-
Notifications
You must be signed in to change notification settings - Fork 0
/
http_server.py
358 lines (320 loc) · 14.2 KB
/
http_server.py
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
#!/usr/bin/env python
# -*- encoding:utf-8 -*-
import sys
from multiprocessing import Process, Event, Queue
import time
import threading
import logging
import uuid
from predict_util import *
from helper import set_logger
import zmq
import zmq.decorators as zmqd
from zmq.utils import jsonapi
import numpy as np
import os
import tensorflow as tf
import modeling
import tokenization
class Test(Process):
def __init__(self):
super(Process, self).__init__()
def run(self):
print("work")
time.sleep(3)
class DataItem(object):
def __init__(self, guid, text_a, text_b=None, label=None):
self.guid = guid
self.text_a = text_a
self.text_b = text_b
self.label = label
class HTTPServer(object):
def __init__(self, config, ready_to_classify_que, classify_res_que, num_worker=1, send_port=5664, recv_port=5665, logger=logging.getLogger("sver")):
#super(Process,self).__init__()
#Process.__init__(self)
self.config = config
self.is_ready = Event()
self.classify_res = dict()
self.ready_to_classify_que = ready_to_classify_que
self.classify_res_que = classify_res_que
self.num_worker = num_worker
self.logger=logger
self.send_port=send_port
self.recv_port=recv_port
send_context = zmq.Context()
self.sender = send_context.socket(zmq.PUSH)
self.sender.bind("tcp://127.0.0.1:%d"%(send_port))
recv_context = zmq.Context()
self.receiver = recv_context.socket(zmq.PULL)
self.receiver.bind("tcp://127.0.0.1:%d"%(recv_port))
logger.info("finish init HTTPServer")
def create_flask_app(self):
try:
from flask import Flask, request
from flask_compress import Compress
# from flask_cors import CORS
from flask_json import FlaskJSON, as_json, JsonError
except ImportError:
raise ImportError()
app = Flask(__name__)
#app.config['SWAGGER'] = {
# 'title':'Colors API',
# 'uiversion':3,
# "openapi":"3.0.2"
#}
self.create_classify_worker()
@app.route('/tts-classify', methods=['POST'])
@as_json
def tts_classify():
req_data = request.form if request.form else request.json
# req_data['req_id'] = uuid.uuid1()
# req_id = req_data['req_id']
texts = req_data['texts'] # text list
if not isinstance(texts, list):
texts = [texts]
req_num = len(texts)
text_ids = []
st = int(time.time() * 1000)
for k in range(req_num):
textid = uuid.uuid1().hex
text_a = texts[k]
text_b = None
# input_item = DataItem(textid,text_a,None,None)
text_ids.append(textid)
input_item = {
"guid":textid,
"text_a":text_a,
"text_b":text_b,
}
self.sender.send_json(jsonapi.dumps(input_item))
# self.ready_to_classify_que.put(input_item)
self.logger.debug("put item:%s"%(textid))
req_time = time.time()
req_data['req_time'] = req_time
# self.ready_to_classify_que.put(req_data)
collect_num = 0
pred_labels = [0]*req_num
for k in range(req_num):
while text_ids[k] not in self.classify_res:
continue
pred_labels[k] = self.classify_res[text_ids[k]]
self.classify_res.pop(text_ids[k])
res_data = {
"pred_labels":pred_labels
}
ed = int(time.time()*1000)
cost = ed - st
self.logger.debug("[timecost] %d ms"%(cost))
return res_data
# CORS(app, origins=self.args.cors)
FlaskJSON(app)
Compress().init_app(app)
return app
def collect_worker_res(self):
while True:
#if self.classify_res_que.empty():
# continue
#else:
# try:
# data_item = self.classify_res_que.get_nowait()
# except:
# continue
# self.logger.debug('put %s res back'%(data_item['req_id']))
# self.classify_res[data_item['req_id']] = data_item
events = self.receiver.poll()
if events:
data_item = self.receiver.recv_json()
data_item = jsonapi.loads(data_item)
guid = data_item['guid']
self.logger.debug('put %s res back'%(guid))
self.classify_res[guid] = data_item['label']
def start(self):
# 启动分类结果接收线程
self.logger.info("start run")
receive_thread = threading.Thread(target=self.collect_worker_res)
receive_thread.start()
self.logger.info("start create app")
app = self.create_flask_app()
self.is_ready.set()
app.run(port=self.config["http_port"], threaded=True, host='0.0.0.0')
self.logger.info("list to port:%d"%(self.config["http_port"]))
receive_thread.join()
def create_classify_worker(self):
for i in range(self.num_worker):
tts_server = TtsClassifyWorker(self.ready_to_classify_que, self.classify_res_que, self.config, send_port=self.recv_port, recv_port=self.send_port, workerid=i, logger=self.logger)
tts_server.start()
class TtsClassifyWorker(Process):
def __init__(self, ready_to_classify_que, classify_res_que, config, send_port, recv_port, workerid=0, logger=logging.getLogger("tts")):
logger.info("start init TtsClassifyWorker:%d init"%(workerid))
Process.__init__(self)
os.environ["CUDA_VISIBLE_DEVICES"] = "%d"%(workerid)
self.id = workerid
self.logger = logger
self.send_port=send_port
self.recv_port=recv_port
model_config_file = "/search/odin/liruihong/tts/multi_attn_model/config_data/classify_config.json"
model_config_file = config["model_config_file"]
run_config = {
"max_seq_length":128,
"batch_size":1,
"word2vec_file":"/search/odin/liruihong/tts/multi_attn_model/config_data/100000-small.txt",
"stop_words_file":"/search/odin/liruihong/tts/multi_attn_model/config_data/cn_stopwords.txt"
}
run_config = {
"max_seq_length":config["max_seq_length"],
"batch_size":config["batch_size"],
"word2vec_file":config["word2vec_file"],
"stop_words_file":config["stop_words_file"]
}
self.processor = TtsProcessor()
self.model_config = modeling.ModelConfig.from_json_file(model_config_file)
self.tokenizer = tokenization.Tokenizer(
word2vec_file=run_config["word2vec_file"], stop_words_file=run_config["stop_words_file"])
self.label_list = self.processor.get_labels()
self.run_config = run_config
self.params = {
"max_seq_length":run_config["max_seq_length"],
"batch_size":run_config["batch_size"]
}
self.embedding_table = load_embedding_table(run_config["word2vec_file"])
# self.model_server = ModelServer(model_config_file, run_config, processor, self.logger)
init_checkpoint = "/search/odin/liruihong/tts/bert_output/wordvec_attn/annotate_part_unlimitlen/model.ckpt-4600"
model_output_dir = "/search/odin/liruihong/tts/bert_output/wordvec_attn/annotate_part_unlimitlen"
self.init_checkpoint = config["init_checkpoint"]
self.model_output_dir = config["model_output_dir"]
self.logger.info("Tts worker[%d] start build model"%(self.id))
#self.model_server.build_model(init_checkpoint, model_output_dir)
#self.get_estimator()
self.ready_to_classify_que = ready_to_classify_que
self.classify_res_que = classify_res_que
self.logger.info("finish TtsClassifyWorker:%d init"%(self.id))
def get_estimator(self):
sess_config = tf.ConfigProto()
sess_config.gpu_options.allow_growth = True
sess_config.gpu_options.per_process_gpu_memory_fraction = 0.5
sess_config.log_device_placement = False
run_config = tf.estimator.RunConfig(session_config=sess_config)
model_fn = model_fn_builder(
model_config=self.model_config,
num_labels=len(self.label_list),
init_checkpoint=self.init_checkpoint,
embedding_table_value=self.embedding_table)
self.estimator = tf.estimator.Estimator(
model_fn=model_fn,
config=run_config,
model_dir=self.model_output_dir,
params=self.params)
self.logger.info("finish model building")
return self.estimator
def collect_pred_res(self, pred_generator):
for pred_res in pred_generator:
pred_label = np.argmax(pred_res["probabilities"]) + 1
guid = pred_res["guid"]
self.logger.debug("collect %s, pred:%d"%(guid, pred_label))
data_item = DataItem(guid=guid,text_a=None,text_b=None,label=pred_label)
self.classify_res_que.put(data_item)
def run(self):
self.logger.info("TtsClassifyWorker %d start run"%(self.id))
self._run()
@zmqd.context()
@zmqd.socket(zmq.PULL)
@zmqd.socket(zmq.PUSH)
def _run(self, _, receiver, sender):
estimator = self.get_estimator()
self.logger.info('bind all sockets')
receiver.connect('tcp://127.0.0.1:%d'%(self.recv_port))
sender.connect('tcp://127.0.0.1:%d'%(self.send_port))
# pred_res_generator = self.model_server.predict()
res_generator = estimator.predict(input_fn=self.input_fn_builder(receiver, self.label_list, self.run_config['max_seq_length'], self.tokenizer), yield_single_examples=True)
for res_item in res_generator:
guid = res_item['guid']
self.logger.debug("worker[%d] get pred res %s"%(self.id, guid))
pred_label = np.argmax(res_item['probabilities']) + 1
data_item = {
"guid":guid,
"label":pred_label
}
sender.send_json(jsonapi.dumps(data_item))
# collect_res_thread = threading.Thread(target=self.collect_pred_res, args=(pred_res_generator))
# collect_res_thread.start()
# collect_res_thread.join()
#while True:
# data_item = self.ready_to_classify_que.get()
# self.logger.debug("get data %s"%(data_item["guid"]))
# # tts_labels = self.model_server.predict(texts)
# # data["tts_labels"] = tts_labels
# self.logger.debug("get predict res req_id:%s, res:%s, cost:%d ms"%(data["req_id"], str(tts_labels), cost))
# # self.classify_res_que.put(data)
def input_fn_builder(self, receiver_sock, label_list, max_seq_length, tokenizer):
def generate_fn():
#poller = zmq.Poller()
#poller.register(receiver_sock, zmq.POLLIN)
while True:
events = receiver_sock.poll()
if events:
data_item = receiver_sock.recv_json()
data_item = jsonapi.loads(data_item)
data_example = InputExample(guid=data_item['guid'],text_a=data_item['text_a'],text_b=data_item['text_b'],label="1")
feature = convert_single_example(data_example,label_list,max_seq_length,tokenizer)
self.logger.debug("input_fn yield %s"%(feature.guid))
yield {
"guid":[feature.guid],
"input_ids":[feature.input_ids],
"input_mask":[feature.input_mask],
"segment_ids":[feature.segment_ids],
"label_ids":[[feature.label_id]]
}
def input_fn(params):
max_seq_length = params["max_seq_length"]
feature_data = tf.data.Dataset.from_generator(
generate_fn,
output_types={
"guid":tf.string,
"input_ids":tf.int32,
"input_mask":tf.int32,
"segment_ids":tf.int32,
"label_ids":tf.int32
},
output_shapes={
"guid":(None),
"input_ids":(None,max_seq_length),
"input_mask":(None,max_seq_length),
"segment_ids":(None,max_seq_length),
"label_ids":(None,1)
}
)
#feature_data = feature_data.batch(params['batch_size'])
#iter = feature_data.make_one_shot_iterator()
#batch_data = iter.get_next()
#feature_dict = {
# 'guid':batch_data['guid'],
# 'input_ids':batch_data['input_ids'],
# 'input_mask':batch_data['input_mask'],
# 'segment_ids':batch_data['segment_ids'],
# 'label_ids':batch_data['label_ids']
#}
#return feature_dict,None
return feature_data
return input_fn
def main():
config = {
"model_config_file":"/search/odin/liruihong/tts/multi_attn_model/config_data/classify_config.json",
"max_seq_length":128,
"batch_size":32,
"word2vec_file":"/search/odin/liruihong/tts/multi_attn_model/config_data/100000-small.txt",
"stop_words_file":"/search/odin/liruihong/tts/multi_attn_model/config_data/cn_stopwords.txt",
"init_checkpoint":"/search/odin/liruihong/tts/bert_output/wordvec_attn/annotate_part_unlimitlen/model.ckpt-4600",
"model_output_dir":"/search/odin/liruihong/tts/bert_output/wordvec_attn/annotate_part_unlimitlen",
"http_port":9001,
}
logger = set_logger("root", verbose=True, handler=logging.StreamHandler())
ready_to_classify_que = Queue()
classify_res_que = Queue()
http_server = HTTPServer(config, ready_to_classify_que, classify_res_que, 1, 5664,5665, logger)
logger.info("start server")
http_server.start()
if __name__ == "__main__":
main()
#test_server = Test()
#test_server.start()
#test_server.join()