|
1
|
|
|
import os |
|
2
|
|
|
import logging |
|
3
|
|
|
import pdb |
|
4
|
|
|
import string |
|
5
|
|
|
import time |
|
6
|
|
|
import re |
|
7
|
|
|
import random |
|
8
|
|
|
import traceback |
|
9
|
|
|
import json |
|
10
|
|
|
import csv |
|
11
|
|
|
from multiprocessing import Process |
|
12
|
|
|
import numpy as np |
|
13
|
|
|
from yaml import full_load, dump |
|
14
|
|
|
from concurrent import futures |
|
15
|
|
|
from client import MilvusClient |
|
16
|
|
|
import parser |
|
17
|
|
|
from runner import Runner |
|
18
|
|
|
from milvus_metrics.api import report |
|
19
|
|
|
from milvus_metrics.models import Env, Hardware, Server, Metric |
|
20
|
|
|
import utils |
|
21
|
|
|
|
|
22
|
|
|
logger = logging.getLogger("milvus_benchmark.k8s_runner") |
|
23
|
|
|
namespace = "milvus" |
|
24
|
|
|
default_port = 19530 |
|
25
|
|
|
DELETE_INTERVAL_TIME = 5 |
|
26
|
|
|
# INSERT_INTERVAL = 100000 |
|
27
|
|
|
INSERT_INTERVAL = 50000 |
|
28
|
|
|
timestamp = int(time.time()) |
|
29
|
|
|
default_path = "/var/lib/milvus" |
|
30
|
|
|
|
|
31
|
|
|
|
|
32
|
|
|
class K8sRunner(Runner): |
|
33
|
|
|
def __init__(self): |
|
34
|
|
|
""" |
|
35
|
|
|
Run with helm mode. |
|
36
|
|
|
|
|
37
|
|
|
Upload test result after tests finished |
|
38
|
|
|
""" |
|
39
|
|
|
super(K8sRunner, self).__init__() |
|
40
|
|
|
self.service_name = utils.get_unique_name() |
|
41
|
|
|
self.host = None |
|
42
|
|
|
self.port = default_port |
|
43
|
|
|
self.hostname = None |
|
44
|
|
|
self.env_value = None |
|
45
|
|
|
|
|
46
|
|
|
def init_env(self, server_config, server_host, deploy_mode, image_type, image_tag): |
|
47
|
|
|
""" |
|
48
|
|
|
Deploy start server with using helm and clean up env. |
|
49
|
|
|
|
|
50
|
|
|
If deploy or start failed |
|
51
|
|
|
""" |
|
52
|
|
|
logger.debug("Tests run on server host:") |
|
53
|
|
|
logger.debug(server_host) |
|
54
|
|
|
self.hostname = server_host |
|
55
|
|
|
# update values |
|
56
|
|
|
helm_path = os.path.join(os.getcwd(), "../milvus-helm/charts/milvus") |
|
57
|
|
|
values_file_path = helm_path+"/values.yaml" |
|
58
|
|
|
if not os.path.exists(values_file_path): |
|
59
|
|
|
raise Exception("File %s not existed" % values_file_path) |
|
60
|
|
|
if server_config: |
|
61
|
|
|
utils.update_values(values_file_path, deploy_mode, server_host, server_config) |
|
62
|
|
|
try: |
|
63
|
|
|
logger.debug("Start install server") |
|
64
|
|
|
self.host = utils.helm_install_server(helm_path, deploy_mode, image_tag, image_type, self.service_name, namespace) |
|
65
|
|
|
except Exception as e: |
|
66
|
|
|
logger.error("Helm install server failed: %s" % (str(e))) |
|
67
|
|
|
logger.error(traceback.format_exc()) |
|
68
|
|
|
logger.debug(server_config) |
|
69
|
|
|
self.clean_up() |
|
70
|
|
|
return False |
|
71
|
|
|
# for debugging |
|
72
|
|
|
if not self.host: |
|
73
|
|
|
logger.error("Helm install server failed") |
|
74
|
|
|
self.clean_up() |
|
75
|
|
|
return False |
|
76
|
|
|
return True |
|
77
|
|
|
|
|
78
|
|
|
def clean_up(self): |
|
79
|
|
|
""" |
|
80
|
|
|
Stop server with using helm. |
|
81
|
|
|
|
|
82
|
|
|
""" |
|
83
|
|
|
logger.debug("Start clean up: %s" % self.service_name) |
|
84
|
|
|
utils.helm_del_server(self.service_name, namespace) |
|
85
|
|
|
|
|
86
|
|
|
def report_wrapper(self, milvus_instance, env_value, hostname, collection_info, index_info, search_params, run_params=None): |
|
87
|
|
|
""" |
|
88
|
|
|
upload test result |
|
89
|
|
|
""" |
|
90
|
|
|
metric = Metric() |
|
91
|
|
|
metric.set_run_id(timestamp) |
|
92
|
|
|
metric.env = Env(env_value) |
|
93
|
|
|
metric.env.OMP_NUM_THREADS = 0 |
|
94
|
|
|
metric.hardware = Hardware(name=hostname) |
|
95
|
|
|
server_version = milvus_instance.get_server_version() |
|
96
|
|
|
server_mode = milvus_instance.get_server_mode() |
|
97
|
|
|
commit = milvus_instance.get_server_commit() |
|
98
|
|
|
metric.server = Server(version=server_version, mode=server_mode, build_commit=commit) |
|
99
|
|
|
metric.collection = collection_info |
|
100
|
|
|
metric.index = index_info |
|
101
|
|
|
metric.search = search_params |
|
102
|
|
|
metric.run_params = run_params |
|
103
|
|
|
return metric |
|
104
|
|
|
|
|
105
|
|
|
def run(self, run_type, collection): |
|
106
|
|
|
""" |
|
107
|
|
|
override runner.run |
|
108
|
|
|
""" |
|
109
|
|
|
logger.debug(run_type) |
|
110
|
|
|
logger.debug(collection) |
|
111
|
|
|
collection_name = collection["collection_name"] if "collection_name" in collection else None |
|
112
|
|
|
milvus_instance = MilvusClient(collection_name=collection_name, host=self.host) |
|
113
|
|
|
self.env_value = milvus_instance.get_server_config() |
|
114
|
|
|
|
|
115
|
|
|
# ugly implemention |
|
116
|
|
|
# remove some parts of result before uploading results |
|
117
|
|
|
self.env_value.pop("logs") |
|
118
|
|
|
if milvus_instance.get_server_mode() == "CPU": |
|
119
|
|
|
if "gpu" in self.env_value: |
|
120
|
|
|
self.env_value.pop("gpu") |
|
121
|
|
|
elif "cache.enable" in self.env_value["gpu"]: |
|
122
|
|
|
self.env_value["gpu"].pop("cache.enable") |
|
123
|
|
|
|
|
124
|
|
|
self.env_value.pop("network") |
|
125
|
|
|
|
|
126
|
|
|
if run_type == "insert_performance": |
|
127
|
|
|
(data_type, collection_size, index_file_size, dimension, metric_type) = parser.collection_parser(collection_name) |
|
128
|
|
|
ni_per = collection["ni_per"] |
|
129
|
|
|
build_index = collection["build_index"] |
|
130
|
|
|
if milvus_instance.exists_collection(): |
|
131
|
|
|
milvus_instance.drop() |
|
132
|
|
|
time.sleep(10) |
|
133
|
|
|
index_info = {} |
|
134
|
|
|
search_params = {} |
|
135
|
|
|
milvus_instance.create_collection(collection_name, dimension, index_file_size, metric_type) |
|
136
|
|
|
if build_index is True: |
|
137
|
|
|
index_type = collection["index_type"] |
|
138
|
|
|
index_param = collection["index_param"] |
|
139
|
|
|
index_info = { |
|
140
|
|
|
"index_type": index_type, |
|
141
|
|
|
"index_param": index_param |
|
142
|
|
|
} |
|
143
|
|
|
milvus_instance.create_index(index_type, index_param) |
|
144
|
|
|
logger.debug(milvus_instance.describe_index()) |
|
145
|
|
|
res = self.do_insert(milvus_instance, collection_name, data_type, dimension, collection_size, ni_per) |
|
146
|
|
|
logger.info(res) |
|
147
|
|
|
if "flush" in collection and collection["flush"] == "no": |
|
148
|
|
|
logger.debug("No manual flush") |
|
149
|
|
|
else: |
|
150
|
|
|
milvus_instance.flush() |
|
151
|
|
|
logger.debug(milvus_instance.count()) |
|
152
|
|
|
collection_info = { |
|
153
|
|
|
"dimension": dimension, |
|
154
|
|
|
"metric_type": metric_type, |
|
155
|
|
|
"dataset_name": collection_name |
|
156
|
|
|
} |
|
157
|
|
|
metric = self.report_wrapper(milvus_instance, self.env_value, self.hostname, collection_info, index_info, search_params) |
|
158
|
|
|
metric.metrics = { |
|
159
|
|
|
"type": run_type, |
|
160
|
|
|
"value": { |
|
161
|
|
|
"total_time": res["total_time"], |
|
162
|
|
|
"qps": res["qps"], |
|
163
|
|
|
"ni_time": res["ni_time"] |
|
164
|
|
|
} |
|
165
|
|
|
} |
|
166
|
|
|
report(metric) |
|
167
|
|
|
if build_index is True: |
|
168
|
|
|
logger.debug("Start build index for last file") |
|
169
|
|
|
milvus_instance.create_index(index_type, index_param) |
|
|
|
|
|
|
170
|
|
|
logger.debug(milvus_instance.describe_index()) |
|
171
|
|
|
|
|
172
|
|
|
elif run_type == "insert_debug_performance": |
|
173
|
|
|
(data_type, collection_size, index_file_size, dimension, metric_type) = parser.collection_parser(collection_name) |
|
174
|
|
|
ni_per = collection["ni_per"] |
|
175
|
|
|
if milvus_instance.exists_collection(): |
|
176
|
|
|
milvus_instance.drop() |
|
177
|
|
|
time.sleep(10) |
|
178
|
|
|
index_info = {} |
|
179
|
|
|
search_params = {} |
|
180
|
|
|
milvus_instance.create_collection(collection_name, dimension, index_file_size, metric_type) |
|
181
|
|
|
insert_vectors = [[random.random() for _ in range(dimension)] for _ in range(ni_per)] |
|
182
|
|
|
start_time = time.time() |
|
183
|
|
|
i = 0 |
|
184
|
|
|
while time.time() < start_time + 2 * 24 * 3600: |
|
185
|
|
|
i = i + 1 |
|
186
|
|
|
logger.debug(i) |
|
187
|
|
|
logger.debug("Row count: %d" % milvus_instance.count()) |
|
188
|
|
|
milvus_instance.insert(insert_vectors) |
|
189
|
|
|
time.sleep(0.1) |
|
190
|
|
|
|
|
191
|
|
|
elif run_type == "insert_performance_multi_collections": |
|
192
|
|
|
(data_type, collection_size, index_file_size, dimension, metric_type) = parser.collection_parser(collection_name) |
|
193
|
|
|
ni_per = collection["ni_per"] |
|
194
|
|
|
build_index = collection["build_index"] |
|
195
|
|
|
if milvus_instance.exists_collection(): |
|
196
|
|
|
milvus_instance.drop() |
|
197
|
|
|
time.sleep(10) |
|
198
|
|
|
index_info = {} |
|
199
|
|
|
search_params = {} |
|
200
|
|
|
milvus_instance.create_collection(collection_name, dimension, index_file_size, metric_type) |
|
201
|
|
|
if build_index is True: |
|
202
|
|
|
index_type = collection["index_type"] |
|
203
|
|
|
index_param = collection["index_param"] |
|
204
|
|
|
index_info = { |
|
205
|
|
|
"index_type": index_type, |
|
206
|
|
|
"index_param": index_param |
|
207
|
|
|
} |
|
208
|
|
|
milvus_instance.create_index(index_type, index_param) |
|
209
|
|
|
logger.debug(milvus_instance.describe_index()) |
|
210
|
|
|
res = self.do_insert(milvus_instance, collection_name, data_type, dimension, collection_size, ni_per) |
|
211
|
|
|
logger.info(res) |
|
212
|
|
|
milvus_instance.flush() |
|
213
|
|
|
collection_info = { |
|
214
|
|
|
"dimension": dimension, |
|
215
|
|
|
"metric_type": metric_type, |
|
216
|
|
|
"dataset_name": collection_name |
|
217
|
|
|
} |
|
218
|
|
|
metric = self.report_wrapper(milvus_instance, self.env_value, self.hostname, collection_info, index_info, search_params) |
|
219
|
|
|
metric.metrics = { |
|
220
|
|
|
"type": run_type, |
|
221
|
|
|
"value": { |
|
222
|
|
|
"total_time": res["total_time"], |
|
223
|
|
|
"qps": res["qps"], |
|
224
|
|
|
"ni_time": res["ni_time"] |
|
225
|
|
|
} |
|
226
|
|
|
} |
|
227
|
|
|
report(metric) |
|
228
|
|
|
if build_index is True: |
|
229
|
|
|
logger.debug("Start build index for last file") |
|
230
|
|
|
milvus_instance.create_index(index_type, index_param) |
|
231
|
|
|
logger.debug(milvus_instance.describe_index()) |
|
232
|
|
|
|
|
233
|
|
|
elif run_type == "insert_flush_performance": |
|
234
|
|
|
(data_type, collection_size, index_file_size, dimension, metric_type) = parser.collection_parser(collection_name) |
|
235
|
|
|
ni_per = collection["ni_per"] |
|
236
|
|
|
if milvus_instance.exists_collection(): |
|
237
|
|
|
milvus_instance.drop() |
|
238
|
|
|
time.sleep(10) |
|
239
|
|
|
index_info = {} |
|
240
|
|
|
search_params = {} |
|
241
|
|
|
milvus_instance.create_collection(collection_name, dimension, index_file_size, metric_type) |
|
242
|
|
|
res = self.do_insert(milvus_instance, collection_name, data_type, dimension, collection_size, ni_per) |
|
243
|
|
|
logger.info(res) |
|
244
|
|
|
logger.debug(milvus_instance.count()) |
|
245
|
|
|
start_time = time.time() |
|
246
|
|
|
milvus_instance.flush() |
|
247
|
|
|
end_time = time.time() |
|
248
|
|
|
logger.debug(milvus_instance.count()) |
|
249
|
|
|
collection_info = { |
|
250
|
|
|
"dimension": dimension, |
|
251
|
|
|
"metric_type": metric_type, |
|
252
|
|
|
"dataset_name": collection_name |
|
253
|
|
|
} |
|
254
|
|
|
metric = self.report_wrapper(milvus_instance, self.env_value, self.hostname, collection_info, index_info, search_params) |
|
255
|
|
|
metric.metrics = { |
|
256
|
|
|
"type": run_type, |
|
257
|
|
|
"value": { |
|
258
|
|
|
"flush_time": round(end_time - start_time, 1) |
|
259
|
|
|
} |
|
260
|
|
|
} |
|
261
|
|
|
report(metric) |
|
262
|
|
|
|
|
263
|
|
|
elif run_type == "build_performance": |
|
264
|
|
|
(data_type, collection_size, index_file_size, dimension, metric_type) = parser.collection_parser(collection_name) |
|
265
|
|
|
index_type = collection["index_type"] |
|
266
|
|
|
index_param = collection["index_param"] |
|
267
|
|
|
collection_info = { |
|
268
|
|
|
"dimension": dimension, |
|
269
|
|
|
"metric_type": metric_type, |
|
270
|
|
|
"index_file_size": index_file_size, |
|
271
|
|
|
"dataset_name": collection_name |
|
272
|
|
|
} |
|
273
|
|
|
index_info = { |
|
274
|
|
|
"index_type": index_type, |
|
275
|
|
|
"index_param": index_param |
|
276
|
|
|
} |
|
277
|
|
|
if not milvus_instance.exists_collection(): |
|
278
|
|
|
logger.error("Table name: %s not existed" % collection_name) |
|
279
|
|
|
return |
|
280
|
|
|
search_params = {} |
|
281
|
|
|
start_time = time.time() |
|
282
|
|
|
# drop index |
|
283
|
|
|
logger.debug("Drop index") |
|
284
|
|
|
milvus_instance.drop_index() |
|
285
|
|
|
start_mem_usage = milvus_instance.get_mem_info()["memory_used"] |
|
286
|
|
|
milvus_instance.create_index(index_type, index_param) |
|
287
|
|
|
logger.debug(milvus_instance.describe_index()) |
|
288
|
|
|
logger.debug(milvus_instance.count()) |
|
289
|
|
|
end_time = time.time() |
|
290
|
|
|
end_mem_usage = milvus_instance.get_mem_info()["memory_used"] |
|
291
|
|
|
metric = self.report_wrapper(milvus_instance, self.env_value, self.hostname, collection_info, index_info, search_params) |
|
292
|
|
|
metric.metrics = { |
|
293
|
|
|
"type": "build_performance", |
|
294
|
|
|
"value": { |
|
295
|
|
|
"build_time": round(end_time - start_time, 1), |
|
296
|
|
|
"start_mem_usage": start_mem_usage, |
|
297
|
|
|
"end_mem_usage": end_mem_usage, |
|
298
|
|
|
"diff_mem": end_mem_usage - start_mem_usage |
|
299
|
|
|
} |
|
300
|
|
|
} |
|
301
|
|
|
report(metric) |
|
302
|
|
|
|
|
303
|
|
|
elif run_type == "delete_performance": |
|
304
|
|
|
(data_type, collection_size, index_file_size, dimension, metric_type) = parser.collection_parser(collection_name) |
|
305
|
|
|
ni_per = collection["ni_per"] |
|
306
|
|
|
search_params = {} |
|
307
|
|
|
collection_info = { |
|
308
|
|
|
"dimension": dimension, |
|
309
|
|
|
"metric_type": metric_type, |
|
310
|
|
|
"dataset_name": collection_name |
|
311
|
|
|
} |
|
312
|
|
|
if not milvus_instance.exists_collection(): |
|
313
|
|
|
logger.error("Table name: %s not existed" % collection_name) |
|
314
|
|
|
return |
|
315
|
|
|
length = milvus_instance.count() |
|
316
|
|
|
logger.info(length) |
|
317
|
|
|
index_info = milvus_instance.describe_index() |
|
318
|
|
|
logger.info(index_info) |
|
319
|
|
|
ids = [i for i in range(length)] |
|
320
|
|
|
loops = int(length / ni_per) |
|
321
|
|
|
milvus_instance.preload_collection() |
|
322
|
|
|
start_mem_usage = milvus_instance.get_mem_info()["memory_used"] |
|
323
|
|
|
start_time = time.time() |
|
324
|
|
|
for i in range(loops): |
|
325
|
|
|
delete_ids = ids[i*ni_per : i*ni_per+ni_per] |
|
326
|
|
|
logger.debug("Delete %d - %d" % (delete_ids[0], delete_ids[-1])) |
|
327
|
|
|
milvus_instance.delete(delete_ids) |
|
328
|
|
|
# milvus_instance.flush() |
|
329
|
|
|
logger.debug("Table row counts: %d" % milvus_instance.count()) |
|
330
|
|
|
logger.debug("Table row counts: %d" % milvus_instance.count()) |
|
331
|
|
|
milvus_instance.flush() |
|
332
|
|
|
end_time = time.time() |
|
333
|
|
|
end_mem_usage = milvus_instance.get_mem_info()["memory_used"] |
|
334
|
|
|
logger.debug("Table row counts: %d" % milvus_instance.count()) |
|
335
|
|
|
metric = self.report_wrapper(milvus_instance, self.env_value, self.hostname, collection_info, index_info, search_params) |
|
336
|
|
|
metric.metrics = { |
|
337
|
|
|
"type": "delete_performance", |
|
338
|
|
|
"value": { |
|
339
|
|
|
"delete_time": round(end_time - start_time, 1), |
|
340
|
|
|
"start_mem_usage": start_mem_usage, |
|
341
|
|
|
"end_mem_usage": end_mem_usage, |
|
342
|
|
|
"diff_mem": end_mem_usage - start_mem_usage |
|
343
|
|
|
} |
|
344
|
|
|
} |
|
345
|
|
|
report(metric) |
|
346
|
|
|
|
|
347
|
|
|
elif run_type == "get_ids_performance": |
|
348
|
|
|
(data_type, collection_size, index_file_size, dimension, metric_type) = parser.collection_parser(collection_name) |
|
349
|
|
|
ids_length_per_segment = collection["ids_length_per_segment"] |
|
350
|
|
|
if not milvus_instance.exists_collection(): |
|
351
|
|
|
logger.error("Table name: %s not existed" % collection_name) |
|
352
|
|
|
return |
|
353
|
|
|
collection_info = { |
|
354
|
|
|
"dimension": dimension, |
|
355
|
|
|
"metric_type": metric_type, |
|
356
|
|
|
"index_file_size": index_file_size, |
|
357
|
|
|
"dataset_name": collection_name |
|
358
|
|
|
} |
|
359
|
|
|
search_params = {} |
|
360
|
|
|
logger.info(milvus_instance.count()) |
|
361
|
|
|
index_info = milvus_instance.describe_index() |
|
362
|
|
|
logger.info(index_info) |
|
363
|
|
|
for ids_num in ids_length_per_segment: |
|
364
|
|
|
segment_num, get_ids = milvus_instance.get_rand_ids_each_segment(ids_num) |
|
365
|
|
|
start_time = time.time() |
|
366
|
|
|
_ = milvus_instance.get_entities(get_ids) |
|
367
|
|
|
total_time = time.time() - start_time |
|
368
|
|
|
avg_time = total_time / segment_num |
|
369
|
|
|
run_params = {"ids_num": ids_num} |
|
370
|
|
|
logger.info("Segment num: %d, ids num per segment: %d, run_time: %f" % (segment_num, ids_num, total_time)) |
|
371
|
|
|
metric = self.report_wrapper(milvus_instance, self.env_value, self.hostname, collection_info, index_info, search_params, run_params=run_params) |
|
372
|
|
|
metric.metrics = { |
|
373
|
|
|
"type": run_type, |
|
374
|
|
|
"value": { |
|
375
|
|
|
"total_time": round(total_time, 1), |
|
376
|
|
|
"avg_time": round(avg_time, 1) |
|
377
|
|
|
} |
|
378
|
|
|
} |
|
379
|
|
|
report(metric) |
|
380
|
|
|
|
|
381
|
|
|
elif run_type == "search_performance": |
|
382
|
|
|
(data_type, collection_size, index_file_size, dimension, metric_type) = parser.collection_parser(collection_name) |
|
383
|
|
|
run_count = collection["run_count"] |
|
384
|
|
|
top_ks = collection["top_ks"] |
|
385
|
|
|
nqs = collection["nqs"] |
|
386
|
|
|
search_params = collection["search_params"] |
|
387
|
|
|
collection_info = { |
|
388
|
|
|
"dimension": dimension, |
|
389
|
|
|
"metric_type": metric_type, |
|
390
|
|
|
"index_file_size": index_file_size, |
|
391
|
|
|
"dataset_name": collection_name |
|
392
|
|
|
} |
|
393
|
|
|
if not milvus_instance.exists_collection(): |
|
394
|
|
|
logger.error("Table name: %s not existed" % collection_name) |
|
395
|
|
|
return |
|
396
|
|
|
|
|
397
|
|
|
logger.info(milvus_instance.count()) |
|
398
|
|
|
index_info = milvus_instance.describe_index() |
|
399
|
|
|
logger.info(index_info) |
|
400
|
|
|
milvus_instance.preload_collection() |
|
401
|
|
|
logger.info("Start warm up query") |
|
402
|
|
|
res = self.do_query(milvus_instance, collection_name, [1], [1], 2, search_param=search_params[0]) |
|
403
|
|
|
logger.info("End warm up query") |
|
404
|
|
|
for search_param in search_params: |
|
405
|
|
|
logger.info("Search param: %s" % json.dumps(search_param)) |
|
406
|
|
|
res = self.do_query(milvus_instance, collection_name, top_ks, nqs, run_count, search_param) |
|
407
|
|
|
headers = ["Nq/Top-k"] |
|
408
|
|
|
headers.extend([str(top_k) for top_k in top_ks]) |
|
409
|
|
|
logger.info("Search param: %s" % json.dumps(search_param)) |
|
410
|
|
|
utils.print_table(headers, nqs, res) |
|
411
|
|
|
for index_nq, nq in enumerate(nqs): |
|
412
|
|
|
for index_top_k, top_k in enumerate(top_ks): |
|
413
|
|
|
search_param_group = { |
|
414
|
|
|
"nq": nq, |
|
415
|
|
|
"topk": top_k, |
|
416
|
|
|
"search_param": search_param |
|
417
|
|
|
} |
|
418
|
|
|
search_time = res[index_nq][index_top_k] |
|
419
|
|
|
metric = self.report_wrapper(milvus_instance, self.env_value, self.hostname, collection_info, index_info, search_param_group) |
|
420
|
|
|
metric.metrics = { |
|
421
|
|
|
"type": "search_performance", |
|
422
|
|
|
"value": { |
|
423
|
|
|
"search_time": search_time |
|
424
|
|
|
} |
|
425
|
|
|
} |
|
426
|
|
|
report(metric) |
|
427
|
|
|
|
|
428
|
|
|
elif run_type == "locust_search_performance": |
|
429
|
|
|
(data_type, collection_size, index_file_size, dimension, metric_type) = parser.collection_parser( |
|
430
|
|
|
collection_name) |
|
431
|
|
|
### clear db |
|
432
|
|
|
### spawn locust requests |
|
433
|
|
|
collection_num = collection["collection_num"] |
|
434
|
|
|
task = collection["task"] |
|
435
|
|
|
# . generate task code |
|
436
|
|
|
task_file = utils.get_unique_name() |
|
437
|
|
|
task_file_script = task_file + '.py' |
|
438
|
|
|
task_file_csv = task_file + '_stats.csv' |
|
439
|
|
|
task_type = task["type"] |
|
440
|
|
|
connection_type = "single" |
|
441
|
|
|
connection_num = task["connection_num"] |
|
442
|
|
|
if connection_num > 1: |
|
443
|
|
|
connection_type = "multi" |
|
444
|
|
|
clients_num = task["clients_num"] |
|
445
|
|
|
hatch_rate = task["hatch_rate"] |
|
446
|
|
|
during_time = task["during_time"] |
|
447
|
|
|
def_name = task_type |
|
448
|
|
|
task_params = task["params"] |
|
449
|
|
|
collection_names = [] |
|
450
|
|
|
for i in range(collection_num): |
|
451
|
|
|
suffix = "".join(random.choice(string.ascii_letters + string.digits) for _ in range(5)) |
|
452
|
|
|
collection_names.append(collection_name + "_" + suffix) |
|
453
|
|
|
# ##### |
|
454
|
|
|
ni_per = collection["ni_per"] |
|
455
|
|
|
build_index = collection["build_index"] |
|
456
|
|
|
# TODO: debug |
|
457
|
|
|
for c_name in collection_names: |
|
458
|
|
|
milvus_instance = MilvusClient(collection_name=c_name, host=self.host, port=self.port) |
|
459
|
|
|
if milvus_instance.exists_collection(collection_name=c_name): |
|
460
|
|
|
milvus_instance.drop(name=c_name) |
|
461
|
|
|
time.sleep(10) |
|
462
|
|
|
milvus_instance.create_collection(c_name, dimension, index_file_size, metric_type) |
|
463
|
|
|
index_info = { |
|
464
|
|
|
"build_index": build_index |
|
465
|
|
|
} |
|
466
|
|
|
if build_index is True: |
|
467
|
|
|
index_type = collection["index_type"] |
|
468
|
|
|
index_param = collection["index_param"] |
|
469
|
|
|
index_info.update({ |
|
470
|
|
|
"index_type": index_type, |
|
471
|
|
|
"index_param": index_param |
|
472
|
|
|
}) |
|
473
|
|
|
milvus_instance.create_index(index_type, index_param) |
|
474
|
|
|
logger.debug(milvus_instance.describe_index()) |
|
475
|
|
|
res = self.do_insert(milvus_instance, c_name, data_type, dimension, collection_size, ni_per) |
|
476
|
|
|
logger.info(res) |
|
477
|
|
|
if "flush" in collection and collection["flush"] == "no": |
|
478
|
|
|
logger.debug("No manual flush") |
|
479
|
|
|
else: |
|
480
|
|
|
milvus_instance.flush() |
|
481
|
|
|
logger.debug("Table row counts: %d" % milvus_instance.count(name=c_name)) |
|
482
|
|
|
if build_index is True: |
|
483
|
|
|
logger.debug("Start build index for last file") |
|
484
|
|
|
milvus_instance.create_index(index_type, index_param) |
|
485
|
|
|
logger.debug(milvus_instance.describe_index()) |
|
486
|
|
|
code_str = """ |
|
487
|
|
|
import random |
|
488
|
|
|
import string |
|
489
|
|
|
from locust import User, task, between |
|
490
|
|
|
from locust_task import MilvusTask |
|
491
|
|
|
from client import MilvusClient |
|
492
|
|
|
|
|
493
|
|
|
host = '%s' |
|
494
|
|
|
port = %s |
|
495
|
|
|
dim = %s |
|
496
|
|
|
connection_type = '%s' |
|
497
|
|
|
collection_names = %s |
|
498
|
|
|
m = MilvusClient(host=host, port=port) |
|
499
|
|
|
|
|
500
|
|
|
|
|
501
|
|
|
def get_collection_name(): |
|
502
|
|
|
return random.choice(collection_names) |
|
503
|
|
|
|
|
504
|
|
|
|
|
505
|
|
|
def get_client(collection_name): |
|
506
|
|
|
if connection_type == 'single': |
|
507
|
|
|
return MilvusTask(m=m) |
|
508
|
|
|
elif connection_type == 'multi': |
|
509
|
|
|
return MilvusTask(connection_type='multi', host=host, port=port, collection_name=collection_name) |
|
510
|
|
|
|
|
511
|
|
|
|
|
512
|
|
|
class QueryTask(User): |
|
513
|
|
|
wait_time = between(0.001, 0.002) |
|
514
|
|
|
|
|
515
|
|
|
@task() |
|
516
|
|
|
def %s(self): |
|
517
|
|
|
top_k = %s |
|
518
|
|
|
X = [[random.random() for i in range(dim)] for i in range(%s)] |
|
519
|
|
|
search_param = %s |
|
520
|
|
|
collection_name = get_collection_name() |
|
521
|
|
|
client = get_client(collection_name) |
|
522
|
|
|
client.query(X, top_k, search_param, collection_name=collection_name) |
|
523
|
|
|
""" % (self.host, self.port, dimension, connection_type, collection_names, def_name, task_params["top_k"], task_params["nq"], task_params["search_param"]) |
|
524
|
|
|
with open(task_file_script, 'w+') as fd: |
|
525
|
|
|
fd.write(code_str) |
|
526
|
|
|
locust_cmd = "locust -f %s --headless --csv=%s -u %d -r %d -t %s" % ( |
|
527
|
|
|
task_file_script, |
|
528
|
|
|
task_file, |
|
529
|
|
|
clients_num, |
|
530
|
|
|
hatch_rate, |
|
531
|
|
|
during_time) |
|
532
|
|
|
logger.info(locust_cmd) |
|
533
|
|
|
try: |
|
534
|
|
|
res = os.system(locust_cmd) |
|
535
|
|
|
except Exception as e: |
|
536
|
|
|
logger.error(str(e)) |
|
537
|
|
|
return |
|
538
|
|
|
|
|
539
|
|
|
# . retrieve and collect test statistics |
|
540
|
|
|
locust_stats = None |
|
541
|
|
|
with open(task_file_csv, newline='') as fd: |
|
542
|
|
|
dr = csv.DictReader(fd) |
|
543
|
|
|
for row in dr: |
|
544
|
|
|
if row["Name"] != "Aggregated": |
|
545
|
|
|
continue |
|
546
|
|
|
locust_stats = row |
|
547
|
|
|
logger.info(locust_stats) |
|
548
|
|
|
# clean up temp files |
|
549
|
|
|
search_params = { |
|
550
|
|
|
"top_k": task_params["top_k"], |
|
551
|
|
|
"nq": task_params["nq"], |
|
552
|
|
|
"nprobe": task_params["search_param"]["nprobe"] |
|
553
|
|
|
} |
|
554
|
|
|
run_params = { |
|
555
|
|
|
"connection_num": connection_num, |
|
556
|
|
|
"clients_num": clients_num, |
|
557
|
|
|
"hatch_rate": hatch_rate, |
|
558
|
|
|
"during_time": during_time |
|
559
|
|
|
} |
|
560
|
|
|
collection_info = { |
|
561
|
|
|
"dimension": dimension, |
|
562
|
|
|
"metric_type": metric_type, |
|
563
|
|
|
"index_file_size": index_file_size, |
|
564
|
|
|
"dataset_name": collection_name |
|
565
|
|
|
} |
|
566
|
|
|
metric = self.report_wrapper(milvus_instance, self.env_value, self.hostname, collection_info, index_info, search_params, run_params) |
|
|
|
|
|
|
567
|
|
|
metric.metrics = { |
|
568
|
|
|
"type": run_type, |
|
569
|
|
|
"value": { |
|
570
|
|
|
"during_time": during_time, |
|
571
|
|
|
"request_count": int(locust_stats["Request Count"]), |
|
572
|
|
|
"failure_count": int(locust_stats["Failure Count"]), |
|
573
|
|
|
"qps": locust_stats["Requests/s"], |
|
574
|
|
|
"min_response_time": int(locust_stats["Min Response Time"]), |
|
575
|
|
|
"max_response_time": int(locust_stats["Max Response Time"]), |
|
576
|
|
|
"median_response_time": int(locust_stats["Median Response Time"]), |
|
577
|
|
|
"avg_response_time": int(locust_stats["Average Response Time"]) |
|
578
|
|
|
} |
|
579
|
|
|
} |
|
580
|
|
|
report(metric) |
|
581
|
|
|
|
|
582
|
|
|
elif run_type == "search_ids_stability": |
|
583
|
|
|
(data_type, collection_size, index_file_size, dimension, metric_type) = parser.collection_parser(collection_name) |
|
584
|
|
|
search_params = collection["search_params"] |
|
585
|
|
|
during_time = collection["during_time"] |
|
586
|
|
|
ids_length = collection["ids_length"] |
|
587
|
|
|
ids = collection["ids"] |
|
588
|
|
|
collection_info = { |
|
589
|
|
|
"dimension": dimension, |
|
590
|
|
|
"metric_type": metric_type, |
|
591
|
|
|
"index_file_size": index_file_size, |
|
592
|
|
|
"dataset_name": collection_name |
|
593
|
|
|
} |
|
594
|
|
|
if not milvus_instance.exists_collection(): |
|
595
|
|
|
logger.error("Table name: %s not existed" % collection_name) |
|
596
|
|
|
return |
|
597
|
|
|
logger.info(milvus_instance.count()) |
|
598
|
|
|
index_info = milvus_instance.describe_index() |
|
599
|
|
|
logger.info(index_info) |
|
600
|
|
|
g_top_k = int(collection["top_ks"].split("-")[1]) |
|
601
|
|
|
l_top_k = int(collection["top_ks"].split("-")[0]) |
|
602
|
|
|
# g_id = int(ids.split("-")[1]) |
|
603
|
|
|
# l_id = int(ids.split("-")[0]) |
|
604
|
|
|
g_id_length = int(ids_length.split("-")[1]) |
|
605
|
|
|
l_id_length = int(ids_length.split("-")[0]) |
|
606
|
|
|
|
|
607
|
|
|
milvus_instance.preload_collection() |
|
608
|
|
|
start_mem_usage = milvus_instance.get_mem_info()["memory_used"] |
|
609
|
|
|
logger.debug(start_mem_usage) |
|
610
|
|
|
start_time = time.time() |
|
611
|
|
|
while time.time() < start_time + during_time * 60: |
|
612
|
|
|
search_param = {} |
|
613
|
|
|
top_k = random.randint(l_top_k, g_top_k) |
|
614
|
|
|
ids_num = random.randint(l_id_length, g_id_length) |
|
615
|
|
|
ids_param = [random.randint(l_id_length, g_id_length) for _ in range(ids_num)] |
|
616
|
|
|
for k, v in search_params.items(): |
|
617
|
|
|
search_param[k] = random.randint(int(v.split("-")[0]), int(v.split("-")[1])) |
|
618
|
|
|
logger.debug("Query top-k: %d, ids_num: %d, param: %s" % (top_k, ids_num, json.dumps(search_param))) |
|
619
|
|
|
result = milvus_instance.query_ids(top_k, ids_param, search_param=search_param) |
|
620
|
|
|
end_mem_usage = milvus_instance.get_mem_info()["memory_used"] |
|
621
|
|
|
metric = self.report_wrapper(milvus_instance, self.env_value, self.hostname, collection_info, index_info, {}) |
|
622
|
|
|
metric.metrics = { |
|
623
|
|
|
"type": "search_ids_stability", |
|
624
|
|
|
"value": { |
|
625
|
|
|
"during_time": during_time, |
|
626
|
|
|
"start_mem_usage": start_mem_usage, |
|
627
|
|
|
"end_mem_usage": end_mem_usage, |
|
628
|
|
|
"diff_mem": end_mem_usage - start_mem_usage |
|
629
|
|
|
} |
|
630
|
|
|
} |
|
631
|
|
|
report(metric) |
|
632
|
|
|
|
|
633
|
|
|
# for sift/deep datasets |
|
634
|
|
|
# TODO: enable |
|
635
|
|
|
elif run_type == "accuracy": |
|
636
|
|
|
(data_type, collection_size, index_file_size, dimension, metric_type) = parser.collection_parser(collection_name) |
|
637
|
|
|
search_params = collection["search_params"] |
|
638
|
|
|
# mapping to search param list |
|
639
|
|
|
search_params = self.generate_combinations(search_params) |
|
640
|
|
|
|
|
641
|
|
|
top_ks = collection["top_ks"] |
|
642
|
|
|
nqs = collection["nqs"] |
|
643
|
|
|
collection_info = { |
|
644
|
|
|
"dimension": dimension, |
|
645
|
|
|
"metric_type": metric_type, |
|
646
|
|
|
"index_file_size": index_file_size, |
|
647
|
|
|
"dataset_name": collection_name |
|
648
|
|
|
} |
|
649
|
|
|
if not milvus_instance.exists_collection(): |
|
650
|
|
|
logger.error("Table name: %s not existed" % collection_name) |
|
651
|
|
|
return |
|
652
|
|
|
logger.info(milvus_instance.count()) |
|
653
|
|
|
index_info = milvus_instance.describe_index() |
|
654
|
|
|
logger.info(index_info) |
|
655
|
|
|
milvus_instance.preload_collection() |
|
656
|
|
|
true_ids_all = self.get_groundtruth_ids(collection_size) |
|
657
|
|
|
for search_param in search_params: |
|
658
|
|
|
for top_k in top_ks: |
|
659
|
|
|
for nq in nqs: |
|
660
|
|
|
# total = 0 |
|
661
|
|
|
search_param_group = { |
|
662
|
|
|
"nq": nq, |
|
663
|
|
|
"topk": top_k, |
|
664
|
|
|
"search_param": search_param |
|
665
|
|
|
} |
|
666
|
|
|
logger.info("Query params: %s" % json.dumps(search_param_group)) |
|
667
|
|
|
result_ids, _ = self.do_query_ids(milvus_instance, collection_name, top_k, nq, search_param=search_param) |
|
668
|
|
|
acc_value = self.get_recall_value(true_ids_all[:nq, :top_k].tolist(), result_ids) |
|
669
|
|
|
logger.info("Query accuracy: %s" % acc_value) |
|
670
|
|
|
metric = self.report_wrapper(milvus_instance, self.env_value, self.hostname, collection_info, index_info, search_param_group) |
|
671
|
|
|
metric.metrics = { |
|
672
|
|
|
"type": "accuracy", |
|
673
|
|
|
"value": { |
|
674
|
|
|
"acc": acc_value |
|
675
|
|
|
} |
|
676
|
|
|
} |
|
677
|
|
|
report(metric) |
|
678
|
|
|
|
|
679
|
|
|
elif run_type == "ann_accuracy": |
|
680
|
|
|
hdf5_source_file = collection["source_file"] |
|
681
|
|
|
collection_name = collection["collection_name"] |
|
682
|
|
|
index_file_sizes = collection["index_file_sizes"] |
|
683
|
|
|
index_types = collection["index_types"] |
|
684
|
|
|
index_params = collection["index_params"] |
|
685
|
|
|
top_ks = collection["top_ks"] |
|
686
|
|
|
nqs = collection["nqs"] |
|
687
|
|
|
search_params = collection["search_params"] |
|
688
|
|
|
# mapping to search param list |
|
689
|
|
|
search_params = self.generate_combinations(search_params) |
|
690
|
|
|
# mapping to index param list |
|
691
|
|
|
index_params = self.generate_combinations(index_params) |
|
692
|
|
|
|
|
693
|
|
|
data_type, dimension, metric_type = parser.parse_ann_collection_name(collection_name) |
|
694
|
|
|
dataset = utils.get_dataset(hdf5_source_file) |
|
695
|
|
|
true_ids = np.array(dataset["neighbors"]) |
|
696
|
|
|
for index_file_size in index_file_sizes: |
|
697
|
|
|
collection_info = { |
|
698
|
|
|
"dimension": dimension, |
|
699
|
|
|
"metric_type": metric_type, |
|
700
|
|
|
"index_file_size": index_file_size, |
|
701
|
|
|
"dataset_name": collection_name |
|
702
|
|
|
} |
|
703
|
|
|
if milvus_instance.exists_collection(collection_name): |
|
704
|
|
|
logger.info("Re-create collection: %s" % collection_name) |
|
705
|
|
|
milvus_instance.drop() |
|
706
|
|
|
time.sleep(DELETE_INTERVAL_TIME) |
|
707
|
|
|
|
|
708
|
|
|
milvus_instance.create_collection(collection_name, dimension, index_file_size, metric_type) |
|
709
|
|
|
logger.info(milvus_instance.describe()) |
|
710
|
|
|
insert_vectors = self.normalize(metric_type, np.array(dataset["train"])) |
|
711
|
|
|
# Insert batch once |
|
712
|
|
|
# milvus_instance.insert(insert_vectors) |
|
713
|
|
|
loops = len(insert_vectors) // INSERT_INTERVAL + 1 |
|
714
|
|
|
for i in range(loops): |
|
715
|
|
|
start = i*INSERT_INTERVAL |
|
716
|
|
|
end = min((i+1)*INSERT_INTERVAL, len(insert_vectors)) |
|
717
|
|
|
tmp_vectors = insert_vectors[start:end] |
|
718
|
|
|
if start < end: |
|
719
|
|
|
if not isinstance(tmp_vectors, list): |
|
720
|
|
|
milvus_instance.insert(tmp_vectors.tolist(), ids=[i for i in range(start, end)]) |
|
721
|
|
|
else: |
|
722
|
|
|
milvus_instance.insert(tmp_vectors, ids=[i for i in range(start, end)]) |
|
723
|
|
|
milvus_instance.flush() |
|
724
|
|
|
logger.info("Table: %s, row count: %s" % (collection_name, milvus_instance.count())) |
|
725
|
|
|
if milvus_instance.count() != len(insert_vectors): |
|
726
|
|
|
logger.error("Table row count is not equal to insert vectors") |
|
727
|
|
|
return |
|
728
|
|
|
for index_type in index_types: |
|
729
|
|
|
for index_param in index_params: |
|
730
|
|
|
logger.debug("Building index with param: %s" % json.dumps(index_param)) |
|
731
|
|
|
milvus_instance.create_index(index_type, index_param=index_param) |
|
732
|
|
|
logger.info(milvus_instance.describe_index()) |
|
733
|
|
|
logger.info("Start preload collection: %s" % collection_name) |
|
734
|
|
|
milvus_instance.preload_collection() |
|
735
|
|
|
index_info = { |
|
736
|
|
|
"index_type": index_type, |
|
737
|
|
|
"index_param": index_param |
|
738
|
|
|
} |
|
739
|
|
|
logger.debug(index_info) |
|
740
|
|
|
for search_param in search_params: |
|
741
|
|
|
for nq in nqs: |
|
742
|
|
|
query_vectors = self.normalize(metric_type, np.array(dataset["test"][:nq])) |
|
743
|
|
|
for top_k in top_ks: |
|
744
|
|
|
search_param_group = { |
|
745
|
|
|
"nq": len(query_vectors), |
|
746
|
|
|
"topk": top_k, |
|
747
|
|
|
"search_param": search_param |
|
748
|
|
|
} |
|
749
|
|
|
logger.debug(search_param_group) |
|
750
|
|
|
if not isinstance(query_vectors, list): |
|
751
|
|
|
result = milvus_instance.query(query_vectors.tolist(), top_k, search_param=search_param) |
|
752
|
|
|
else: |
|
753
|
|
|
result = milvus_instance.query(query_vectors, top_k, search_param=search_param) |
|
754
|
|
|
if len(result): |
|
755
|
|
|
logger.debug(len(result)) |
|
756
|
|
|
logger.debug(len(result[0])) |
|
757
|
|
|
result_ids = result.id_array |
|
758
|
|
|
acc_value = self.get_recall_value(true_ids[:nq, :top_k].tolist(), result_ids) |
|
759
|
|
|
logger.info("Query ann_accuracy: %s" % acc_value) |
|
760
|
|
|
metric = self.report_wrapper(milvus_instance, self.env_value, self.hostname, collection_info, index_info, search_param_group) |
|
761
|
|
|
metric.metrics = { |
|
762
|
|
|
"type": "ann_accuracy", |
|
763
|
|
|
"value": { |
|
764
|
|
|
"acc": acc_value |
|
765
|
|
|
} |
|
766
|
|
|
} |
|
767
|
|
|
report(metric) |
|
768
|
|
|
|
|
769
|
|
|
elif run_type == "search_stability": |
|
770
|
|
|
(data_type, collection_size, index_file_size, dimension, metric_type) = parser.collection_parser(collection_name) |
|
771
|
|
|
search_params = collection["search_params"] |
|
772
|
|
|
during_time = collection["during_time"] |
|
773
|
|
|
collection_info = { |
|
774
|
|
|
"dimension": dimension, |
|
775
|
|
|
"metric_type": metric_type, |
|
776
|
|
|
"dataset_name": collection_name |
|
777
|
|
|
} |
|
778
|
|
|
if not milvus_instance.exists_collection(): |
|
779
|
|
|
logger.error("Table name: %s not existed" % collection_name) |
|
780
|
|
|
return |
|
781
|
|
|
logger.info(milvus_instance.count()) |
|
782
|
|
|
index_info = milvus_instance.describe_index() |
|
783
|
|
|
logger.info(index_info) |
|
784
|
|
|
g_top_k = int(collection["top_ks"].split("-")[1]) |
|
785
|
|
|
g_nq = int(collection["nqs"].split("-")[1]) |
|
786
|
|
|
l_top_k = int(collection["top_ks"].split("-")[0]) |
|
787
|
|
|
l_nq = int(collection["nqs"].split("-")[0]) |
|
788
|
|
|
milvus_instance.preload_collection() |
|
789
|
|
|
start_mem_usage = milvus_instance.get_mem_info()["memory_used"] |
|
790
|
|
|
logger.debug(start_mem_usage) |
|
791
|
|
|
start_row_count = milvus_instance.count() |
|
792
|
|
|
logger.debug(milvus_instance.describe_index()) |
|
793
|
|
|
logger.info(start_row_count) |
|
794
|
|
|
start_time = time.time() |
|
795
|
|
|
while time.time() < start_time + during_time * 60: |
|
796
|
|
|
search_param = {} |
|
797
|
|
|
top_k = random.randint(l_top_k, g_top_k) |
|
798
|
|
|
nq = random.randint(l_nq, g_nq) |
|
799
|
|
|
for k, v in search_params.items(): |
|
800
|
|
|
search_param[k] = random.randint(int(v.split("-")[0]), int(v.split("-")[1])) |
|
801
|
|
|
query_vectors = [[random.random() for _ in range(dimension)] for _ in range(nq)] |
|
802
|
|
|
logger.debug("Query nq: %d, top-k: %d, param: %s" % (nq, top_k, json.dumps(search_param))) |
|
803
|
|
|
result = milvus_instance.query(query_vectors, top_k, search_param=search_param) |
|
804
|
|
|
end_mem_usage = milvus_instance.get_mem_info()["memory_used"] |
|
805
|
|
|
metric = self.report_wrapper(milvus_instance, self.env_value, self.hostname, collection_info, index_info, {}) |
|
806
|
|
|
metric.metrics = { |
|
807
|
|
|
"type": "search_stability", |
|
808
|
|
|
"value": { |
|
809
|
|
|
"during_time": during_time, |
|
810
|
|
|
"start_mem_usage": start_mem_usage, |
|
811
|
|
|
"end_mem_usage": end_mem_usage, |
|
812
|
|
|
"diff_mem": end_mem_usage - start_mem_usage |
|
813
|
|
|
} |
|
814
|
|
|
} |
|
815
|
|
|
report(metric) |
|
816
|
|
|
|
|
817
|
|
|
elif run_type == "loop_stability": |
|
818
|
|
|
# init data |
|
819
|
|
|
milvus_instance.clean_db() |
|
820
|
|
|
pull_interval = collection["pull_interval"] |
|
821
|
|
|
collection_num = collection["collection_num"] |
|
822
|
|
|
concurrent = collection["concurrent"] if "concurrent" in collection else False |
|
823
|
|
|
concurrent_num = collection_num |
|
824
|
|
|
dimension = collection["dimension"] if "dimension" in collection else 128 |
|
825
|
|
|
insert_xb = collection["insert_xb"] if "insert_xb" in collection else 100000 |
|
826
|
|
|
index_types = collection["index_types"] if "index_types" in collection else ['ivf_sq8'] |
|
827
|
|
|
index_param = {"nlist": 2048} |
|
828
|
|
|
collection_names = [] |
|
829
|
|
|
milvus_instances_map = {} |
|
830
|
|
|
insert_vectors = [[random.random() for _ in range(dimension)] for _ in range(insert_xb)] |
|
831
|
|
|
for i in range(collection_num): |
|
832
|
|
|
name = utils.get_unique_name(prefix="collection_") |
|
833
|
|
|
collection_names.append(name) |
|
834
|
|
|
metric_type = random.choice(["l2", "ip"]) |
|
835
|
|
|
index_file_size = random.randint(10, 20) |
|
836
|
|
|
milvus_instance.create_collection(name, dimension, index_file_size, metric_type) |
|
837
|
|
|
milvus_instance = MilvusClient(collection_name=name, host=self.host) |
|
838
|
|
|
index_type = random.choice(index_types) |
|
839
|
|
|
milvus_instance.create_index(index_type, index_param=index_param) |
|
840
|
|
|
logger.info(milvus_instance.describe_index()) |
|
841
|
|
|
insert_vectors = utils.normalize(metric_type, insert_vectors) |
|
842
|
|
|
milvus_instance.insert(insert_vectors) |
|
843
|
|
|
milvus_instance.flush() |
|
844
|
|
|
milvus_instances_map.update({name: milvus_instance}) |
|
845
|
|
|
logger.info(milvus_instance.describe_index()) |
|
846
|
|
|
logger.info(milvus_instance.describe()) |
|
847
|
|
|
|
|
848
|
|
|
# loop time unit: min -> s |
|
849
|
|
|
pull_interval_seconds = pull_interval * 60 |
|
850
|
|
|
tasks = ["insert_rand", "delete_rand", "query_rand", "flush", "compact"] |
|
851
|
|
|
i = 1 |
|
852
|
|
|
while True: |
|
853
|
|
|
logger.info("Loop time: %d" % i) |
|
854
|
|
|
start_time = time.time() |
|
855
|
|
|
while time.time() - start_time < pull_interval_seconds: |
|
856
|
|
|
if concurrent: |
|
857
|
|
|
mp = [] |
|
858
|
|
|
for _ in range(concurrent_num): |
|
859
|
|
|
tmp_collection_name = random.choice(collection_names) |
|
860
|
|
|
task_name = random.choice(tasks) |
|
861
|
|
|
mp.append((tmp_collection_name, task_name)) |
|
862
|
|
|
|
|
863
|
|
|
with futures.ThreadPoolExecutor(max_workers=concurrent_num) as executor: |
|
864
|
|
|
future_results = {executor.submit(getattr(milvus_instances_map[mp[j][0]], mp[j][1])): j for j in range(concurrent_num)} |
|
865
|
|
|
for future in futures.as_completed(future_results): |
|
866
|
|
|
future.result() |
|
867
|
|
|
|
|
868
|
|
|
else: |
|
869
|
|
|
tmp_collection_name = random.choice(collection_names) |
|
870
|
|
|
task_name = random.choice(tasks) |
|
871
|
|
|
logger.info(tmp_collection_name) |
|
872
|
|
|
logger.info(task_name) |
|
873
|
|
|
task_run = getattr(milvus_instances_map[tmp_collection_name], task_name) |
|
874
|
|
|
task_run() |
|
875
|
|
|
|
|
876
|
|
|
logger.debug("Restart server") |
|
877
|
|
|
utils.restart_server(self.service_name, namespace) |
|
878
|
|
|
# new connection |
|
879
|
|
|
for name in collection_names: |
|
880
|
|
|
milvus_instance = MilvusClient(collection_name=name, host=self.host) |
|
881
|
|
|
milvus_instances_map.update({name: milvus_instance}) |
|
882
|
|
|
i = i + 1 |
|
883
|
|
|
|
|
884
|
|
|
elif run_type == "stability": |
|
885
|
|
|
(data_type, collection_size, index_file_size, dimension, metric_type) = parser.collection_parser(collection_name) |
|
886
|
|
|
search_params = collection["search_params"] |
|
887
|
|
|
insert_xb = collection["insert_xb"] |
|
888
|
|
|
insert_interval = collection["insert_interval"] |
|
889
|
|
|
delete_xb = collection["delete_xb"] |
|
890
|
|
|
during_time = collection["during_time"] |
|
891
|
|
|
collection_info = { |
|
892
|
|
|
"dimension": dimension, |
|
893
|
|
|
"metric_type": metric_type, |
|
894
|
|
|
"dataset_name": collection_name |
|
895
|
|
|
} |
|
896
|
|
|
if not milvus_instance.exists_collection(): |
|
897
|
|
|
logger.error("Table name: %s not existed" % collection_name) |
|
898
|
|
|
return |
|
899
|
|
|
logger.info(milvus_instance.count()) |
|
900
|
|
|
index_info = milvus_instance.describe_index() |
|
901
|
|
|
logger.info(index_info) |
|
902
|
|
|
g_top_k = int(collection["top_ks"].split("-")[1]) |
|
903
|
|
|
g_nq = int(collection["nqs"].split("-")[1]) |
|
904
|
|
|
l_top_k = int(collection["top_ks"].split("-")[0]) |
|
905
|
|
|
l_nq = int(collection["nqs"].split("-")[0]) |
|
906
|
|
|
milvus_instance.preload_collection() |
|
907
|
|
|
start_mem_usage = milvus_instance.get_mem_info()["memory_used"] |
|
908
|
|
|
start_row_count = milvus_instance.count() |
|
909
|
|
|
logger.debug(milvus_instance.describe_index()) |
|
910
|
|
|
logger.info(start_row_count) |
|
911
|
|
|
start_time = time.time() |
|
912
|
|
|
i = 0 |
|
913
|
|
|
ids = [] |
|
914
|
|
|
insert_vectors = [[random.random() for _ in range(dimension)] for _ in range(insert_xb)] |
|
915
|
|
|
query_vectors = [[random.random() for _ in range(dimension)] for _ in range(10000)] |
|
916
|
|
View Code Duplication |
while time.time() < start_time + during_time * 60: |
|
|
|
|
|
|
917
|
|
|
i = i + 1 |
|
918
|
|
|
for j in range(insert_interval): |
|
919
|
|
|
top_k = random.randint(l_top_k, g_top_k) |
|
920
|
|
|
nq = random.randint(l_nq, g_nq) |
|
921
|
|
|
search_param = {} |
|
922
|
|
|
for k, v in search_params.items(): |
|
923
|
|
|
search_param[k] = random.randint(int(v.split("-")[0]), int(v.split("-")[1])) |
|
924
|
|
|
logger.debug("Query nq: %d, top-k: %d, param: %s" % (nq, top_k, json.dumps(search_param))) |
|
925
|
|
|
result = milvus_instance.query(query_vectors[0:nq], top_k, search_param=search_param) |
|
926
|
|
|
count = milvus_instance.count() |
|
927
|
|
|
insert_ids = [(count+x) for x in range(len(insert_vectors))] |
|
928
|
|
|
ids.extend(insert_ids) |
|
929
|
|
|
status, res = milvus_instance.insert(insert_vectors, ids=insert_ids) |
|
930
|
|
|
logger.debug("%d, row_count: %d" % (i, milvus_instance.count())) |
|
931
|
|
|
milvus_instance.delete(ids[-delete_xb:]) |
|
932
|
|
|
milvus_instance.flush() |
|
933
|
|
|
milvus_instance.compact() |
|
934
|
|
|
end_mem_usage = milvus_instance.get_mem_info()["memory_used"] |
|
935
|
|
|
end_row_count = milvus_instance.count() |
|
936
|
|
|
metric = self.report_wrapper(milvus_instance, self.env_value, self.hostname, collection_info, index_info, {}) |
|
937
|
|
|
metric.metrics = { |
|
938
|
|
|
"type": "stability", |
|
939
|
|
|
"value": { |
|
940
|
|
|
"during_time": during_time, |
|
941
|
|
|
"start_mem_usage": start_mem_usage, |
|
942
|
|
|
"end_mem_usage": end_mem_usage, |
|
943
|
|
|
"diff_mem": end_mem_usage - start_mem_usage, |
|
944
|
|
|
"row_count_increments": end_row_count - start_row_count |
|
945
|
|
|
} |
|
946
|
|
|
} |
|
947
|
|
|
report(metric) |
|
948
|
|
|
|
|
949
|
|
|
elif run_type == "locust_mix_performance": |
|
950
|
|
|
(data_type, collection_size, index_file_size, dimension, metric_type) = parser.collection_parser( |
|
951
|
|
|
collection_name) |
|
952
|
|
|
ni_per = collection["ni_per"] |
|
953
|
|
|
build_index = collection["build_index"] |
|
954
|
|
|
# # TODO: debug |
|
955
|
|
|
if milvus_instance.exists_collection(): |
|
956
|
|
|
milvus_instance.drop() |
|
957
|
|
|
time.sleep(10) |
|
958
|
|
|
index_info = {} |
|
959
|
|
|
search_params = {} |
|
960
|
|
|
milvus_instance.create_collection(collection_name, dimension, index_file_size, metric_type) |
|
961
|
|
|
if build_index is True: |
|
962
|
|
|
index_type = collection["index_type"] |
|
963
|
|
|
index_param = collection["index_param"] |
|
964
|
|
|
index_info = { |
|
965
|
|
|
"index_tyoe": index_type, |
|
966
|
|
|
"index_param": index_param |
|
967
|
|
|
} |
|
968
|
|
|
milvus_instance.create_index(index_type, index_param) |
|
969
|
|
|
logger.debug(milvus_instance.describe_index()) |
|
970
|
|
|
res = self.do_insert(milvus_instance, collection_name, data_type, dimension, collection_size, ni_per) |
|
971
|
|
|
logger.info(res) |
|
972
|
|
|
if "flush" in collection and collection["flush"] == "no": |
|
973
|
|
|
logger.debug("No manual flush") |
|
974
|
|
|
else: |
|
975
|
|
|
milvus_instance.flush() |
|
976
|
|
|
if build_index is True: |
|
977
|
|
|
logger.debug("Start build index for last file") |
|
978
|
|
|
milvus_instance.create_index(index_type, index_param) |
|
979
|
|
|
logger.debug(milvus_instance.describe_index()) |
|
980
|
|
|
### spawn locust requests |
|
981
|
|
|
task = collection["tasks"] |
|
982
|
|
|
# generate task code |
|
983
|
|
|
task_file = utils.get_unique_name() |
|
984
|
|
|
task_file_script = task_file + '.py' |
|
985
|
|
|
task_file_csv = task_file + '_stats.csv' |
|
986
|
|
|
task_types = task["types"] |
|
987
|
|
|
connection_type = "single" |
|
988
|
|
|
connection_num = task["connection_num"] |
|
989
|
|
|
if connection_num > 1: |
|
990
|
|
|
connection_type = "multi" |
|
991
|
|
|
clients_num = task["clients_num"] |
|
992
|
|
|
hatch_rate = task["hatch_rate"] |
|
993
|
|
|
during_time = task["during_time"] |
|
994
|
|
|
def_strs = "" |
|
995
|
|
|
for task_type in task_types: |
|
996
|
|
|
_type = task_type["type"] |
|
997
|
|
|
weight = task_type["weight"] |
|
998
|
|
|
if _type == "flush": |
|
999
|
|
|
def_str = """ |
|
1000
|
|
|
@task(%d) |
|
1001
|
|
|
def flush(self): |
|
1002
|
|
|
client = get_client(collection_name) |
|
1003
|
|
|
client.flush(collection_name=collection_name) |
|
1004
|
|
|
""" % weight |
|
1005
|
|
|
if _type == "compact": |
|
1006
|
|
|
def_str = """ |
|
1007
|
|
|
@task(%d) |
|
1008
|
|
|
def compact(self): |
|
1009
|
|
|
client = get_client(collection_name) |
|
1010
|
|
|
client.compact(collection_name) |
|
1011
|
|
|
""" % weight |
|
1012
|
|
|
if _type == "query": |
|
1013
|
|
|
def_str = """ |
|
1014
|
|
|
@task(%d) |
|
1015
|
|
|
def query(self): |
|
1016
|
|
|
client = get_client(collection_name) |
|
1017
|
|
|
params = %s |
|
1018
|
|
|
X = [[random.random() for i in range(dim)] for i in range(params["nq"])] |
|
1019
|
|
|
client.query(X, params["top_k"], params["search_param"], collection_name=collection_name) |
|
1020
|
|
|
""" % (weight, task_type["params"]) |
|
1021
|
|
|
if _type == "insert": |
|
1022
|
|
|
def_str = """ |
|
1023
|
|
|
@task(%d) |
|
1024
|
|
|
def insert(self): |
|
1025
|
|
|
client = get_client(collection_name) |
|
1026
|
|
|
params = %s |
|
1027
|
|
|
ids = [random.randint(10, 1000000) for i in range(params["nb"])] |
|
1028
|
|
|
X = [[random.random() for i in range(dim)] for i in range(params["nb"])] |
|
1029
|
|
|
client.insert(X,ids=ids, collection_name=collection_name) |
|
1030
|
|
|
""" % (weight, task_type["params"]) |
|
1031
|
|
|
if _type == "delete": |
|
1032
|
|
|
def_str = """ |
|
1033
|
|
|
@task(%d) |
|
1034
|
|
|
def delete(self): |
|
1035
|
|
|
client = get_client(collection_name) |
|
1036
|
|
|
ids = [random.randint(1, 1000000) for i in range(1)] |
|
1037
|
|
|
client.delete(ids, collection_name) |
|
1038
|
|
|
""" % weight |
|
1039
|
|
|
def_strs += def_str |
|
|
|
|
|
|
1040
|
|
|
code_str = """ |
|
1041
|
|
|
import random |
|
1042
|
|
|
import json |
|
1043
|
|
|
from locust import User, task, between |
|
1044
|
|
|
from locust_task import MilvusTask |
|
1045
|
|
|
from client import MilvusClient |
|
1046
|
|
|
|
|
1047
|
|
|
host = '%s' |
|
1048
|
|
|
port = %s |
|
1049
|
|
|
collection_name = '%s' |
|
1050
|
|
|
dim = %s |
|
1051
|
|
|
connection_type = '%s' |
|
1052
|
|
|
m = MilvusClient(host=host, port=port) |
|
1053
|
|
|
|
|
1054
|
|
|
def get_client(collection_name): |
|
1055
|
|
|
if connection_type == 'single': |
|
1056
|
|
|
return MilvusTask(m=m) |
|
1057
|
|
|
elif connection_type == 'multi': |
|
1058
|
|
|
return MilvusTask(connection_type='multi', host=host, port=port, collection_name=collection_name) |
|
1059
|
|
|
|
|
1060
|
|
|
|
|
1061
|
|
|
class MixTask(User): |
|
1062
|
|
|
wait_time = between(0.001, 0.002) |
|
1063
|
|
|
%s |
|
1064
|
|
|
""" % (self.host, self.port, collection_name, dimension, connection_type, def_strs) |
|
1065
|
|
|
print(def_strs) |
|
1066
|
|
|
with open(task_file_script, "w+") as fd: |
|
1067
|
|
|
fd.write(code_str) |
|
|
|
|
|
|
1068
|
|
|
locust_cmd = "locust -f %s --headless --csv=%s -u %d -r %d -t %s" % ( |
|
1069
|
|
|
task_file_script, |
|
1070
|
|
|
task_file, |
|
1071
|
|
|
clients_num, |
|
1072
|
|
|
hatch_rate, |
|
1073
|
|
|
during_time) |
|
1074
|
|
|
logger.info(locust_cmd) |
|
1075
|
|
|
try: |
|
1076
|
|
|
res = os.system(locust_cmd) |
|
1077
|
|
|
except Exception as e: |
|
1078
|
|
|
logger.error(str(e)) |
|
1079
|
|
|
return |
|
1080
|
|
|
# . retrieve and collect test statistics |
|
1081
|
|
|
locust_stats = None |
|
1082
|
|
|
with open(task_file_csv, newline='') as fd: |
|
1083
|
|
|
dr = csv.DictReader(fd) |
|
1084
|
|
|
for row in dr: |
|
1085
|
|
|
if row["Name"] != "Aggregated": |
|
1086
|
|
|
continue |
|
1087
|
|
|
locust_stats = row |
|
1088
|
|
|
logger.info(locust_stats) |
|
1089
|
|
|
collection_info = { |
|
1090
|
|
|
"dimension": dimension, |
|
1091
|
|
|
"metric_type": metric_type, |
|
1092
|
|
|
"dataset_name": collection_name |
|
1093
|
|
|
} |
|
1094
|
|
|
metric = self.report_wrapper(milvus_instance, self.env_value, self.hostname, collection_info, index_info, search_params) |
|
1095
|
|
|
metric.metrics = { |
|
1096
|
|
|
"type": run_type, |
|
1097
|
|
|
"value": { |
|
1098
|
|
|
"during_time": during_time, |
|
1099
|
|
|
"request_count": int(locust_stats["Request Count"]), |
|
1100
|
|
|
"failure_count": int(locust_stats["Failure Count"]), |
|
1101
|
|
|
"qps": locust_stats["Requests/s"], |
|
1102
|
|
|
"min_response_time": int(locust_stats["Min Response Time"]), |
|
1103
|
|
|
"max_response_time": int(locust_stats["Max Response Time"]), |
|
1104
|
|
|
"median_response_time": int(locust_stats["Median Response Time"]), |
|
1105
|
|
|
"avg_response_time": int(locust_stats["Average Response Time"]) |
|
1106
|
|
|
} |
|
1107
|
|
|
} |
|
1108
|
|
|
report(metric) |
|
1109
|
|
|
|
|
1110
|
|
|
else: |
|
1111
|
|
|
logger.warning("Run type: %s not defined" % run_type) |
|
1112
|
|
|
return |
|
1113
|
|
|
logger.debug("Test finished") |
|
1114
|
|
|
|
milvus-io /
milvus
Loading history...
