-
Notifications
You must be signed in to change notification settings - Fork 4.2k
/
dataflow.go
328 lines (295 loc) · 12.7 KB
/
dataflow.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
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
// Licensed to the Apache Software Foundation (ASF) under one or more
// contributor license agreements. See the NOTICE file distributed with
// this work for additional information regarding copyright ownership.
// The ASF licenses this file to You under the Apache License, Version 2.0
// (the "License"); you may not use this file except in compliance with
// the License. You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
// Package dataflow contains the Dataflow runner for submitting pipelines
// to Google Cloud Dataflow.
//
// This package infers Pipeline Options from flags automatically on job
// submission, for display in the Dataflow UI.
// Use the DontUseFlagAsPipelineOption function to prevent using a given
// flag as a PipelineOption.
package dataflow
import (
"context"
"encoding/json"
"flag"
"fmt"
"io"
"path"
"strings"
"sync/atomic"
"time"
"cloud.google.com/go/storage"
"github.com/apache/beam/sdks/v2/go/pkg/beam"
"github.com/apache/beam/sdks/v2/go/pkg/beam/core/runtime/graphx"
"github.com/apache/beam/sdks/v2/go/pkg/beam/core/runtime/pipelinex"
"github.com/apache/beam/sdks/v2/go/pkg/beam/core/util/hooks"
"github.com/apache/beam/sdks/v2/go/pkg/beam/internal/errors"
"github.com/apache/beam/sdks/v2/go/pkg/beam/log"
"github.com/apache/beam/sdks/v2/go/pkg/beam/options/gcpopts"
"github.com/apache/beam/sdks/v2/go/pkg/beam/options/jobopts"
"github.com/apache/beam/sdks/v2/go/pkg/beam/runners/dataflow/dataflowlib"
"github.com/apache/beam/sdks/v2/go/pkg/beam/util/gcsx"
"github.com/apache/beam/sdks/v2/go/pkg/beam/x/hooks/perf"
"github.com/golang/protobuf/proto"
)
// TODO(herohde) 5/16/2017: the Dataflow flags should match the other SDKs.
var (
endpoint = flag.String("dataflow_endpoint", "", "Dataflow endpoint (optional).")
stagingLocation = flag.String("staging_location", "", "GCS staging location (required).")
image = flag.String("worker_harness_container_image", "", "Worker harness container image (required).")
labels = flag.String("labels", "", "JSON-formatted map[string]string of job labels (optional).")
serviceAccountEmail = flag.String("service_account_email", "", "Service account email (optional).")
numWorkers = flag.Int64("num_workers", 0, "Number of workers (optional).")
maxNumWorkers = flag.Int64("max_num_workers", 0, "Maximum number of workers during scaling (optional).")
diskSizeGb = flag.Int64("disk_size_gb", 0, "Size of root disk for VMs, in GB (optional).")
diskType = flag.String("disk_type", "", "Type of root disk for VMs (optional).")
autoscalingAlgorithm = flag.String("autoscaling_algorithm", "", "Autoscaling mode to use (optional).")
zone = flag.String("zone", "", "GCP zone (optional)")
network = flag.String("network", "", "GCP network (optional)")
subnetwork = flag.String("subnetwork", "", "GCP subnetwork (optional)")
noUsePublicIPs = flag.Bool("no_use_public_ips", false, "Workers must not use public IP addresses (optional)")
tempLocation = flag.String("temp_location", "", "Temp location (optional)")
machineType = flag.String("worker_machine_type", "", "GCE machine type (optional)")
minCPUPlatform = flag.String("min_cpu_platform", "", "GCE minimum cpu platform (optional)")
workerJar = flag.String("dataflow_worker_jar", "", "Dataflow worker jar (optional)")
workerRegion = flag.String("worker_region", "", "Dataflow worker region (optional)")
workerZone = flag.String("worker_zone", "", "Dataflow worker zone (optional)")
executeAsync = flag.Bool("execute_async", false, "Asynchronous execution. Submit the job and return immediately.")
dryRun = flag.Bool("dry_run", false, "Dry run. Just print the job, but don't submit it.")
teardownPolicy = flag.String("teardown_policy", "", "Job teardown policy (internal only).")
// SDK options
cpuProfiling = flag.String("cpu_profiling", "", "Job records CPU profiles to this GCS location (optional)")
sessionRecording = flag.String("session_recording", "", "Job records session transcripts")
)
// flagFilter filters flags that are already represented by the above flags
// or in the JobOpts to prevent them from appearing duplicated
// as PipelineOption display data.
//
// New flags that are already put into pipeline options
// should be added to this map.
var flagFilter = map[string]bool{
"dataflow_endpoint": true,
"staging_location": true,
"worker_harness_container_image": true,
"labels": true,
"service_account_email": true,
"num_workers": true,
"max_num_workers": true,
"disk_size_gb": true,
"disk_type": true,
"autoscaling_algorithm": true,
"zone": true,
"network": true,
"subnetwork": true,
"no_use_public_ips": true,
"temp_location": true,
"worker_machine_type": true,
"min_cpu_platform": true,
"dataflow_worker_jar": true,
"worker_region": true,
"worker_zone": true,
"teardown_policy": true,
"cpu_profiling": true,
"session_recording": true,
// Job Options flags
"endpoint": true,
"job_name": true,
"environment_type": true,
"environment_config": true,
"experiments": true,
"async": true,
"retain_docker_containers": true,
"parallelism": true,
// GCP opts
"project": true,
"region": true,
// Other common beam flags.
"runner": true,
// Don't filter these to note override.
// "beam_strict": true,
// "sdk_harness_container_image_override": true,
// "worker_binary": true,
}
// DontUseFlagAsPipelineOption prevents a set flag from appearing
// as a PipelineOption in the Dataflow UI. Useful for sensitive,
// noisy, or irrelevant configuration.
func DontUseFlagAsPipelineOption(s string) {
flagFilter[s] = true
}
func init() {
// Note that we also _ import harness/init to setup the remote execution hook.
beam.RegisterRunner("dataflow", Execute)
beam.RegisterRunner("DataflowRunner", Execute)
perf.RegisterProfCaptureHook("gcs_profile_writer", gcsRecorderHook)
}
var unique int32
// Execute runs the given pipeline on Google Cloud Dataflow. It uses the
// default application credentials to submit the job.
func Execute(ctx context.Context, p *beam.Pipeline) (beam.PipelineResult, error) {
// (1) Gather job options
project := gcpopts.GetProjectFromFlagOrEnvironment(ctx)
if project == "" {
return nil, errors.New("no Google Cloud project specified. Use --project=<project>")
}
region := gcpopts.GetRegion(ctx)
if region == "" {
return nil, errors.New("No Google Cloud region specified. Use --region=<region>. See https://cloud.google.com/dataflow/docs/concepts/regional-endpoints")
}
if *stagingLocation == "" {
return nil, errors.New("no GCS staging location specified. Use --staging_location=gs://<bucket>/<path>")
}
var jobLabels map[string]string
if *labels != "" {
if err := json.Unmarshal([]byte(*labels), &jobLabels); err != nil {
return nil, errors.Wrapf(err, "error reading --label flag as JSON")
}
}
if *cpuProfiling != "" {
perf.EnableProfCaptureHook("gcs_profile_writer", *cpuProfiling)
}
if *sessionRecording != "" {
// TODO(wcn): BEAM-4017
// It's a bit inconvenient for GCS because the whole object is written in
// one pass, whereas the session logs are constantly appended. We wouldn't
// want to hold all the logs in memory to flush at the end of the pipeline
// as we'd blow out memory on the worker. The implementation of the
// CaptureHook should create an internal buffer and write chunks out to GCS
// once they get to an appropriate size (50M or so?)
}
if *autoscalingAlgorithm != "" {
if *autoscalingAlgorithm != "NONE" && *autoscalingAlgorithm != "THROUGHPUT_BASED" {
return nil, errors.New("invalid autoscaling algorithm. Use --autoscaling_algorithm=(NONE|THROUGHPUT_BASED)")
}
}
hooks.SerializeHooksToOptions()
experiments := jobopts.GetExperiments()
// Always use runner v2, unless set already.
var v2set, portaSubmission bool
for _, e := range experiments {
if strings.Contains(e, "use_runner_v2") || strings.Contains(e, "use_unified_worker") {
v2set = true
}
if strings.Contains(e, "use_portable_job_submission") {
portaSubmission = true
}
}
// Enable by default unified worker, and portable job submission.
if !v2set {
experiments = append(experiments, "use_unified_worker")
}
if !portaSubmission {
experiments = append(experiments, "use_portable_job_submission")
}
if *minCPUPlatform != "" {
experiments = append(experiments, fmt.Sprintf("min_cpu_platform=%v", *minCPUPlatform))
}
opts := &dataflowlib.JobOptions{
Name: jobopts.GetJobName(),
Experiments: experiments,
Options: beam.PipelineOptions.Export(),
Project: project,
Region: region,
Zone: *zone,
Network: *network,
Subnetwork: *subnetwork,
NoUsePublicIPs: *noUsePublicIPs,
NumWorkers: *numWorkers,
MaxNumWorkers: *maxNumWorkers,
DiskSizeGb: *diskSizeGb,
DiskType: *diskType,
Algorithm: *autoscalingAlgorithm,
MachineType: *machineType,
Labels: jobLabels,
ServiceAccountEmail: *serviceAccountEmail,
TempLocation: *tempLocation,
Worker: *jobopts.WorkerBinary,
WorkerJar: *workerJar,
WorkerRegion: *workerRegion,
WorkerZone: *workerZone,
TeardownPolicy: *teardownPolicy,
ContainerImage: getContainerImage(ctx),
}
if opts.TempLocation == "" {
opts.TempLocation = gcsx.Join(*stagingLocation, "tmp")
}
// (1) Build and submit
// NOTE(herohde) 10/8/2018: the last segment of the names must be "worker" and "dataflow-worker.jar".
id := fmt.Sprintf("go-%v-%v", atomic.AddInt32(&unique, 1), time.Now().UnixNano())
modelURL := gcsx.Join(*stagingLocation, id, "model")
workerURL := gcsx.Join(*stagingLocation, id, "worker")
jarURL := gcsx.Join(*stagingLocation, id, "dataflow-worker.jar")
xlangURL := gcsx.Join(*stagingLocation, id, "xlang")
edges, _, err := p.Build()
if err != nil {
return nil, err
}
artifactURLs, err := dataflowlib.ResolveXLangArtifacts(ctx, edges, opts.Project, xlangURL)
if err != nil {
return nil, errors.WithContext(err, "resolving cross-language artifacts")
}
opts.ArtifactURLs = artifactURLs
environment, err := graphx.CreateEnvironment(ctx, jobopts.GetEnvironmentUrn(ctx), getContainerImage)
if err != nil {
return nil, errors.WithContext(err, "creating environment for model pipeline")
}
model, err := graphx.Marshal(edges, &graphx.Options{Environment: environment})
if err != nil {
return nil, errors.WithContext(err, "generating model pipeline")
}
err = pipelinex.ApplySdkImageOverrides(model, jobopts.GetSdkImageOverrides())
if err != nil {
return nil, errors.WithContext(err, "applying container image overrides")
}
// Apply the all the as Go Options
flag.Visit(func(f *flag.Flag) {
if !flagFilter[f.Name] {
opts.Options.Options[f.Name] = f.Value.String()
}
})
if *dryRun {
log.Info(ctx, "Dry-run: not submitting job!")
log.Info(ctx, proto.MarshalTextString(model))
job, err := dataflowlib.Translate(ctx, model, opts, workerURL, jarURL, modelURL)
if err != nil {
return nil, err
}
dataflowlib.PrintJob(ctx, job)
return nil, nil
}
return dataflowlib.Execute(ctx, model, opts, workerURL, jarURL, modelURL, *endpoint, *executeAsync)
}
func gcsRecorderHook(opts []string) perf.CaptureHook {
bucket, prefix, err := gcsx.ParseObject(opts[0])
if err != nil {
panic(fmt.Sprintf("Invalid hook configuration for gcsRecorderHook: %s", opts))
}
return func(ctx context.Context, spec string, r io.Reader) error {
client, err := gcsx.NewClient(ctx, storage.ScopeReadWrite)
if err != nil {
return errors.WithContext(err, "establishing GCS client")
}
return gcsx.WriteObject(ctx, client, bucket, path.Join(prefix, spec), r)
}
}
func getContainerImage(ctx context.Context) string {
urn := jobopts.GetEnvironmentUrn(ctx)
if urn == "" || urn == "beam:env:docker:v1" {
if *image != "" {
return *image
}
return jobopts.GetEnvironmentConfig(ctx)
}
panic(fmt.Sprintf("Unsupported environment %v", urn))
}