-
Notifications
You must be signed in to change notification settings - Fork 58
/
ModelManager.java
449 lines (418 loc) · 17.8 KB
/
ModelManager.java
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
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
/*
* Copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved.
*
* Licensed under the Apache License, Version 2.0 (the "License"). You may not use this file except in compliance
* with the License. A copy of the License is located at
*
* http://aws.amazon.com/apache2.0/
*
* or in the "license" file accompanying this file. This file 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 ai.djl.serving.models;
import ai.djl.ModelException;
import ai.djl.metric.Dimension;
import ai.djl.metric.Metric;
import ai.djl.metric.Unit;
import ai.djl.modality.Input;
import ai.djl.modality.Output;
import ai.djl.repository.zoo.ModelNotFoundException;
import ai.djl.serving.http.BadRequestException;
import ai.djl.serving.http.DescribeWorkflowResponse;
import ai.djl.serving.http.StatusResponse;
import ai.djl.serving.plugins.DependencyManager;
import ai.djl.serving.util.MutableClassLoader;
import ai.djl.serving.wlm.ModelInfo;
import ai.djl.serving.wlm.WorkLoadManager;
import ai.djl.serving.wlm.WorkerPool;
import ai.djl.serving.wlm.WorkerPoolConfig;
import ai.djl.serving.workflow.Workflow;
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;
import java.io.IOException;
import java.util.ArrayList;
import java.util.Arrays;
import java.util.Collections;
import java.util.HashSet;
import java.util.LinkedHashMap;
import java.util.List;
import java.util.Map;
import java.util.Optional;
import java.util.Set;
import java.util.concurrent.CompletableFuture;
import java.util.concurrent.CompletionException;
import java.util.concurrent.ConcurrentHashMap;
import java.util.concurrent.ExecutionException;
import java.util.concurrent.ExecutorService;
import java.util.concurrent.Executors;
import java.util.concurrent.Future;
import java.util.stream.Collectors;
/** A class that in charge of managing models. */
public final class ModelManager {
private static final Logger logger = LoggerFactory.getLogger(ModelManager.class);
private static final Logger MODEL_METRIC = LoggerFactory.getLogger("model_metric");
private static ModelManager modelManager = new ModelManager();
private WorkLoadManager wlm;
private Map<String, Endpoint> endpoints;
private Set<String> startupWorkflows;
private ModelManager() {
wlm = new WorkLoadManager();
endpoints = new ConcurrentHashMap<>();
startupWorkflows = new HashSet<>();
}
/**
* Returns the singleton {@code ModelManager} instance.
*
* @return the singleton {@code ModelManager} instance
*/
public static ModelManager getInstance() {
return modelManager;
}
/**
* Registers and loads a {@link Workflow}.
*
* @param workflow the workflow to register
* @return a {@code CompletableFuture} instance
*/
public CompletableFuture<Void> registerWorkflow(Workflow workflow) {
Endpoint endpoint = endpoints.computeIfAbsent(workflow.getName(), k -> new Endpoint());
if (!endpoint.add(workflow)) {
// workflow already exists
throw new BadRequestException(409, "Workflow " + workflow + " is already registered.");
}
return CompletableFuture.supplyAsync(
() -> {
long begin = System.nanoTime();
Map<String, WorkerPoolConfig<Input, Output>> wpcs = workflow.getWpcMap();
for (Map.Entry<String, WorkerPoolConfig<Input, Output>> entry :
wpcs.entrySet()) {
String key = entry.getKey();
WorkerPoolConfig<Input, Output> workerPoolConfig = entry.getValue();
try {
// download model and configure per model settings
workerPoolConfig.initialize();
// Install engine if necessary
String engine = null;
if (workerPoolConfig instanceof ModelInfo) {
ModelInfo<Input, Output> model =
(ModelInfo<Input, Output>) workerPoolConfig;
engine = model.getEngineName();
DependencyManager dm = DependencyManager.getInstance();
dm.installEngine(engine);
Thread.currentThread()
.setContextClassLoader(MutableClassLoader.getInstance());
WorkerPool<Input, Output> wp = wlm.getWorkerPool(model);
if (wp != null) {
wpcs.put(key, wp.getWpc());
wp.increaseRef();
logger.info("Model {} is registered by other workflow", model);
continue;
}
}
wlm.registerWorkerPool(workerPoolConfig);
String[] devices = workerPoolConfig.getLoadOnDevices();
if (engine != null) {
logger.info(
"Loading model on {}:{}", engine, Arrays.toString(devices));
} else {
logger.info("Loading worker: {}", Arrays.toString(devices));
}
ExecutorService pool = null;
List<Future<?>> futures = new ArrayList<>();
if (workerPoolConfig.isParallelLoading()) {
pool = Executors.newFixedThreadPool(devices.length);
}
for (String deviceName : devices) {
if (pool != null) {
futures.add(
pool.submit(
() ->
initWorkers(
workerPoolConfig, deviceName)));
} else {
initWorkers(workerPoolConfig, deviceName);
}
}
if (pool != null) {
pool.shutdown();
for (Future<?> future : futures) {
try {
future.get();
} catch (ExecutionException e) {
throw new CompletionException(e.getCause()); // NOPMD
} catch (InterruptedException e) {
throw new AssertionError("Worker startup interrupted.", e);
}
}
}
} catch (IOException | ModelException e) {
throw new CompletionException(e);
}
}
workflow.prepare(wlm);
long duration = (System.nanoTime() - begin) / 1000;
Dimension dimension = new Dimension("Model", workflow.getName());
Metric metric =
new Metric("RegisterWorkflow", duration, Unit.MICROSECONDS, dimension);
MODEL_METRIC.info("{}", metric);
return null;
});
}
/**
* Unregisters a workflow by its name and version.
*
* @param workflowName the workflow name to be unregistered (may also be the same as a model
* name)
* @param version the model version
* @return {@code true} if unregister success
*/
public boolean unregisterWorkflow(String workflowName, String version) {
Endpoint endpoint = endpoints.get(workflowName);
if (endpoint == null) {
logger.warn("Model not found: {}", workflowName);
return false;
}
Set<WorkerPoolConfig<Input, Output>> candidateWpcsToUnregister = new HashSet<>();
if (version == null) {
// unregister all versions
for (Workflow workflow : endpoint.getWorkflows()) {
candidateWpcsToUnregister.addAll(workflow.getWpcs());
workflow.close();
}
startupWorkflows.remove(workflowName);
endpoint.getWorkflows().clear();
logger.info("Model {} unregistered.", workflowName);
} else {
Workflow workflow = endpoint.remove(version);
if (workflow == null) {
logger.warn("Workflow not found: {}:{}", workflowName, version);
return false;
}
candidateWpcsToUnregister.addAll(workflow.getWpcs());
workflow.close();
startupWorkflows.remove(workflowName);
logger.info("Model {}/{} unregistered.", workflowName, version);
}
if (endpoint.getWorkflows().isEmpty()) {
endpoints.remove(workflowName);
}
// Unregister candidate models if they are not used for a remaining endpoint
candidateWpcsToUnregister.removeAll(getWpcs());
for (WorkerPoolConfig<Input, Output> wpc : candidateWpcsToUnregister) {
wlm.unregisterWorkerPool(wpc);
}
return true;
}
/**
* Initializes the workers for a workerPoolConfig.
*
* @param wpc the workerPoolConfig to scale workers for
* @param deviceName the device for the workerPoolConfig
* @see WorkerPool#initWorkers(String)
*/
public void initWorkers(WorkerPoolConfig<Input, Output> wpc, String deviceName) {
Thread.currentThread().setContextClassLoader(MutableClassLoader.getInstance());
wlm.getWorkerPool(wpc).initWorkers(deviceName);
}
/**
* Scales the workers for a model.
*
* @param wpc the model to scale workers for
* @param deviceName the device for the model
* @param minWorkers the min workers, -1 for auto-scale
* @param maxWorkers the max workers, -1 for auto-scale
* @see WorkerPool#scaleWorkers(String, int, int)
*/
public void scaleWorkers(
WorkerPoolConfig<Input, Output> wpc,
String deviceName,
int minWorkers,
int maxWorkers) {
Thread.currentThread().setContextClassLoader(MutableClassLoader.getInstance());
wlm.getWorkerPool(wpc).scaleWorkers(deviceName, minWorkers, maxWorkers);
}
/**
* Returns the registry of all endpoints.
*
* @return the registry of all endpoints
*/
public Map<String, Endpoint> getEndpoints() {
return endpoints;
}
/**
* Returns all {@link WorkerPoolConfig}s in an endpoint.
*
* @return all {@link WorkerPoolConfig}s in an endpoint
*/
public Set<WorkerPoolConfig<Input, Output>> getWpcs() {
return getEndpoints().values().stream()
.flatMap(e -> e.getWorkflows().stream())
.flatMap(w -> w.getWpcs().stream())
.collect(Collectors.toSet());
}
/**
* Returns a version of workflow.
*
* @param workflowName the workflow name
* @param version the model version
* @param predict ture for selecting a model in load balance fashion
* @return the model
*/
public Workflow getWorkflow(String workflowName, String version, boolean predict) {
Endpoint endpoint = endpoints.get(workflowName);
if (endpoint == null) {
return null;
}
if (version == null) {
if (endpoint.getWorkflows().isEmpty()) {
return null;
}
if (predict) {
return endpoint.next();
}
return endpoint.getWorkflows().get(0);
}
return endpoint.get(version);
}
/**
* Returns the {@link WorkLoadManager}.
*
* @return the {@link WorkLoadManager}
*/
public WorkLoadManager getWorkLoadManager() {
return wlm;
}
/**
* Returns a set of models or workflows that were loaded at startup.
*
* @return a set of models or workflows that were loaded at startup
*/
public Set<String> getStartupWorkflows() {
return startupWorkflows;
}
/**
* Returns the single startup workflow.
*
* <p>Returns only if there was exactly 1 startup workflow passed in. Used with integration of
* SageMaker SME and single model services.
*
* @return the workflow name
*/
public Optional<String> getSingleStartupWorkflow() {
Set<String> startModels = getStartupWorkflows();
if (startModels.size() == 1) {
return Optional.ofNullable(startModels.iterator().next());
}
return Optional.empty();
}
/**
* Runs an inference job by assigning the job to the next free worker.
*
* @param workflow the workflow to run
* @param input the input to the task
* @return the {@code CompletableFuture}
*/
public CompletableFuture<Output> runJob(Workflow workflow, Input input) {
return workflow.execute(wlm, input);
}
/**
* Returns a list of worker information for specified workflow.
*
* @param workflowName the workflow name to be queried
* @param version the model version to be queried
* @return model and workers information for specified workflow
* @throws ModelNotFoundException if specified workflow not found
*/
public DescribeWorkflowResponse[] describeWorkflow(String workflowName, String version)
throws ModelNotFoundException {
Endpoint endpoint = endpoints.get(workflowName);
if (endpoint == null) {
throw new ModelNotFoundException("Workflow not found: " + workflowName);
}
List<Workflow> list = null;
if (version == null) {
list = endpoint.getWorkflows();
} else {
Workflow wf = endpoint.get(version);
if (wf != null) {
list = Collections.singletonList(wf);
}
}
if (list == null || list.isEmpty()) {
StringBuilder sb = new StringBuilder("Workflow not found: ");
sb.append(workflowName);
if (version != null) {
sb.append('/').append(version);
}
throw new ModelNotFoundException("Workflow not found: " + sb);
}
DescribeWorkflowResponse[] array = new DescribeWorkflowResponse[list.size()];
int index = 0;
for (Workflow workflow : list) {
array[index++] = new DescribeWorkflowResponse(workflow);
}
return array;
}
/**
* Sends model server health status to client.
*
* @return completableFuture with eventually result in the future after async execution
*/
public CompletableFuture<Map<String, Object>> workerStatus() {
return CompletableFuture.supplyAsync(
() -> {
boolean hasFailure = false;
boolean hasPending = false;
Map<String, StatusResponse> data = new LinkedHashMap<>(); // NOPMD
for (Endpoint endpoint : endpoints.values()) {
for (Workflow wf : endpoint.getWorkflows()) {
String workflowName = wf.getName();
for (WorkerPoolConfig<Input, Output> wpc : wf.getWpcs()) {
String modelName = wpc.getId();
if (!modelName.equals(workflowName)) {
modelName = workflowName + ':' + modelName; // NOPMD
}
WorkerPoolConfig.Status status = wpc.getStatus();
switch (status) {
case FAILED:
data.put(modelName, new StatusResponse(status.name()));
hasFailure = true;
break;
case PENDING:
data.put(modelName, new StatusResponse(status.name()));
hasPending = true;
break;
default:
if (wlm.getWorkerPool(wpc).isFullyScaled()) {
data.put(modelName, new StatusResponse("Healthy"));
} else {
data.put(modelName, new StatusResponse("Unhealthy"));
}
break;
}
}
}
}
Map<String, Object> modelInfos = new LinkedHashMap<>(); // NOPMD
modelInfos.put("hasFailure", hasFailure);
modelInfos.put("hasPending", hasPending);
modelInfos.put("data", data);
return modelInfos;
});
}
/**
* Clears everything in the {@link ModelManager}.
*
* <p>Can be run between tests.
*/
public void clear() {
wlm.close();
for (Endpoint endpoint : endpoints.values()) {
endpoint.close();
}
wlm = new WorkLoadManager();
endpoints = new ConcurrentHashMap<>();
startupWorkflows = new HashSet<>();
}
}