers[0]))
partialsWritten = block.ParseRanges(trailers[0])
}
return partialsWritten, nil
}
jobLogger.Warn("worker done on stream error", zap.Error(err))
return nil, &RetryableErr{cause: fmt.Errorf("receiving stream resp: %w", err)}
}
}
}