__init__.py 27.7 KB
Newer Older
G
guru4elephant 已提交
1 2 3 4 5 6 7 8 9 10 11 12 13
#   Copyright (c) 2020 PaddlePaddle Authors. 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.
# 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.
14
# pylint: disable=doc-string-missing
G
guru4elephant 已提交
15

M
MRXLT 已提交
16 17
import paddle_serving_client
import os
18 19 20
from .proto import sdk_configure_pb2 as sdk
from .proto import general_model_config_pb2 as m_config
import google.protobuf.text_format
D
dongdaxiang 已提交
21 22
import numpy as np
import time
23
import sys
G
guru4elephant 已提交
24

B
barrierye 已提交
25
import grpc
B
barrierye 已提交
26
from .proto import multi_lang_general_model_service_pb2
B
barrierye 已提交
27 28
sys.path.append(
    os.path.join(os.path.abspath(os.path.dirname(__file__)), 'proto'))
B
barrierye 已提交
29
from .proto import multi_lang_general_model_service_pb2_grpc
B
barrierye 已提交
30

M
MRXLT 已提交
31 32 33 34 35
int64_type = 0
float32_type = 1
int32_type = 2
int_type = set([int64_type, int32_type])
float_type = set([float32_type])
G
guru4elephant 已提交
36

M
MRXLT 已提交
37

W
WangXi 已提交
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
class _NOPProfiler(object):
    def record(self, name):
        pass

    def print_profile(self):
        pass


class _TimeProfiler(object):
    def __init__(self):
        self.pid = os.getpid()
        self.print_head = 'PROFILE\tpid:{}\t'.format(self.pid)
        self.time_record = [self.print_head]

    def record(self, name):
        self.time_record.append('{}:{} '.format(
            name, int(round(time.time() * 1000000))))

    def print_profile(self):
        self.time_record.append('\n')
        sys.stderr.write(''.join(self.time_record))
        self.time_record = [self.print_head]


_is_profile = int(os.environ.get('FLAGS_profile_client', 0))
_Profiler = _TimeProfiler if _is_profile else _NOPProfiler


G
guru4elephant 已提交
66 67 68
class SDKConfig(object):
    def __init__(self):
        self.sdk_desc = sdk.SDKConf()
69 70 71
        self.tag_list = []
        self.cluster_list = []
        self.variant_weight_list = []
M
MRXLT 已提交
72 73
        self.rpc_timeout_ms = 20000
        self.load_balance_strategy = "la"
G
guru4elephant 已提交
74

75 76 77 78
    def add_server_variant(self, tag, cluster, variant_weight):
        self.tag_list.append(tag)
        self.cluster_list.append(cluster)
        self.variant_weight_list.append(variant_weight)
G
guru4elephant 已提交
79

M
MRXLT 已提交
80 81 82 83
    def set_load_banlance_strategy(self, strategy):
        self.load_balance_strategy = strategy

    def gen_desc(self, rpc_timeout_ms):
G
guru4elephant 已提交
84 85 86 87 88
        predictor_desc = sdk.Predictor()
        predictor_desc.name = "general_model"
        predictor_desc.service_name = \
            "baidu.paddle_serving.predictor.general_model.GeneralModelService"
        predictor_desc.endpoint_router = "WeightedRandomRender"
89 90
        predictor_desc.weighted_random_render_conf.variant_weight_list = "|".join(
            self.variant_weight_list)
G
guru4elephant 已提交
91

92 93 94 95 96 97
        for idx, tag in enumerate(self.tag_list):
            variant_desc = sdk.VariantConf()
            variant_desc.tag = tag
            variant_desc.naming_conf.cluster = "list://{}".format(",".join(
                self.cluster_list[idx]))
            predictor_desc.variants.extend([variant_desc])
G
guru4elephant 已提交
98 99 100 101

        self.sdk_desc.predictors.extend([predictor_desc])
        self.sdk_desc.default_variant_conf.tag = "default"
        self.sdk_desc.default_variant_conf.connection_conf.connect_timeout_ms = 2000
M
MRXLT 已提交
102
        self.sdk_desc.default_variant_conf.connection_conf.rpc_timeout_ms = rpc_timeout_ms
G
guru4elephant 已提交
103 104 105 106 107
        self.sdk_desc.default_variant_conf.connection_conf.connect_retry_count = 2
        self.sdk_desc.default_variant_conf.connection_conf.max_connection_per_host = 100
        self.sdk_desc.default_variant_conf.connection_conf.hedge_request_timeout_ms = -1
        self.sdk_desc.default_variant_conf.connection_conf.hedge_fetch_retry_count = 2
        self.sdk_desc.default_variant_conf.connection_conf.connection_type = "pooled"
M
MRXLT 已提交
108

G
guru4elephant 已提交
109 110 111 112 113 114 115 116
        self.sdk_desc.default_variant_conf.naming_conf.cluster_filter_strategy = "Default"
        self.sdk_desc.default_variant_conf.naming_conf.load_balance_strategy = "la"

        self.sdk_desc.default_variant_conf.rpc_parameter.compress_type = 0
        self.sdk_desc.default_variant_conf.rpc_parameter.package_size = 20
        self.sdk_desc.default_variant_conf.rpc_parameter.protocol = "baidu_std"
        self.sdk_desc.default_variant_conf.rpc_parameter.max_channel_per_request = 3

G
guru4elephant 已提交
117
        return self.sdk_desc
G
guru4elephant 已提交
118

G
guru4elephant 已提交
119 120 121 122 123 124

class Client(object):
    def __init__(self):
        self.feed_names_ = []
        self.fetch_names_ = []
        self.client_handle_ = None
M
MRXLT 已提交
125
        self.feed_shapes_ = {}
G
guru4elephant 已提交
126
        self.feed_types_ = {}
G
guru4elephant 已提交
127
        self.feed_names_to_idx_ = {}
M
MRXLT 已提交
128
        self.pid = os.getpid()
B
barrierye 已提交
129
        self.predictor_sdk_ = None
G
guru4elephant 已提交
130 131
        self.producers = []
        self.consumer = None
W
WangXi 已提交
132
        self.profile_ = _Profiler()
M
MRXLT 已提交
133 134
        self.all_numpy_input = True
        self.has_numpy_input = False
M
MRXLT 已提交
135
        self.rpc_timeout_ms = 20000
136 137
        from .serving_client import PredictorRes
        self.predictorres_constructor = PredictorRes
M
MRXLT 已提交
138

G
guru4elephant 已提交
139
    def load_client_config(self, path):
M
MRXLT 已提交
140
        from .serving_client import PredictorClient
141 142 143 144 145
        model_conf = m_config.GeneralModelConfig()
        f = open(path, 'r')
        model_conf = google.protobuf.text_format.Merge(
            str(f.read()), model_conf)

G
guru4elephant 已提交
146 147 148 149
        # load configuraion here
        # get feed vars, fetch vars
        # get feed shapes, feed types
        # map feed names to index
G
guru4elephant 已提交
150 151
        self.client_handle_ = PredictorClient()
        self.client_handle_.init(path)
M
bug fix  
MRXLT 已提交
152 153
        if "FLAGS_max_body_size" not in os.environ:
            os.environ["FLAGS_max_body_size"] = str(512 * 1024 * 1024)
M
MRXLT 已提交
154
        read_env_flags = ["profile_client", "profile_server", "max_body_size"]
M
MRXLT 已提交
155 156
        self.client_handle_.init_gflags([sys.argv[
            0]] + ["--tryfromenv=" + ",".join(read_env_flags)])
157 158
        self.feed_names_ = [var.alias_name for var in model_conf.feed_var]
        self.fetch_names_ = [var.alias_name for var in model_conf.fetch_var]
G
guru4elephant 已提交
159
        self.feed_names_to_idx_ = {}
G
guru4elephant 已提交
160 161
        self.fetch_names_to_type_ = {}
        self.fetch_names_to_idx_ = {}
M
MRXLT 已提交
162
        self.lod_tensor_set = set()
M
MRXLT 已提交
163
        self.feed_tensor_len = {}
164

165 166 167
        for i, var in enumerate(model_conf.feed_var):
            self.feed_names_to_idx_[var.alias_name] = i
            self.feed_types_[var.alias_name] = var.feed_type
M
MRXLT 已提交
168
            self.feed_shapes_[var.alias_name] = var.shape
M
MRXLT 已提交
169

M
MRXLT 已提交
170 171
            if var.is_lod_tensor:
                self.lod_tensor_set.add(var.alias_name)
M
MRXLT 已提交
172 173 174 175 176
            else:
                counter = 1
                for dim in self.feed_shapes_[var.alias_name]:
                    counter *= dim
                self.feed_tensor_len[var.alias_name] = counter
G
guru4elephant 已提交
177 178 179
        for i, var in enumerate(model_conf.fetch_var):
            self.fetch_names_to_idx_[var.alias_name] = i
            self.fetch_names_to_type_[var.alias_name] = var.fetch_type
180 181
            if var.is_lod_tensor:
                self.lod_tensor_set.add(var.alias_name)
G
guru4elephant 已提交
182 183
        return

184
    def add_variant(self, tag, cluster, variant_weight):
B
barrierye 已提交
185 186
        if self.predictor_sdk_ is None:
            self.predictor_sdk_ = SDKConfig()
187 188 189
        self.predictor_sdk_.add_server_variant(tag, cluster,
                                               str(variant_weight))

M
MRXLT 已提交
190 191 192 193 194 195
    def set_rpc_timeout_ms(self, rpc_timeout):
        if not isinstance(rpc_timeout, int):
            raise ValueError("rpc_timeout must be int type.")
        else:
            self.rpc_timeout_ms = rpc_timeout

B
barrierye 已提交
196
    def connect(self, endpoints=None):
G
guru4elephant 已提交
197 198 199
        # check whether current endpoint is available
        # init from client config
        # create predictor here
B
barrierye 已提交
200 201
        if endpoints is None:
            if self.predictor_sdk_ is None:
M
MRXLT 已提交
202
                raise ValueError(
B
barrierye 已提交
203 204 205 206
                    "You must set the endpoints parameter or use add_variant function to create a variant."
                )
        else:
            if self.predictor_sdk_ is None:
207
                self.add_variant('default_tag_{}'.format(id(self)), endpoints,
208
                                 100)
B
barrierye 已提交
209 210
            else:
                print(
211
                    "parameter endpoints({}) will not take effect, because you use the add_variant function.".
B
barrierye 已提交
212
                    format(endpoints))
M
MRXLT 已提交
213
        sdk_desc = self.predictor_sdk_.gen_desc(self.rpc_timeout_ms)
M
MRXLT 已提交
214 215
        self.client_handle_.create_predictor_by_desc(sdk_desc.SerializeToString(
        ))
G
guru4elephant 已提交
216 217 218 219 220 221 222

    def get_feed_names(self):
        return self.feed_names_

    def get_fetch_names(self):
        return self.fetch_names_

M
MRXLT 已提交
223 224 225
    def shape_check(self, feed, key):
        if key in self.lod_tensor_set:
            return
M
MRXLT 已提交
226 227
        if isinstance(feed[key],
                      list) and len(feed[key]) != self.feed_tensor_len[key]:
M
MRXLT 已提交
228
            raise ValueError("The shape of feed tensor {} not match.".format(
M
MRXLT 已提交
229 230 231
                key))
        if type(feed[key]).__module__ == np.__name__ and np.size(feed[
                key]) != self.feed_tensor_len[key]:
M
MRXLT 已提交
232 233 234
            #raise SystemExit("The shape of feed tensor {} not match.".format(
            #    key))
            pass
M
MRXLT 已提交
235

236
    def predict(self, feed=None, fetch=None, need_variant_tag=False, log_id=0):
W
WangXi 已提交
237 238
        self.profile_.record('py_prepro_0')

G
guru4elephant 已提交
239 240 241
        if feed is None or fetch is None:
            raise ValueError("You should specify feed and fetch for prediction")

242 243 244 245 246 247
        fetch_list = []
        if isinstance(fetch, str):
            fetch_list = [fetch]
        elif isinstance(fetch, list):
            fetch_list = fetch
        else:
M
MRXLT 已提交
248
            raise ValueError("Fetch only accepts string and list of string")
249 250 251 252 253 254 255

        feed_batch = []
        if isinstance(feed, dict):
            feed_batch.append(feed)
        elif isinstance(feed, list):
            feed_batch = feed
        else:
M
MRXLT 已提交
256
            raise ValueError("Feed only accepts dict and list of dict")
G
guru4elephant 已提交
257

M
MRXLT 已提交
258 259 260 261
        int_slot_batch = []
        float_slot_batch = []
        int_feed_names = []
        float_feed_names = []
D
dongdaxiang 已提交
262 263
        int_shape = []
        float_shape = []
M
MRXLT 已提交
264
        fetch_names = []
M
MRXLT 已提交
265
        counter = 0
M
MRXLT 已提交
266
        batch_size = len(feed_batch)
267 268 269 270 271 272 273

        for key in fetch_list:
            if key in self.fetch_names_:
                fetch_names.append(key)

        if len(fetch_names) == 0:
            raise ValueError(
M
MRXLT 已提交
274
                "Fetch names should not be empty or out of saved fetch list.")
275 276
            return {}

G
guru4elephant 已提交
277
        for i, feed_i in enumerate(feed_batch):
M
MRXLT 已提交
278 279
            int_slot = []
            float_slot = []
280
            for key in feed_i:
M
MRXLT 已提交
281
                if key not in self.feed_names_:
M
MRXLT 已提交
282
                    raise ValueError("Wrong feed name: {}.".format(key))
M
MRXLT 已提交
283 284
                #if not isinstance(feed_i[key], np.ndarray):
                self.shape_check(feed_i, key)
M
MRXLT 已提交
285
                if self.feed_types_[key] in int_type:
G
guru4elephant 已提交
286
                    if i == 0:
M
MRXLT 已提交
287
                        int_feed_names.append(key)
D
dongdaxiang 已提交
288
                        if isinstance(feed_i[key], np.ndarray):
289
                            int_shape.append(list(feed_i[key].shape))
D
dongdaxiang 已提交
290 291
                        else:
                            int_shape.append(self.feed_shapes_[key])
D
dongdaxiang 已提交
292
                    if isinstance(feed_i[key], np.ndarray):
M
MRXLT 已提交
293
                        int_slot.append(feed_i[key])
M
MRXLT 已提交
294
                        self.has_numpy_input = True
D
dongdaxiang 已提交
295 296
                    else:
                        int_slot.append(feed_i[key])
M
MRXLT 已提交
297
                        self.all_numpy_input = False
M
MRXLT 已提交
298
                elif self.feed_types_[key] in float_type:
G
guru4elephant 已提交
299
                    if i == 0:
M
MRXLT 已提交
300
                        float_feed_names.append(key)
D
dongdaxiang 已提交
301
                        if isinstance(feed_i[key], np.ndarray):
302
                            float_shape.append(list(feed_i[key].shape))
D
dongdaxiang 已提交
303 304
                        else:
                            float_shape.append(self.feed_shapes_[key])
D
dongdaxiang 已提交
305
                    if isinstance(feed_i[key], np.ndarray):
M
MRXLT 已提交
306
                        float_slot.append(feed_i[key])
M
MRXLT 已提交
307
                        self.has_numpy_input = True
D
dongdaxiang 已提交
308 309
                    else:
                        float_slot.append(feed_i[key])
M
MRXLT 已提交
310
                        self.all_numpy_input = False
M
MRXLT 已提交
311 312 313
            int_slot_batch.append(int_slot)
            float_slot_batch.append(float_slot)

W
WangXi 已提交
314 315 316
        self.profile_.record('py_prepro_1')
        self.profile_.record('py_client_infer_0')

317
        result_batch_handle = self.predictorres_constructor()
M
MRXLT 已提交
318
        if self.all_numpy_input:
M
MRXLT 已提交
319 320
            res = self.client_handle_.numpy_predict(
                float_slot_batch, float_feed_names, float_shape, int_slot_batch,
321
                int_feed_names, int_shape, fetch_names, result_batch_handle,
322
                self.pid, log_id)
M
MRXLT 已提交
323
        elif self.has_numpy_input == False:
M
MRXLT 已提交
324 325
            res = self.client_handle_.batch_predict(
                float_slot_batch, float_feed_names, float_shape, int_slot_batch,
326
                int_feed_names, int_shape, fetch_names, result_batch_handle,
327
                self.pid, log_id)
M
MRXLT 已提交
328
        else:
M
MRXLT 已提交
329
            raise ValueError(
M
MRXLT 已提交
330 331
                "Please make sure the inputs are all in list type or all in numpy.array type"
            )
M
MRXLT 已提交
332

W
WangXi 已提交
333 334 335
        self.profile_.record('py_client_infer_1')
        self.profile_.record('py_postpro_0')

336 337 338
        if res == -1:
            return None

B
barrierye 已提交
339
        multi_result_map = []
340
        model_engine_names = result_batch_handle.get_engine_names()
B
barrierye 已提交
341
        for mi, engine_name in enumerate(model_engine_names):
B
barrierye 已提交
342
            result_map = {}
B
barrierye 已提交
343
            # result map needs to be a numpy array
B
barrierye 已提交
344
            for i, name in enumerate(fetch_names):
M
MRXLT 已提交
345
                if self.fetch_names_to_type_[name] == int64_type:
B
barrierye 已提交
346
                    # result_map[name] will be py::array(numpy array)
347 348 349
                    result_map[name] = result_batch_handle.get_int64_by_name(
                        mi, name)
                    shape = result_batch_handle.get_shape(mi, name)
B
barriery 已提交
350 351 352 353 354
                    if result_map[name].size == 0:
                        raise ValueError(
                            "Failed to fetch, maybe the type of [{}]"
                            " is wrong, please check the model file".format(
                                name))
B
barrierye 已提交
355 356
                    result_map[name].shape = shape
                    if name in self.lod_tensor_set:
W
wangjiawei04 已提交
357 358 359
                        tmp_lod = result_batch_handle.get_lod(mi, name)
                        if np.size(tmp_lod) > 0:
                            result_map["{}.lod".format(name)] = tmp_lod
M
MRXLT 已提交
360
                elif self.fetch_names_to_type_[name] == float32_type:
361 362
                    result_map[name] = result_batch_handle.get_float_by_name(
                        mi, name)
B
barriery 已提交
363 364 365 366 367
                    if result_map[name].size == 0:
                        raise ValueError(
                            "Failed to fetch, maybe the type of [{}]"
                            " is wrong, please check the model file".format(
                                name))
368
                    shape = result_batch_handle.get_shape(mi, name)
B
barrierye 已提交
369 370
                    result_map[name].shape = shape
                    if name in self.lod_tensor_set:
W
wangjiawei04 已提交
371 372 373
                        tmp_lod = result_batch_handle.get_lod(mi, name)
                        if np.size(tmp_lod) > 0:
                            result_map["{}.lod".format(name)] = tmp_lod
M
MRXLT 已提交
374 375 376 377
                elif self.fetch_names_to_type_[name] == int32_type:
                    # result_map[name] will be py::array(numpy array)
                    result_map[name] = result_batch_handle.get_int32_by_name(
                        mi, name)
B
barriery 已提交
378 379 380 381 382
                    if result_map[name].size == 0:
                        raise ValueError(
                            "Failed to fetch, maybe the type of [{}]"
                            " is wrong, please check the model file".format(
                                name))
M
MRXLT 已提交
383 384 385
                    shape = result_batch_handle.get_shape(mi, name)
                    result_map[name].shape = shape
                    if name in self.lod_tensor_set:
W
wangjiawei04 已提交
386 387 388
                        tmp_lod = result_batch_handle.get_lod(mi, name)
                        if np.size(tmp_lod) > 0:
                            result_map["{}.lod".format(name)] = tmp_lod
B
barrierye 已提交
389
            multi_result_map.append(result_map)
B
barrierye 已提交
390 391
        ret = None
        if len(model_engine_names) == 1:
B
barrierye 已提交
392 393
            # If only one model result is returned, the format of ret is result_map
            ret = multi_result_map[0]
G
guru4elephant 已提交
394
        else:
B
barrierye 已提交
395 396 397 398 399 400
            # If multiple model results are returned, the format of ret is {name: result_map}
            ret = {
                engine_name: multi_result_map[mi]
                for mi, engine_name in enumerate(model_engine_names)
            }

W
WangXi 已提交
401 402 403
        self.profile_.record('py_postpro_1')
        self.profile_.print_profile()

B
barrierye 已提交
404
        # When using the A/B test, the tag of variant needs to be returned
B
barrierye 已提交
405
        return ret if not need_variant_tag else [
406
            ret, result_batch_handle.variant_tag()
B
barrierye 已提交
407
        ]
B
barrierye 已提交
408

409 410
    def release(self):
        self.client_handle_.destroy_predictor()
G
guru4elephant 已提交
411
        self.client_handle_ = None
B
barrierye 已提交
412 413


414
class MultiLangClient(object):
B
barrierye 已提交
415 416
    def __init__(self):
        self.channel_ = None
417
        self.stub_ = None
B
barrierye 已提交
418
        self.rpc_timeout_s_ = 2
B
barrierye 已提交
419
        self.profile_ = _Profiler()
B
barrierye 已提交
420

B
barrierye 已提交
421 422
    def add_variant(self, tag, cluster, variant_weight):
        # TODO
B
barrierye 已提交
423
        raise Exception("cannot support ABtest yet")
B
barrierye 已提交
424 425

    def set_rpc_timeout_ms(self, rpc_timeout):
426 427 428 429 430
        if self.stub_ is None:
            raise Exception("set timeout must be set after connect.")
        if not isinstance(rpc_timeout, int):
            # for bclient
            raise ValueError("rpc_timeout must be int type.")
B
barrierye 已提交
431
        self.rpc_timeout_s_ = rpc_timeout / 1000.0
432 433 434 435
        timeout_req = multi_lang_general_model_service_pb2.SetTimeoutRequest()
        timeout_req.timeout_ms = rpc_timeout
        resp = self.stub_.SetTimeout(timeout_req)
        return resp.err_code == 0
B
barrierye 已提交
436 437

    def connect(self, endpoints):
W
WangXi 已提交
438 439
        # https://github.com/tensorflow/serving/issues/1382
        options = [('grpc.max_receive_message_length', 512 * 1024 * 1024),
440 441
                   ('grpc.max_send_message_length', 512 * 1024 * 1024),
                   ('grpc.lb_policy_name', 'round_robin')]
B
barrierye 已提交
442
        # TODO: weight round robin
443
        g_endpoint = 'ipv4:{}'.format(','.join(endpoints))
B
barrierye 已提交
444
        self.channel_ = grpc.insecure_channel(g_endpoint, options=options)
445
        self.stub_ = multi_lang_general_model_service_pb2_grpc.MultiLangGeneralModelServiceStub(
B
barrierye 已提交
446
            self.channel_)
447 448 449 450 451 452
        # get client model config
        get_client_config_req = multi_lang_general_model_service_pb2.GetClientConfigRequest(
        )
        resp = self.stub_.GetClientConfig(get_client_config_req)
        model_config_str = resp.client_config_str
        self._parse_model_config(model_config_str)
B
barrierye 已提交
453

B
barrierye 已提交
454 455 456 457 458 459 460 461
    def _flatten_list(self, nested_list):
        for item in nested_list:
            if isinstance(item, (list, tuple)):
                for sub_item in self._flatten_list(item):
                    yield sub_item
            else:
                yield item

462
    def _parse_model_config(self, model_config_str):
B
barrierye 已提交
463
        model_conf = m_config.GeneralModelConfig()
464 465
        model_conf = google.protobuf.text_format.Merge(model_config_str,
                                                       model_conf)
B
barrierye 已提交
466 467
        self.feed_names_ = [var.alias_name for var in model_conf.feed_var]
        self.feed_types_ = {}
B
barrierye 已提交
468
        self.feed_shapes_ = {}
B
barrierye 已提交
469
        self.fetch_names_ = [var.alias_name for var in model_conf.fetch_var]
B
barrierye 已提交
470 471
        self.fetch_types_ = {}
        self.lod_tensor_set_ = set()
B
barrierye 已提交
472 473 474
        for i, var in enumerate(model_conf.feed_var):
            self.feed_types_[var.alias_name] = var.feed_type
            self.feed_shapes_[var.alias_name] = var.shape
B
barrierye 已提交
475
            if var.is_lod_tensor:
B
barrierye 已提交
476
                self.lod_tensor_set_.add(var.alias_name)
B
barrierye 已提交
477 478 479 480
            else:
                counter = 1
                for dim in self.feed_shapes_[var.alias_name]:
                    counter *= dim
B
barrierye 已提交
481
        for i, var in enumerate(model_conf.fetch_var):
B
barrierye 已提交
482 483 484
            self.fetch_types_[var.alias_name] = var.fetch_type
            if var.is_lod_tensor:
                self.lod_tensor_set_.add(var.alias_name)
B
barrierye 已提交
485

B
barriery 已提交
486
    def _pack_inference_request(self, feed, fetch, is_python, log_id):
487
        req = multi_lang_general_model_service_pb2.InferenceRequest()
B
barrierye 已提交
488
        req.fetch_var_names.extend(fetch)
B
barrierye 已提交
489
        req.is_python = is_python
B
barriery 已提交
490
        req.log_id = log_id
B
barrierye 已提交
491 492 493 494 495 496 497
        feed_batch = None
        if isinstance(feed, dict):
            feed_batch = [feed]
        elif isinstance(feed, list):
            feed_batch = feed
        else:
            raise Exception("{} not support".format(type(feed)))
W
WangXi 已提交
498
        req.feed_var_names.extend(feed_batch[0].keys())
B
barrierye 已提交
499
        init_feed_names = False
B
barrierye 已提交
500
        for feed_data in feed_batch:
501
            inst = multi_lang_general_model_service_pb2.FeedInst()
B
barrierye 已提交
502
            for name in req.feed_var_names:
503
                tensor = multi_lang_general_model_service_pb2.Tensor()
B
barrierye 已提交
504 505
                var = feed_data[name]
                v_type = self.feed_types_[name]
B
barrierye 已提交
506 507 508 509 510 511 512
                if is_python:
                    data = None
                    if isinstance(var, list):
                        if v_type == 0:  # int64
                            data = np.array(var, dtype="int64")
                        elif v_type == 1:  # float32
                            data = np.array(var, dtype="float32")
B
barrierye 已提交
513 514
                        elif v_type == 2:  # int32
                            data = np.array(var, dtype="int32")
B
barrierye 已提交
515
                        else:
B
barrierye 已提交
516 517
                            raise Exception("error tensor value type.")
                    elif isinstance(var, np.ndarray):
B
barrierye 已提交
518
                        data = var
B
barrierye 已提交
519 520 521 522 523 524 525 526 527
                        if v_type == 0:
                            if data.dtype != 'int64':
                                data = data.astype("int64")
                        elif v_type == 1:
                            if data.dtype != 'float32':
                                data = data.astype("float32")
                        elif v_type == 2:
                            if data.dtype != 'int32':
                                data = data.astype("int32")
B
barrierye 已提交
528 529 530 531
                        else:
                            raise Exception("error tensor value type.")
                    else:
                        raise Exception("var must be list or ndarray.")
B
barrierye 已提交
532
                    tensor.data = data.tobytes()
B
barrierye 已提交
533
                else:
B
barrierye 已提交
534 535 536 537 538 539 540 541
                    if isinstance(var, np.ndarray):
                        if v_type == 0:  # int64
                            tensor.int64_data.extend(
                                var.reshape(-1).astype("int64").tolist())
                        elif v_type == 1:
                            tensor.float_data.extend(
                                var.reshape(-1).astype('float32').tolist())
                        elif v_type == 2:
542
                            tensor.int_data.extend(
B
barrierye 已提交
543
                                var.reshape(-1).astype('int32').tolist())
B
barrierye 已提交
544
                        else:
B
barrierye 已提交
545 546 547
                            raise Exception("error tensor value type.")
                    elif isinstance(var, list):
                        if v_type == 0:
B
barrierye 已提交
548
                            tensor.int64_data.extend(self._flatten_list(var))
B
barrierye 已提交
549
                        elif v_type == 1:
B
barrierye 已提交
550
                            tensor.float_data.extend(self._flatten_list(var))
B
barrierye 已提交
551
                        elif v_type == 2:
552
                            tensor.int_data.extend(self._flatten_list(var))
B
barrierye 已提交
553 554
                        else:
                            raise Exception("error tensor value type.")
B
barrierye 已提交
555
                    else:
B
barrierye 已提交
556
                        raise Exception("var must be list or ndarray.")
B
barrierye 已提交
557
                if isinstance(var, np.ndarray):
B
barrierye 已提交
558
                    tensor.shape.extend(list(var.shape))
B
barrierye 已提交
559
                else:
B
barrierye 已提交
560 561 562
                    tensor.shape.extend(self.feed_shapes_[name])
                inst.tensor_array.append(tensor)
            req.insts.append(inst)
B
barrierye 已提交
563
        return req
B
barrierye 已提交
564

565 566 567
    def _unpack_inference_response(self, resp, fetch, is_python,
                                   need_variant_tag):
        if resp.err_code != 0:
B
fix bug  
barrierye 已提交
568 569
            return None
        tag = resp.tag
B
barrierye 已提交
570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585
        multi_result_map = {}
        for model_result in resp.outputs:
            inst = model_result.insts[0]
            result_map = {}
            for i, name in enumerate(fetch):
                var = inst.tensor_array[i]
                v_type = self.fetch_types_[name]
                if is_python:
                    if v_type == 0:  # int64
                        result_map[name] = np.frombuffer(
                            var.data, dtype="int64")
                    elif v_type == 1:  # float32
                        result_map[name] = np.frombuffer(
                            var.data, dtype="float32")
                    else:
                        raise Exception("error type.")
B
barrierye 已提交
586
                else:
B
barrierye 已提交
587 588 589 590 591 592 593 594 595 596 597 598 599 600
                    if v_type == 0:  # int64
                        result_map[name] = np.array(
                            list(var.int64_data), dtype="int64")
                    elif v_type == 1:  # float32
                        result_map[name] = np.array(
                            list(var.float_data), dtype="float32")
                    else:
                        raise Exception("error type.")
                result_map[name].shape = list(var.shape)
                if name in self.lod_tensor_set_:
                    result_map["{}.lod".format(name)] = np.array(list(var.lod))
            multi_result_map[model_result.engine_name] = result_map
        ret = None
        if len(resp.outputs) == 1:
B
barrierye 已提交
601
            ret = list(multi_result_map.values())[0]
B
barrierye 已提交
602 603
        else:
            ret = multi_result_map
B
barrierye 已提交
604

605
        ret["serving_status_code"] = 0
B
barrierye 已提交
606
        return ret if not need_variant_tag else [ret, tag]
607

B
barrierye 已提交
608
    def _done_callback_func(self, fetch, is_python, need_variant_tag):
609
        def unpack_resp(resp):
610 611
            return self._unpack_inference_response(resp, fetch, is_python,
                                                   need_variant_tag)
B
barrierye 已提交
612

613 614
        return unpack_resp

W
WangXi 已提交
615 616 617
    def get_feed_names(self):
        return self.feed_names_

B
barrierye 已提交
618 619 620 621 622
    def predict(self,
                feed,
                fetch,
                need_variant_tag=False,
                asyn=False,
B
barriery 已提交
623 624
                is_python=True,
                log_id=0):
625
        if not asyn:
B
barrierye 已提交
626
            try:
B
barrierye 已提交
627 628
                self.profile_.record('py_prepro_0')
                req = self._pack_inference_request(
B
barriery 已提交
629
                    feed, fetch, is_python=is_python, log_id=log_id)
B
barrierye 已提交
630 631 632
                self.profile_.record('py_prepro_1')

                self.profile_.record('py_client_infer_0')
B
barrierye 已提交
633
                resp = self.stub_.Inference(req, timeout=self.rpc_timeout_s_)
B
barrierye 已提交
634 635 636 637
                self.profile_.record('py_client_infer_1')

                self.profile_.record('py_postpro_0')
                ret = self._unpack_inference_response(
B
barrierye 已提交
638 639 640 641
                    resp,
                    fetch,
                    is_python=is_python,
                    need_variant_tag=need_variant_tag)
B
barrierye 已提交
642 643 644
                self.profile_.record('py_postpro_1')
                self.profile_.print_profile()
                return ret
B
barrierye 已提交
645
            except grpc.RpcError as e:
M
fix ut  
MRXLT 已提交
646
                return {"serving_status_code": e.code()}
647
        else:
B
barriery 已提交
648 649
            req = self._pack_inference_request(
                feed, fetch, is_python=is_python, log_id=log_id)
650 651
            call_future = self.stub_.Inference.future(
                req, timeout=self.rpc_timeout_s_)
652
            return MultiLangPredictFuture(
B
barrierye 已提交
653 654 655 656 657
                call_future,
                self._done_callback_func(
                    fetch,
                    is_python=is_python,
                    need_variant_tag=need_variant_tag))
658 659 660 661 662 663 664 665


class MultiLangPredictFuture(object):
    def __init__(self, call_future, callback_func):
        self.call_future_ = call_future
        self.callback_func_ = callback_func

    def result(self):
B
barrierye 已提交
666 667 668
        try:
            resp = self.call_future_.result()
        except grpc.RpcError as e:
669
            return {"serving_status_code": e.code()}
670
        return self.callback_func_(resp)
W
WangXi 已提交
671 672 673 674 675 676 677

    def add_done_callback(self, fn):
        def __fn__(call_future):
            assert call_future == self.call_future_
            fn(self)

        self.call_future_.add_done_callback(__fn__)