|
1
|
|
|
import pdb |
|
2
|
|
|
import struct |
|
3
|
|
|
from random import sample |
|
4
|
|
|
import threading |
|
5
|
|
|
import datetime |
|
6
|
|
|
import logging |
|
7
|
|
|
from time import sleep |
|
8
|
|
|
import concurrent.futures |
|
9
|
|
|
from multiprocessing import Process |
|
10
|
|
|
import pytest |
|
11
|
|
|
import numpy |
|
12
|
|
|
import sklearn.preprocessing |
|
13
|
|
|
from milvus import IndexType, MetricType |
|
14
|
|
|
from utils import * |
|
15
|
|
|
|
|
16
|
|
|
dim = 128 |
|
17
|
|
|
collection_id = "test_search" |
|
18
|
|
|
add_interval_time = 2 |
|
19
|
|
|
vectors = gen_vectors(6000, dim) |
|
20
|
|
|
vectors = sklearn.preprocessing.normalize(vectors, axis=1, norm='l2') |
|
21
|
|
|
vectors = vectors.tolist() |
|
22
|
|
|
top_k = 1 |
|
23
|
|
|
nprobe = 1 |
|
24
|
|
|
epsilon = 0.001 |
|
25
|
|
|
tag = "1970-01-01" |
|
26
|
|
|
raw_vectors, binary_vectors = gen_binary_vectors(6000, dim) |
|
27
|
|
|
|
|
28
|
|
|
|
|
29
|
|
|
class TestSearchBase: |
|
30
|
|
|
def init_data(self, connect, collection, nb=6000, dim=dim, partition_tags=None): |
|
31
|
|
|
''' |
|
32
|
|
|
Generate vectors and add it in collection, before search vectors |
|
33
|
|
|
''' |
|
34
|
|
|
global vectors |
|
35
|
|
|
if nb == 6000: |
|
36
|
|
|
add_vectors = vectors |
|
37
|
|
|
else: |
|
38
|
|
|
add_vectors = gen_vectors(nb, dim) |
|
39
|
|
|
add_vectors = sklearn.preprocessing.normalize(add_vectors, axis=1, norm='l2') |
|
40
|
|
|
add_vectors = add_vectors.tolist() |
|
41
|
|
|
if partition_tags is None: |
|
42
|
|
|
status, ids = connect.insert(collection, add_vectors) |
|
43
|
|
|
assert status.OK() |
|
44
|
|
|
else: |
|
45
|
|
|
status, ids = connect.insert(collection, add_vectors, partition_tag=partition_tags) |
|
46
|
|
|
assert status.OK() |
|
47
|
|
|
connect.flush([collection]) |
|
48
|
|
|
return add_vectors, ids |
|
49
|
|
|
|
|
50
|
|
|
def init_binary_data(self, connect, collection, nb=6000, dim=dim, insert=True, partition_tags=None): |
|
51
|
|
|
''' |
|
52
|
|
|
Generate vectors and add it in collection, before search vectors |
|
53
|
|
|
''' |
|
54
|
|
|
ids = [] |
|
55
|
|
|
global binary_vectors |
|
56
|
|
|
global raw_vectors |
|
57
|
|
|
if nb == 6000: |
|
58
|
|
|
add_vectors = binary_vectors |
|
59
|
|
|
add_raw_vectors = raw_vectors |
|
60
|
|
|
else: |
|
61
|
|
|
add_raw_vectors, add_vectors = gen_binary_vectors(nb, dim) |
|
62
|
|
|
if insert is True: |
|
63
|
|
|
if partition_tags is None: |
|
64
|
|
|
status, ids = connect.insert(collection, add_vectors) |
|
65
|
|
|
assert status.OK() |
|
66
|
|
|
else: |
|
67
|
|
|
status, ids = connect.insert(collection, add_vectors, partition_tag=partition_tags) |
|
68
|
|
|
assert status.OK() |
|
69
|
|
|
connect.flush([collection]) |
|
70
|
|
|
return add_raw_vectors, add_vectors, ids |
|
71
|
|
|
|
|
72
|
|
|
""" |
|
73
|
|
|
generate valid create_index params |
|
74
|
|
|
""" |
|
75
|
|
|
|
|
76
|
|
|
@pytest.fixture( |
|
77
|
|
|
scope="function", |
|
78
|
|
|
params=gen_index() |
|
79
|
|
|
) |
|
80
|
|
|
def get_index(self, request, connect): |
|
81
|
|
|
if str(connect._cmd("mode")[1]) == "CPU": |
|
82
|
|
|
if request.param["index_type"] == IndexType.IVF_SQ8H: |
|
83
|
|
|
pytest.skip("sq8h not support in CPU mode") |
|
84
|
|
|
return request.param |
|
85
|
|
|
|
|
86
|
|
|
@pytest.fixture( |
|
87
|
|
|
scope="function", |
|
88
|
|
|
params=gen_simple_index() |
|
89
|
|
|
) |
|
90
|
|
|
def get_simple_index(self, request, connect): |
|
91
|
|
|
if str(connect._cmd("mode")[1]) == "CPU": |
|
92
|
|
|
if request.param["index_type"] == IndexType.IVF_SQ8H: |
|
93
|
|
|
pytest.skip("sq8h not support in CPU mode") |
|
94
|
|
|
return request.param |
|
95
|
|
|
|
|
96
|
|
View Code Duplication |
@pytest.fixture( |
|
|
|
|
|
|
97
|
|
|
scope="function", |
|
98
|
|
|
params=gen_simple_index() |
|
99
|
|
|
) |
|
100
|
|
|
def get_jaccard_index(self, request, connect): |
|
101
|
|
|
logging.getLogger().info(request.param) |
|
102
|
|
|
if request.param["index_type"] == IndexType.IVFLAT or request.param["index_type"] == IndexType.FLAT: |
|
103
|
|
|
return request.param |
|
104
|
|
|
else: |
|
105
|
|
|
pytest.skip("Skip index Temporary") |
|
106
|
|
|
|
|
107
|
|
View Code Duplication |
@pytest.fixture( |
|
|
|
|
|
|
108
|
|
|
scope="function", |
|
109
|
|
|
params=gen_simple_index() |
|
110
|
|
|
) |
|
111
|
|
|
def get_hamming_index(self, request, connect): |
|
112
|
|
|
logging.getLogger().info(request.param) |
|
113
|
|
|
if request.param["index_type"] == IndexType.IVFLAT or request.param["index_type"] == IndexType.FLAT: |
|
114
|
|
|
return request.param |
|
115
|
|
|
else: |
|
116
|
|
|
pytest.skip("Skip index Temporary") |
|
117
|
|
|
|
|
118
|
|
|
@pytest.fixture( |
|
119
|
|
|
scope="function", |
|
120
|
|
|
params=gen_simple_index() |
|
121
|
|
|
) |
|
122
|
|
|
def get_structure_index(self, request, connect): |
|
123
|
|
|
logging.getLogger().info(request.param) |
|
124
|
|
|
if request.param["index_type"] == IndexType.FLAT: |
|
125
|
|
|
return request.param |
|
126
|
|
|
else: |
|
127
|
|
|
pytest.skip("Skip index Temporary") |
|
128
|
|
|
|
|
129
|
|
|
""" |
|
130
|
|
|
generate top-k params |
|
131
|
|
|
""" |
|
132
|
|
|
|
|
133
|
|
|
@pytest.fixture( |
|
134
|
|
|
scope="function", |
|
135
|
|
|
params=[1, 99, 1024, 2049, 16385] |
|
136
|
|
|
) |
|
137
|
|
|
def get_top_k(self, request): |
|
138
|
|
|
yield request.param |
|
139
|
|
|
|
|
140
|
|
|
def test_search_top_k_flat_index(self, connect, collection, get_top_k): |
|
141
|
|
|
''' |
|
142
|
|
|
target: test basic search fuction, all the search params is corrent, change top-k value |
|
143
|
|
|
method: search with the given vectors, check the result |
|
144
|
|
|
expected: search status ok, and the length of the result is top_k |
|
145
|
|
|
''' |
|
146
|
|
|
vectors, ids = self.init_data(connect, collection) |
|
147
|
|
|
query_vec = [vectors[0]] |
|
148
|
|
|
top_k = get_top_k |
|
149
|
|
|
status, result = connect.search(collection, top_k, query_vec) |
|
150
|
|
|
if top_k <= 16384: |
|
151
|
|
|
assert status.OK() |
|
152
|
|
|
assert len(result[0]) == min(len(vectors), top_k) |
|
153
|
|
|
assert result[0][0].distance <= epsilon |
|
154
|
|
|
assert check_result(result[0], ids[0]) |
|
155
|
|
|
else: |
|
156
|
|
|
assert not status.OK() |
|
157
|
|
|
|
|
158
|
|
|
def test_search_top_k_flat_index_metric_type(self, connect, collection): |
|
159
|
|
|
''' |
|
160
|
|
|
target: test basic search fuction, all the search params is corrent, change top-k value |
|
161
|
|
|
method: search with the given vectors, check the result |
|
162
|
|
|
expected: search status ok, and the length of the result is top_k |
|
163
|
|
|
''' |
|
164
|
|
|
vectors, ids = self.init_data(connect, collection) |
|
165
|
|
|
query_vec = [vectors[0]] |
|
166
|
|
|
status, result = connect.search(collection, top_k, query_vec, params={"metric_type": MetricType.IP.value}) |
|
167
|
|
|
assert status.OK() |
|
168
|
|
|
assert len(result[0]) == min(len(vectors), top_k) |
|
169
|
|
|
assert result[0][0].distance >= 1 - epsilon |
|
170
|
|
|
assert check_result(result[0], ids[0]) |
|
171
|
|
|
|
|
172
|
|
|
@pytest.mark.level(2) |
|
173
|
|
|
def test_search_top_k_flat_index_metric_type_invalid(self, connect, collection): |
|
174
|
|
|
''' |
|
175
|
|
|
target: test basic search fuction, all the search params is corrent, change top-k value |
|
176
|
|
|
method: search with the given vectors, check the result |
|
177
|
|
|
expected: search status ok, and the length of the result is top_k |
|
178
|
|
|
''' |
|
179
|
|
|
vectors, ids = self.init_data(connect, collection) |
|
180
|
|
|
query_vec = [vectors[0]] |
|
181
|
|
|
status, result = connect.search(collection, top_k, query_vec, params={"metric_type": MetricType.JACCARD.value}) |
|
182
|
|
|
assert not status.OK() |
|
183
|
|
|
|
|
184
|
|
|
def test_search_l2_index_params(self, connect, collection, get_simple_index): |
|
185
|
|
|
''' |
|
186
|
|
|
target: test basic search fuction, all the search params is corrent, test all index params, and build |
|
187
|
|
|
method: search with the given vectors, check the result |
|
188
|
|
|
expected: search status ok, and the length of the result is top_k |
|
189
|
|
|
''' |
|
190
|
|
|
top_k = 10 |
|
191
|
|
|
index_param = get_simple_index["index_param"] |
|
192
|
|
|
index_type = get_simple_index["index_type"] |
|
193
|
|
|
logging.getLogger().info(get_simple_index) |
|
194
|
|
|
vectors, ids = self.init_data(connect, collection) |
|
195
|
|
|
status = connect.create_index(collection, index_type, index_param) |
|
196
|
|
|
query_vec = [vectors[0], vectors[1]] |
|
197
|
|
|
search_param = get_search_param(index_type) |
|
198
|
|
|
status, result = connect.search(collection, top_k, query_vec, params=search_param) |
|
199
|
|
|
logging.getLogger().info(result) |
|
200
|
|
|
if top_k <= 1024: |
|
201
|
|
|
assert status.OK() |
|
202
|
|
|
assert len(result[0]) == min(len(vectors), top_k) |
|
203
|
|
|
assert check_result(result[0], ids[0]) |
|
204
|
|
|
assert result[0][0].distance < result[0][1].distance |
|
205
|
|
|
assert result[1][0].distance < result[1][1].distance |
|
206
|
|
|
else: |
|
207
|
|
|
assert not status.OK() |
|
208
|
|
|
|
|
209
|
|
|
def test_search_l2_large_nq_index_params(self, connect, collection, get_simple_index): |
|
210
|
|
|
''' |
|
211
|
|
|
target: test basic search fuction, all the search params is corrent, test all index params, and build |
|
212
|
|
|
method: search with the given vectors, check the result |
|
213
|
|
|
expected: search status ok, and the length of the result is top_k |
|
214
|
|
|
''' |
|
215
|
|
|
top_k = 10 |
|
216
|
|
|
index_param = get_simple_index["index_param"] |
|
217
|
|
|
index_type = get_simple_index["index_type"] |
|
218
|
|
|
logging.getLogger().info(get_simple_index) |
|
219
|
|
|
if index_type == IndexType.IVF_PQ: |
|
220
|
|
|
pytest.skip("Skip PQ") |
|
221
|
|
|
|
|
222
|
|
|
vectors, ids = self.init_data(connect, collection) |
|
223
|
|
|
status = connect.create_index(collection, index_type, index_param) |
|
224
|
|
|
query_vec = vectors[:1000] |
|
225
|
|
|
search_param = get_search_param(index_type) |
|
226
|
|
|
status, result = connect.search(collection, top_k, query_vec, params=search_param) |
|
227
|
|
|
logging.getLogger().info(result) |
|
228
|
|
|
assert status.OK() |
|
229
|
|
|
assert len(result[0]) == min(len(vectors), top_k) |
|
230
|
|
|
assert check_result(result[0], ids[0]) |
|
231
|
|
|
assert result[0][0].distance <= epsilon |
|
232
|
|
|
|
|
233
|
|
|
def test_search_with_multi_partitions(self, connect, collection): |
|
234
|
|
|
''' |
|
235
|
|
|
target: test search with multi partition which contains default tag and other tags |
|
236
|
|
|
method: insert vectors into e partition and search with partitions [_default, tag] |
|
237
|
|
|
expected: search result is correct |
|
238
|
|
|
''' |
|
239
|
|
|
connect.create_partition(collection, tag) |
|
240
|
|
|
vectors, ids = self.init_data(connect, collection, nb=10, partition_tags=tag) |
|
241
|
|
|
query_vec = [vectors[0]] |
|
242
|
|
|
search_param = get_search_param(IndexType.FLAT) |
|
243
|
|
|
status, result = connect.search(collection, top_k, query_vec, partition_tags=["_default", tag], |
|
244
|
|
|
params=search_param) |
|
245
|
|
|
assert status.OK() |
|
246
|
|
|
logging.getLogger().info(result) |
|
247
|
|
|
assert len(result[0]) == min(len(vectors), top_k) |
|
248
|
|
|
assert check_result(result[0], ids[0]) |
|
249
|
|
|
assert result[0][0].distance <= epsilon |
|
250
|
|
|
|
|
251
|
|
|
def test_search_l2_index_params_partition(self, connect, collection, get_simple_index): |
|
252
|
|
|
''' |
|
253
|
|
|
target: test basic search fuction, all the search params is corrent, test all index params, and build |
|
254
|
|
|
method: add vectors into collection, search with the given vectors, check the result |
|
255
|
|
|
expected: search status ok, and the length of the result is top_k, search collection with partition tag return empty |
|
256
|
|
|
''' |
|
257
|
|
|
top_k = 10 |
|
258
|
|
|
index_param = get_simple_index["index_param"] |
|
259
|
|
|
index_type = get_simple_index["index_type"] |
|
260
|
|
|
logging.getLogger().info(get_simple_index) |
|
261
|
|
|
if index_type == IndexType.IVF_PQ: |
|
262
|
|
|
pytest.skip("Skip PQ") |
|
263
|
|
|
status = connect.create_partition(collection, tag) |
|
264
|
|
|
vectors, ids = self.init_data(connect, collection) |
|
265
|
|
|
status = connect.create_index(collection, index_type, index_param) |
|
266
|
|
|
query_vec = [vectors[0]] |
|
267
|
|
|
search_param = get_search_param(index_type) |
|
268
|
|
|
status, result = connect.search(collection, top_k, query_vec, params=search_param) |
|
269
|
|
|
logging.getLogger().info(result) |
|
270
|
|
|
assert status.OK() |
|
271
|
|
|
assert len(result[0]) == min(len(vectors), top_k) |
|
272
|
|
|
assert check_result(result[0], ids[0]) |
|
273
|
|
|
assert result[0][0].distance <= epsilon |
|
274
|
|
|
status, result = connect.search(collection, top_k, query_vec, partition_tags=[tag], params=search_param) |
|
275
|
|
|
logging.getLogger().info(result) |
|
276
|
|
|
assert status.OK() |
|
277
|
|
|
assert len(result) == 0 |
|
278
|
|
|
|
|
279
|
|
|
def test_search_l2_index_params_partition_A(self, connect, collection, get_simple_index): |
|
280
|
|
|
''' |
|
281
|
|
|
target: test basic search fuction, all the search params is corrent, test all index params, and build |
|
282
|
|
|
method: search partition with the given vectors, check the result |
|
283
|
|
|
expected: search status ok, and the length of the result is 0 |
|
284
|
|
|
''' |
|
285
|
|
|
top_k = 10 |
|
286
|
|
|
index_param = get_simple_index["index_param"] |
|
287
|
|
|
index_type = get_simple_index["index_type"] |
|
288
|
|
|
logging.getLogger().info(get_simple_index) |
|
289
|
|
|
if index_type == IndexType.IVF_PQ: |
|
290
|
|
|
pytest.skip("Skip PQ") |
|
291
|
|
|
|
|
292
|
|
|
status = connect.create_partition(collection, tag) |
|
293
|
|
|
vectors, ids = self.init_data(connect, collection) |
|
294
|
|
|
status = connect.create_index(collection, index_type, index_param) |
|
295
|
|
|
query_vec = [vectors[0]] |
|
296
|
|
|
search_param = get_search_param(index_type) |
|
297
|
|
|
status, result = connect.search(collection, top_k, query_vec, partition_tags=[tag], params=search_param) |
|
298
|
|
|
logging.getLogger().info(result) |
|
299
|
|
|
assert status.OK() |
|
300
|
|
|
assert len(result) == 0 |
|
301
|
|
|
|
|
302
|
|
|
def test_search_l2_index_params_partition_B(self, connect, collection, get_simple_index): |
|
303
|
|
|
''' |
|
304
|
|
|
target: test basic search fuction, all the search params is corrent, test all index params, and build |
|
305
|
|
|
method: search with the given vectors, check the result |
|
306
|
|
|
expected: search status ok, and the length of the result is top_k |
|
307
|
|
|
''' |
|
308
|
|
|
top_k = 10 |
|
309
|
|
|
index_param = get_simple_index["index_param"] |
|
310
|
|
|
index_type = get_simple_index["index_type"] |
|
311
|
|
|
logging.getLogger().info(get_simple_index) |
|
312
|
|
|
if index_type == IndexType.IVF_PQ: |
|
313
|
|
|
pytest.skip("Skip PQ") |
|
314
|
|
|
status = connect.create_partition(collection, tag) |
|
315
|
|
|
vectors, ids = self.init_data(connect, collection, partition_tags=tag) |
|
316
|
|
|
status = connect.create_index(collection, index_type, index_param) |
|
317
|
|
|
query_vec = [vectors[0]] |
|
318
|
|
|
search_param = get_search_param(index_type) |
|
319
|
|
|
status, result = connect.search(collection, top_k, query_vec, params=search_param) |
|
320
|
|
|
logging.getLogger().info(result) |
|
321
|
|
|
assert status.OK() |
|
322
|
|
|
assert len(result[0]) == min(len(vectors), top_k) |
|
323
|
|
|
assert check_result(result[0], ids[0]) |
|
324
|
|
|
assert result[0][0].distance <= epsilon |
|
325
|
|
|
status, result = connect.search(collection, top_k, query_vec, partition_tags=[tag], params=search_param) |
|
326
|
|
|
logging.getLogger().info(result) |
|
327
|
|
|
assert status.OK() |
|
328
|
|
|
assert len(result[0]) == min(len(vectors), top_k) |
|
329
|
|
|
assert check_result(result[0], ids[0]) |
|
330
|
|
|
assert result[0][0].distance <= epsilon |
|
331
|
|
|
|
|
332
|
|
|
def test_search_l2_index_params_partition_C(self, connect, collection, get_simple_index): |
|
333
|
|
|
''' |
|
334
|
|
|
target: test basic search fuction, all the search params is corrent, test all index params, and build |
|
335
|
|
|
method: search with the given vectors and tags (one of the tags not existed in collection), check the result |
|
336
|
|
|
expected: search status ok, and the length of the result is top_k |
|
337
|
|
|
''' |
|
338
|
|
|
index_param = get_simple_index["index_param"] |
|
339
|
|
|
index_type = get_simple_index["index_type"] |
|
340
|
|
|
logging.getLogger().info(get_simple_index) |
|
341
|
|
|
if index_type == IndexType.IVF_PQ: |
|
342
|
|
|
pytest.skip("Skip PQ") |
|
343
|
|
|
status = connect.create_partition(collection, tag) |
|
344
|
|
|
vectors, ids = self.init_data(connect, collection, partition_tags=tag) |
|
345
|
|
|
status = connect.create_index(collection, index_type, index_param) |
|
346
|
|
|
query_vec = [vectors[0]] |
|
347
|
|
|
top_k = 10 |
|
348
|
|
|
search_param = get_search_param(index_type) |
|
349
|
|
|
status, result = connect.search(collection, top_k, query_vec, partition_tags=[tag, "new_tag"], |
|
350
|
|
|
params=search_param) |
|
351
|
|
|
logging.getLogger().info(result) |
|
352
|
|
|
assert status.OK() |
|
353
|
|
|
assert len(result[0]) == min(len(vectors), top_k) |
|
354
|
|
|
assert check_result(result[0], ids[0]) |
|
355
|
|
|
assert result[0][0].distance <= epsilon |
|
356
|
|
|
|
|
357
|
|
|
@pytest.mark.level(2) |
|
358
|
|
|
def test_search_l2_index_params_partition_D(self, connect, collection, get_simple_index): |
|
359
|
|
|
''' |
|
360
|
|
|
target: test basic search fuction, all the search params is corrent, test all index params, and build |
|
361
|
|
|
method: search with the given vectors and tag (tag name not existed in collection), check the result |
|
362
|
|
|
expected: search status ok, and the length of the result is top_k |
|
363
|
|
|
''' |
|
364
|
|
|
index_param = get_simple_index["index_param"] |
|
365
|
|
|
index_type = get_simple_index["index_type"] |
|
366
|
|
|
logging.getLogger().info(get_simple_index) |
|
367
|
|
|
status = connect.create_partition(collection, tag) |
|
368
|
|
|
vectors, ids = self.init_data(connect, collection, partition_tags=tag) |
|
369
|
|
|
status = connect.create_index(collection, index_type, index_param) |
|
370
|
|
|
query_vec = [vectors[0]] |
|
371
|
|
|
top_k = 10 |
|
372
|
|
|
search_param = get_search_param(index_type) |
|
373
|
|
|
status, result = connect.search(collection, top_k, query_vec, partition_tags=["new_tag"], params=search_param) |
|
374
|
|
|
logging.getLogger().info(result) |
|
375
|
|
|
assert not status.OK() |
|
376
|
|
|
|
|
377
|
|
|
@pytest.mark.level(2) |
|
378
|
|
|
def test_search_l2_index_params_partition_E(self, connect, collection, get_simple_index): |
|
379
|
|
|
''' |
|
380
|
|
|
target: test basic search fuction, all the search params is corrent, test all index params, and build |
|
381
|
|
|
method: search collection with the given vectors and tags, check the result |
|
382
|
|
|
expected: search status ok, and the length of the result is top_k |
|
383
|
|
|
''' |
|
384
|
|
|
top_k = 10 |
|
385
|
|
|
new_tag = "new_tag" |
|
386
|
|
|
index_type = get_simple_index["index_type"] |
|
387
|
|
|
index_param = get_simple_index["index_param"] |
|
388
|
|
|
if index_type == IndexType.IVF_PQ: |
|
389
|
|
|
pytest.skip("Skip PQ") |
|
390
|
|
|
logging.getLogger().info(get_simple_index) |
|
391
|
|
|
status = connect.create_partition(collection, tag) |
|
392
|
|
|
status = connect.create_partition(collection, new_tag) |
|
393
|
|
|
vectors, ids = self.init_data(connect, collection, partition_tags=tag) |
|
394
|
|
|
new_vectors, new_ids = self.init_data(connect, collection, nb=6001, partition_tags=new_tag) |
|
395
|
|
|
status = connect.create_index(collection, index_type, index_param) |
|
396
|
|
|
query_vec = [vectors[0], new_vectors[0]] |
|
397
|
|
|
search_param = get_search_param(index_type) |
|
398
|
|
|
status, result = connect.search(collection, top_k, query_vec, partition_tags=[tag, new_tag], |
|
399
|
|
|
params=search_param) |
|
400
|
|
|
logging.getLogger().info(result) |
|
401
|
|
|
assert status.OK() |
|
402
|
|
|
assert len(result[0]) == min(len(vectors), top_k) |
|
403
|
|
|
assert check_result(result[0], ids[0]) |
|
404
|
|
|
assert check_result(result[1], new_ids[0]) |
|
405
|
|
|
assert result[0][0].distance <= epsilon |
|
406
|
|
|
assert result[1][0].distance <= epsilon |
|
407
|
|
|
status, result = connect.search(collection, top_k, query_vec, partition_tags=[new_tag], params=search_param) |
|
408
|
|
|
logging.getLogger().info(result) |
|
409
|
|
|
assert status.OK() |
|
410
|
|
|
assert len(result[0]) == min(len(vectors), top_k) |
|
411
|
|
|
assert check_result(result[1], new_ids[0]) |
|
412
|
|
|
assert result[1][0].distance <= epsilon |
|
413
|
|
|
|
|
414
|
|
|
def test_search_l2_index_params_partition_F(self, connect, collection, get_simple_index): |
|
415
|
|
|
''' |
|
416
|
|
|
target: test basic search fuction, all the search params is corrent, test all index params, and build |
|
417
|
|
|
method: search collection with the given vectors and tags with "re" expr, check the result |
|
418
|
|
|
expected: search status ok, and the length of the result is top_k |
|
419
|
|
|
''' |
|
420
|
|
|
tag = "atag" |
|
421
|
|
|
new_tag = "new_tag" |
|
422
|
|
|
index_param = get_simple_index["index_param"] |
|
423
|
|
|
index_type = get_simple_index["index_type"] |
|
424
|
|
|
logging.getLogger().info(get_simple_index) |
|
425
|
|
|
if index_type == IndexType.IVF_PQ: |
|
426
|
|
|
pytest.skip("Skip PQ") |
|
427
|
|
|
status = connect.create_partition(collection, tag) |
|
428
|
|
|
status = connect.create_partition(collection, new_tag) |
|
429
|
|
|
vectors, ids = self.init_data(connect, collection, partition_tags=tag) |
|
430
|
|
|
new_vectors, new_ids = self.init_data(connect, collection, nb=6001, partition_tags=new_tag) |
|
431
|
|
|
status = connect.create_index(collection, index_type, index_param) |
|
432
|
|
|
query_vec = [vectors[0], new_vectors[0]] |
|
433
|
|
|
top_k = 10 |
|
434
|
|
|
search_param = get_search_param(index_type) |
|
435
|
|
|
status, result = connect.search(collection, top_k, query_vec, partition_tags=["new(.*)"], params=search_param) |
|
436
|
|
|
logging.getLogger().info(result) |
|
437
|
|
|
assert status.OK() |
|
438
|
|
|
assert result[0][0].distance > epsilon |
|
439
|
|
|
assert result[1][0].distance <= epsilon |
|
440
|
|
|
status, result = connect.search(collection, top_k, query_vec, partition_tags=["(.*)tag"], params=search_param) |
|
441
|
|
|
logging.getLogger().info(result) |
|
442
|
|
|
assert status.OK() |
|
443
|
|
|
assert result[0][0].distance <= epsilon |
|
444
|
|
|
assert result[1][0].distance <= epsilon |
|
445
|
|
|
|
|
446
|
|
|
@pytest.mark.level(2) |
|
447
|
|
|
def test_search_ip_index_params(self, connect, ip_collection, get_simple_index): |
|
448
|
|
|
''' |
|
449
|
|
|
target: test basic search fuction, all the search params is corrent, test all index params, and build |
|
450
|
|
|
method: search with the given vectors, check the result |
|
451
|
|
|
expected: search status ok, and the length of the result is top_k |
|
452
|
|
|
''' |
|
453
|
|
|
top_k = 10 |
|
454
|
|
|
index_param = get_simple_index["index_param"] |
|
455
|
|
|
index_type = get_simple_index["index_type"] |
|
456
|
|
|
logging.getLogger().info(get_simple_index) |
|
457
|
|
|
vectors, ids = self.init_data(connect, ip_collection) |
|
458
|
|
|
status = connect.create_index(ip_collection, index_type, index_param) |
|
459
|
|
|
query_vec = [vectors[0]] |
|
460
|
|
|
search_param = get_search_param(index_type) |
|
461
|
|
|
status, result = connect.search(ip_collection, top_k, query_vec, params=search_param) |
|
462
|
|
|
logging.getLogger().info(result) |
|
463
|
|
|
assert status.OK() |
|
464
|
|
|
assert len(result[0]) == min(len(vectors), top_k) |
|
465
|
|
|
assert check_result(result[0], ids[0]) |
|
466
|
|
|
assert result[0][0].distance >= result[0][1].distance |
|
467
|
|
|
|
|
468
|
|
|
def test_search_ip_large_nq_index_params(self, connect, ip_collection, get_simple_index): |
|
469
|
|
|
''' |
|
470
|
|
|
target: test basic search fuction, all the search params is corrent, test all index params, and build |
|
471
|
|
|
method: search with the given vectors, check the result |
|
472
|
|
|
expected: search status ok, and the length of the result is top_k |
|
473
|
|
|
''' |
|
474
|
|
|
index_param = get_simple_index["index_param"] |
|
475
|
|
|
index_type = get_simple_index["index_type"] |
|
476
|
|
|
logging.getLogger().info(get_simple_index) |
|
477
|
|
|
if index_type in [IndexType.RNSG, IndexType.IVF_PQ]: |
|
478
|
|
|
pytest.skip("rnsg not support in ip, skip pq") |
|
479
|
|
|
vectors, ids = self.init_data(connect, ip_collection) |
|
480
|
|
|
status = connect.create_index(ip_collection, index_type, index_param) |
|
481
|
|
|
query_vec = [] |
|
482
|
|
|
for i in range(1200): |
|
483
|
|
|
query_vec.append(vectors[i]) |
|
484
|
|
|
top_k = 10 |
|
485
|
|
|
search_param = get_search_param(index_type) |
|
486
|
|
|
status, result = connect.search(ip_collection, top_k, query_vec, params=search_param) |
|
487
|
|
|
logging.getLogger().info(result) |
|
488
|
|
|
assert status.OK() |
|
489
|
|
|
assert len(result[0]) == min(len(vectors), top_k) |
|
490
|
|
|
assert check_result(result[0], ids[0]) |
|
491
|
|
|
assert result[0][0].distance >= 1 - gen_inaccuracy(result[0][0].distance) |
|
492
|
|
|
|
|
493
|
|
|
@pytest.mark.level(2) |
|
494
|
|
|
def test_search_ip_index_params_partition(self, connect, ip_collection, get_simple_index): |
|
495
|
|
|
''' |
|
496
|
|
|
target: test basic search fuction, all the search params is corrent, test all index params, and build |
|
497
|
|
|
method: search with the given vectors, check the result |
|
498
|
|
|
expected: search status ok, and the length of the result is top_k |
|
499
|
|
|
''' |
|
500
|
|
|
top_k = 10 |
|
501
|
|
|
index_param = get_simple_index["index_param"] |
|
502
|
|
|
index_type = get_simple_index["index_type"] |
|
503
|
|
|
logging.getLogger().info(index_param) |
|
504
|
|
|
if index_type in [IndexType.RNSG, IndexType.IVF_PQ]: |
|
505
|
|
|
pytest.skip("rnsg not support in ip, skip pq") |
|
506
|
|
|
|
|
507
|
|
|
status = connect.create_partition(ip_collection, tag) |
|
508
|
|
|
vectors, ids = self.init_data(connect, ip_collection) |
|
509
|
|
|
status = connect.create_index(ip_collection, index_type, index_param) |
|
510
|
|
|
query_vec = [vectors[0]] |
|
511
|
|
|
search_param = get_search_param(index_type) |
|
512
|
|
|
status, result = connect.search(ip_collection, top_k, query_vec, params=search_param) |
|
513
|
|
|
logging.getLogger().info(result) |
|
514
|
|
|
assert status.OK() |
|
515
|
|
|
assert len(result[0]) == min(len(vectors), top_k) |
|
516
|
|
|
assert check_result(result[0], ids[0]) |
|
517
|
|
|
assert result[0][0].distance >= 1 - gen_inaccuracy(result[0][0].distance) |
|
518
|
|
|
status, result = connect.search(ip_collection, top_k, query_vec, partition_tags=[tag], params=search_param) |
|
519
|
|
|
logging.getLogger().info(result) |
|
520
|
|
|
assert status.OK() |
|
521
|
|
|
assert len(result) == 0 |
|
522
|
|
|
|
|
523
|
|
|
@pytest.mark.level(2) |
|
524
|
|
|
def test_search_ip_index_params_partition_A(self, connect, ip_collection, get_simple_index): |
|
525
|
|
|
''' |
|
526
|
|
|
target: test basic search fuction, all the search params is corrent, test all index params, and build |
|
527
|
|
|
method: search with the given vectors and tag, check the result |
|
528
|
|
|
expected: search status ok, and the length of the result is top_k |
|
529
|
|
|
''' |
|
530
|
|
|
top_k = 10 |
|
531
|
|
|
index_param = get_simple_index["index_param"] |
|
532
|
|
|
index_type = get_simple_index["index_type"] |
|
533
|
|
|
logging.getLogger().info(index_param) |
|
534
|
|
|
if index_type in [IndexType.RNSG, IndexType.IVF_PQ]: |
|
535
|
|
|
pytest.skip("rnsg not support in ip, skip pq") |
|
536
|
|
|
|
|
537
|
|
|
status = connect.create_partition(ip_collection, tag) |
|
538
|
|
|
vectors, ids = self.init_data(connect, ip_collection, partition_tags=tag) |
|
539
|
|
|
status = connect.create_index(ip_collection, index_type, index_param) |
|
540
|
|
|
query_vec = [vectors[0]] |
|
541
|
|
|
search_param = get_search_param(index_type) |
|
542
|
|
|
status, result = connect.search(ip_collection, top_k, query_vec, partition_tags=[tag], params=search_param) |
|
543
|
|
|
logging.getLogger().info(result) |
|
544
|
|
|
assert status.OK() |
|
545
|
|
|
assert len(result[0]) == min(len(vectors), top_k) |
|
546
|
|
|
assert check_result(result[0], ids[0]) |
|
547
|
|
|
assert result[0][0].distance >= 1 - gen_inaccuracy(result[0][0].distance) |
|
548
|
|
|
|
|
549
|
|
|
@pytest.mark.level(2) |
|
550
|
|
|
def test_search_vectors_without_connect(self, dis_connect, collection): |
|
551
|
|
|
''' |
|
552
|
|
|
target: test search vectors without connection |
|
553
|
|
|
method: use dis connected instance, call search method and check if search successfully |
|
554
|
|
|
expected: raise exception |
|
555
|
|
|
''' |
|
556
|
|
|
query_vectors = [vectors[0]] |
|
557
|
|
|
nprobe = 1 |
|
558
|
|
|
with pytest.raises(Exception) as e: |
|
559
|
|
|
status, ids = dis_connect.search(collection, top_k, query_vectors) |
|
560
|
|
|
|
|
561
|
|
|
def test_search_collection_name_not_existed(self, connect, collection): |
|
562
|
|
|
''' |
|
563
|
|
|
target: search collection not existed |
|
564
|
|
|
method: search with the random collection_name, which is not in db |
|
565
|
|
|
expected: status not ok |
|
566
|
|
|
''' |
|
567
|
|
|
collection_name = gen_unique_str("not_existed_collection") |
|
568
|
|
|
nprobe = 1 |
|
569
|
|
|
query_vecs = [vectors[0]] |
|
570
|
|
|
status, result = connect.search(collection_name, top_k, query_vecs) |
|
571
|
|
|
assert not status.OK() |
|
572
|
|
|
|
|
573
|
|
|
def test_search_collection_name_None(self, connect, collection): |
|
574
|
|
|
''' |
|
575
|
|
|
target: search collection that collection name is None |
|
576
|
|
|
method: search with the collection_name: None |
|
577
|
|
|
expected: status not ok |
|
578
|
|
|
''' |
|
579
|
|
|
collection_name = None |
|
580
|
|
|
nprobe = 1 |
|
581
|
|
|
query_vecs = [vectors[0]] |
|
582
|
|
|
with pytest.raises(Exception) as e: |
|
583
|
|
|
status, result = connect.search(collection_name, top_k, query_vecs) |
|
584
|
|
|
|
|
585
|
|
|
def test_search_top_k_query_records(self, connect, collection): |
|
586
|
|
|
''' |
|
587
|
|
|
target: test search fuction, with search params: query_records |
|
588
|
|
|
method: search with the given query_records, which are subarrays of the inserted vectors |
|
589
|
|
|
expected: status ok and the returned vectors should be query_records |
|
590
|
|
|
''' |
|
591
|
|
|
top_k = 10 |
|
592
|
|
|
vectors, ids = self.init_data(connect, collection) |
|
593
|
|
|
query_vecs = [vectors[0], vectors[55], vectors[99]] |
|
594
|
|
|
status, result = connect.search(collection, top_k, query_vecs) |
|
595
|
|
|
assert status.OK() |
|
596
|
|
|
assert len(result) == len(query_vecs) |
|
597
|
|
|
for i in range(len(query_vecs)): |
|
598
|
|
|
assert len(result[i]) == top_k |
|
599
|
|
|
assert result[i][0].distance <= epsilon |
|
600
|
|
|
|
|
601
|
|
|
def test_search_distance_l2_flat_index(self, connect, collection): |
|
602
|
|
|
''' |
|
603
|
|
|
target: search collection, and check the result: distance |
|
604
|
|
|
method: compare the return distance value with value computed with Euclidean |
|
605
|
|
|
expected: the return distance equals to the computed value |
|
606
|
|
|
''' |
|
607
|
|
|
nb = 2 |
|
608
|
|
|
vectors, ids = self.init_data(connect, collection, nb=nb) |
|
609
|
|
|
query_vecs = [[0.50 for i in range(dim)]] |
|
610
|
|
|
distance_0 = numpy.linalg.norm(numpy.array(query_vecs[0]) - numpy.array(vectors[0])) |
|
611
|
|
|
distance_1 = numpy.linalg.norm(numpy.array(query_vecs[0]) - numpy.array(vectors[1])) |
|
612
|
|
|
status, result = connect.search(collection, top_k, query_vecs) |
|
613
|
|
|
assert abs(numpy.sqrt(result[0][0].distance) - min(distance_0, distance_1)) <= gen_inaccuracy( |
|
614
|
|
|
result[0][0].distance) |
|
615
|
|
|
|
|
616
|
|
|
def test_search_distance_ip_flat_index(self, connect, ip_collection): |
|
617
|
|
|
''' |
|
618
|
|
|
target: search ip_collection, and check the result: distance |
|
619
|
|
|
method: compare the return distance value with value computed with Inner product |
|
620
|
|
|
expected: the return distance equals to the computed value |
|
621
|
|
|
''' |
|
622
|
|
|
nb = 2 |
|
623
|
|
|
nprobe = 1 |
|
624
|
|
|
vectors, ids = self.init_data(connect, ip_collection, nb=nb) |
|
625
|
|
|
index_type = IndexType.FLAT |
|
626
|
|
|
index_param = { |
|
627
|
|
|
"nlist": 16384 |
|
628
|
|
|
} |
|
629
|
|
|
connect.create_index(ip_collection, index_type, index_param) |
|
630
|
|
|
logging.getLogger().info(connect.get_index_info(ip_collection)) |
|
631
|
|
|
query_vecs = [[0.50 for i in range(dim)]] |
|
632
|
|
|
distance_0 = numpy.inner(numpy.array(query_vecs[0]), numpy.array(vectors[0])) |
|
633
|
|
|
distance_1 = numpy.inner(numpy.array(query_vecs[0]), numpy.array(vectors[1])) |
|
634
|
|
|
search_param = get_search_param(index_type) |
|
635
|
|
|
status, result = connect.search(ip_collection, top_k, query_vecs, params=search_param) |
|
636
|
|
|
assert abs(result[0][0].distance - max(distance_0, distance_1)) <= gen_inaccuracy(result[0][0].distance) |
|
637
|
|
|
|
|
638
|
|
View Code Duplication |
def test_search_distance_jaccard_flat_index(self, connect, jac_collection): |
|
|
|
|
|
|
639
|
|
|
''' |
|
640
|
|
|
target: search ip_collection, and check the result: distance |
|
641
|
|
|
method: compare the return distance value with value computed with Inner product |
|
642
|
|
|
expected: the return distance equals to the computed value |
|
643
|
|
|
''' |
|
644
|
|
|
# from scipy.spatial import distance |
|
645
|
|
|
nprobe = 512 |
|
646
|
|
|
int_vectors, vectors, ids = self.init_binary_data(connect, jac_collection, nb=2) |
|
647
|
|
|
index_type = IndexType.FLAT |
|
648
|
|
|
index_param = { |
|
649
|
|
|
"nlist": 16384 |
|
650
|
|
|
} |
|
651
|
|
|
connect.create_index(jac_collection, index_type, index_param) |
|
652
|
|
|
logging.getLogger().info(connect.get_collection_info(jac_collection)) |
|
653
|
|
|
logging.getLogger().info(connect.get_index_info(jac_collection)) |
|
654
|
|
|
query_int_vectors, query_vecs, tmp_ids = self.init_binary_data(connect, jac_collection, nb=1, insert=False) |
|
655
|
|
|
distance_0 = jaccard(query_int_vectors[0], int_vectors[0]) |
|
656
|
|
|
distance_1 = jaccard(query_int_vectors[0], int_vectors[1]) |
|
657
|
|
|
search_param = get_search_param(index_type) |
|
658
|
|
|
status, result = connect.search(jac_collection, top_k, query_vecs, params=search_param) |
|
659
|
|
|
logging.getLogger().info(status) |
|
660
|
|
|
logging.getLogger().info(result) |
|
661
|
|
|
assert abs(result[0][0].distance - min(distance_0, distance_1)) <= epsilon |
|
662
|
|
|
|
|
663
|
|
|
def test_search_distance_jaccard_flat_index_metric_type(self, connect, jac_collection): |
|
664
|
|
|
''' |
|
665
|
|
|
target: search ip_collection, and check the result: distance |
|
666
|
|
|
method: compare the return distance value with value computed with HAMMING |
|
667
|
|
|
expected: the return distance equals to the computed value |
|
668
|
|
|
''' |
|
669
|
|
|
# from scipy.spatial import distance |
|
670
|
|
|
nprobe = 512 |
|
671
|
|
|
int_vectors, vectors, ids = self.init_binary_data(connect, jac_collection, nb=2) |
|
672
|
|
|
index_type = IndexType.FLAT |
|
673
|
|
|
index_param = { |
|
674
|
|
|
"nlist": 16384 |
|
675
|
|
|
} |
|
676
|
|
|
connect.create_index(jac_collection, index_type, index_param) |
|
677
|
|
|
logging.getLogger().info(connect.get_collection_info(jac_collection)) |
|
678
|
|
|
logging.getLogger().info(connect.get_index_info(jac_collection)) |
|
679
|
|
|
query_int_vectors, query_vecs, tmp_ids = self.init_binary_data(connect, jac_collection, nb=1, insert=False) |
|
680
|
|
|
distance_0 = hamming(query_int_vectors[0], int_vectors[0]) |
|
681
|
|
|
distance_1 = hamming(query_int_vectors[0], int_vectors[1]) |
|
682
|
|
|
search_param = get_search_param(index_type) |
|
683
|
|
|
search_param.update({"metric_type": MetricType.HAMMING.value}) |
|
684
|
|
|
status, result = connect.search(jac_collection, top_k, query_vecs, params=search_param) |
|
685
|
|
|
assert status.OK() |
|
686
|
|
|
logging.getLogger().info(status) |
|
687
|
|
|
logging.getLogger().info(result) |
|
688
|
|
|
assert abs(result[0][0].distance - min(distance_0, distance_1).astype(float)) <= epsilon |
|
689
|
|
|
|
|
690
|
|
View Code Duplication |
def test_search_distance_hamming_flat_index(self, connect, ham_collection): |
|
|
|
|
|
|
691
|
|
|
''' |
|
692
|
|
|
target: search ip_collection, and check the result: distance |
|
693
|
|
|
method: compare the return distance value with value computed with Inner product |
|
694
|
|
|
expected: the return distance equals to the computed value |
|
695
|
|
|
''' |
|
696
|
|
|
# from scipy.spatial import distance |
|
697
|
|
|
nprobe = 512 |
|
698
|
|
|
int_vectors, vectors, ids = self.init_binary_data(connect, ham_collection, nb=2) |
|
699
|
|
|
index_type = IndexType.FLAT |
|
700
|
|
|
index_param = { |
|
701
|
|
|
"nlist": 16384 |
|
702
|
|
|
} |
|
703
|
|
|
connect.create_index(ham_collection, index_type, index_param) |
|
704
|
|
|
logging.getLogger().info(connect.get_collection_info(ham_collection)) |
|
705
|
|
|
logging.getLogger().info(connect.get_index_info(ham_collection)) |
|
706
|
|
|
query_int_vectors, query_vecs, tmp_ids = self.init_binary_data(connect, ham_collection, nb=1, insert=False) |
|
707
|
|
|
distance_0 = hamming(query_int_vectors[0], int_vectors[0]) |
|
708
|
|
|
distance_1 = hamming(query_int_vectors[0], int_vectors[1]) |
|
709
|
|
|
search_param = get_search_param(index_type) |
|
710
|
|
|
status, result = connect.search(ham_collection, top_k, query_vecs, params=search_param) |
|
711
|
|
|
logging.getLogger().info(status) |
|
712
|
|
|
logging.getLogger().info(result) |
|
713
|
|
|
assert abs(result[0][0].distance - min(distance_0, distance_1).astype(float)) <= epsilon |
|
714
|
|
|
|
|
715
|
|
|
def test_search_distance_substructure_flat_index(self, connect, substructure_collection): |
|
716
|
|
|
''' |
|
717
|
|
|
target: search ip_collection, and check the result: distance |
|
718
|
|
|
method: compare the return distance value with value computed with Inner product |
|
719
|
|
|
expected: the return distance equals to the computed value |
|
720
|
|
|
''' |
|
721
|
|
|
# from scipy.spatial import distance |
|
722
|
|
|
nprobe = 512 |
|
723
|
|
|
int_vectors, vectors, ids = self.init_binary_data(connect, substructure_collection, nb=2) |
|
724
|
|
|
index_type = IndexType.FLAT |
|
725
|
|
|
index_param = { |
|
726
|
|
|
"nlist": 16384 |
|
727
|
|
|
} |
|
728
|
|
|
connect.create_index(substructure_collection, index_type, index_param) |
|
729
|
|
|
logging.getLogger().info(connect.get_collection_info(substructure_collection)) |
|
730
|
|
|
logging.getLogger().info(connect.get_index_info(substructure_collection)) |
|
731
|
|
|
query_int_vectors, query_vecs, tmp_ids = self.init_binary_data(connect, substructure_collection, nb=1, |
|
732
|
|
|
insert=False) |
|
733
|
|
|
distance_0 = substructure(query_int_vectors[0], int_vectors[0]) |
|
734
|
|
|
distance_1 = substructure(query_int_vectors[0], int_vectors[1]) |
|
735
|
|
|
search_param = get_search_param(index_type) |
|
736
|
|
|
status, result = connect.search(substructure_collection, top_k, query_vecs, params=search_param) |
|
737
|
|
|
logging.getLogger().info(status) |
|
738
|
|
|
logging.getLogger().info(result) |
|
739
|
|
|
assert len(result[0]) == 0 |
|
740
|
|
|
|
|
741
|
|
View Code Duplication |
def test_search_distance_substructure_flat_index_B(self, connect, substructure_collection): |
|
|
|
|
|
|
742
|
|
|
''' |
|
743
|
|
|
target: search ip_collection, and check the result: distance |
|
744
|
|
|
method: compare the return distance value with value computed with SUB |
|
745
|
|
|
expected: the return distance equals to the computed value |
|
746
|
|
|
''' |
|
747
|
|
|
# from scipy.spatial import distance |
|
748
|
|
|
top_k = 3 |
|
749
|
|
|
nprobe = 512 |
|
750
|
|
|
int_vectors, vectors, ids = self.init_binary_data(connect, substructure_collection, nb=2) |
|
751
|
|
|
index_type = IndexType.FLAT |
|
752
|
|
|
index_param = { |
|
753
|
|
|
"nlist": 16384 |
|
754
|
|
|
} |
|
755
|
|
|
connect.create_index(substructure_collection, index_type, index_param) |
|
756
|
|
|
logging.getLogger().info(connect.get_collection_info(substructure_collection)) |
|
757
|
|
|
logging.getLogger().info(connect.get_index_info(substructure_collection)) |
|
758
|
|
|
query_int_vectors, query_vecs = gen_binary_sub_vectors(int_vectors, 2) |
|
759
|
|
|
search_param = get_search_param(index_type) |
|
760
|
|
|
status, result = connect.search(substructure_collection, top_k, query_vecs, params=search_param) |
|
761
|
|
|
logging.getLogger().info(status) |
|
762
|
|
|
logging.getLogger().info(result) |
|
763
|
|
|
assert len(result[0]) == 1 |
|
764
|
|
|
assert len(result[1]) == 1 |
|
765
|
|
|
assert result[0][0].distance <= epsilon |
|
766
|
|
|
assert result[0][0].id == ids[0] |
|
767
|
|
|
assert result[1][0].distance <= epsilon |
|
768
|
|
|
assert result[1][0].id == ids[1] |
|
769
|
|
|
|
|
770
|
|
|
def test_search_distance_superstructure_flat_index(self, connect, superstructure_collection): |
|
771
|
|
|
''' |
|
772
|
|
|
target: search ip_collection, and check the result: distance |
|
773
|
|
|
method: compare the return distance value with value computed with Inner product |
|
774
|
|
|
expected: the return distance equals to the computed value |
|
775
|
|
|
''' |
|
776
|
|
|
# from scipy.spatial import distance |
|
777
|
|
|
nprobe = 512 |
|
778
|
|
|
int_vectors, vectors, ids = self.init_binary_data(connect, superstructure_collection, nb=2) |
|
779
|
|
|
index_type = IndexType.FLAT |
|
780
|
|
|
index_param = { |
|
781
|
|
|
"nlist": 16384 |
|
782
|
|
|
} |
|
783
|
|
|
connect.create_index(superstructure_collection, index_type, index_param) |
|
784
|
|
|
logging.getLogger().info(connect.get_collection_info(superstructure_collection)) |
|
785
|
|
|
logging.getLogger().info(connect.get_index_info(superstructure_collection)) |
|
786
|
|
|
query_int_vectors, query_vecs, tmp_ids = self.init_binary_data(connect, superstructure_collection, nb=1, |
|
787
|
|
|
insert=False) |
|
788
|
|
|
distance_0 = superstructure(query_int_vectors[0], int_vectors[0]) |
|
789
|
|
|
distance_1 = superstructure(query_int_vectors[0], int_vectors[1]) |
|
790
|
|
|
search_param = get_search_param(index_type) |
|
791
|
|
|
status, result = connect.search(superstructure_collection, top_k, query_vecs, params=search_param) |
|
792
|
|
|
logging.getLogger().info(status) |
|
793
|
|
|
logging.getLogger().info(result) |
|
794
|
|
|
assert len(result[0]) == 0 |
|
795
|
|
|
|
|
796
|
|
View Code Duplication |
def test_search_distance_superstructure_flat_index_B(self, connect, superstructure_collection): |
|
|
|
|
|
|
797
|
|
|
''' |
|
798
|
|
|
target: search ip_collection, and check the result: distance |
|
799
|
|
|
method: compare the return distance value with value computed with SUPER |
|
800
|
|
|
expected: the return distance equals to the computed value |
|
801
|
|
|
''' |
|
802
|
|
|
# from scipy.spatial import distance |
|
803
|
|
|
top_k = 3 |
|
804
|
|
|
nprobe = 512 |
|
805
|
|
|
int_vectors, vectors, ids = self.init_binary_data(connect, superstructure_collection, nb=2) |
|
806
|
|
|
index_type = IndexType.FLAT |
|
807
|
|
|
index_param = { |
|
808
|
|
|
"nlist": 16384 |
|
809
|
|
|
} |
|
810
|
|
|
connect.create_index(superstructure_collection, index_type, index_param) |
|
811
|
|
|
logging.getLogger().info(connect.get_collection_info(superstructure_collection)) |
|
812
|
|
|
logging.getLogger().info(connect.get_index_info(superstructure_collection)) |
|
813
|
|
|
query_int_vectors, query_vecs = gen_binary_super_vectors(int_vectors, 2) |
|
814
|
|
|
search_param = get_search_param(index_type) |
|
815
|
|
|
status, result = connect.search(superstructure_collection, top_k, query_vecs, params=search_param) |
|
816
|
|
|
logging.getLogger().info(status) |
|
817
|
|
|
logging.getLogger().info(result) |
|
818
|
|
|
assert len(result[0]) == 2 |
|
819
|
|
|
assert len(result[1]) == 2 |
|
820
|
|
|
assert result[0][0].id in ids |
|
821
|
|
|
assert result[0][0].distance <= epsilon |
|
822
|
|
|
assert result[1][0].id in ids |
|
823
|
|
|
assert result[1][0].distance <= epsilon |
|
824
|
|
|
|
|
825
|
|
View Code Duplication |
def test_search_distance_tanimoto_flat_index(self, connect, tanimoto_collection): |
|
|
|
|
|
|
826
|
|
|
''' |
|
827
|
|
|
target: search ip_collection, and check the result: distance |
|
828
|
|
|
method: compare the return distance value with value computed with Inner product |
|
829
|
|
|
expected: the return distance equals to the computed value |
|
830
|
|
|
''' |
|
831
|
|
|
# from scipy.spatial import distance |
|
832
|
|
|
nprobe = 512 |
|
833
|
|
|
int_vectors, vectors, ids = self.init_binary_data(connect, tanimoto_collection, nb=2) |
|
834
|
|
|
index_type = IndexType.FLAT |
|
835
|
|
|
index_param = { |
|
836
|
|
|
"nlist": 16384 |
|
837
|
|
|
} |
|
838
|
|
|
connect.create_index(tanimoto_collection, index_type, index_param) |
|
839
|
|
|
logging.getLogger().info(connect.get_collection_info(tanimoto_collection)) |
|
840
|
|
|
logging.getLogger().info(connect.get_index_info(tanimoto_collection)) |
|
841
|
|
|
query_int_vectors, query_vecs, tmp_ids = self.init_binary_data(connect, tanimoto_collection, nb=1, insert=False) |
|
842
|
|
|
distance_0 = tanimoto(query_int_vectors[0], int_vectors[0]) |
|
843
|
|
|
distance_1 = tanimoto(query_int_vectors[0], int_vectors[1]) |
|
844
|
|
|
search_param = get_search_param(index_type) |
|
845
|
|
|
status, result = connect.search(tanimoto_collection, top_k, query_vecs, params=search_param) |
|
846
|
|
|
logging.getLogger().info(status) |
|
847
|
|
|
logging.getLogger().info(result) |
|
848
|
|
|
assert abs(result[0][0].distance - min(distance_0, distance_1)) <= epsilon |
|
849
|
|
|
|
|
850
|
|
|
def test_search_distance_ip_index_params(self, connect, ip_collection, get_index): |
|
851
|
|
|
''' |
|
852
|
|
|
target: search collection, and check the result: distance |
|
853
|
|
|
method: compare the return distance value with value computed with Inner product |
|
854
|
|
|
expected: the return distance equals to the computed value |
|
855
|
|
|
''' |
|
856
|
|
|
top_k = 2 |
|
857
|
|
|
nprobe = 1 |
|
858
|
|
|
index_param = get_index["index_param"] |
|
859
|
|
|
index_type = get_index["index_type"] |
|
860
|
|
|
if index_type == IndexType.RNSG: |
|
861
|
|
|
pytest.skip("rnsg not support in ip") |
|
862
|
|
|
vectors, ids = self.init_data(connect, ip_collection, nb=2) |
|
863
|
|
|
connect.create_index(ip_collection, index_type, index_param) |
|
864
|
|
|
logging.getLogger().info(connect.get_index_info(ip_collection)) |
|
865
|
|
|
query_vecs = [[0.50 for i in range(dim)]] |
|
866
|
|
|
search_param = get_search_param(index_type) |
|
867
|
|
|
status, result = connect.search(ip_collection, top_k, query_vecs, params=search_param) |
|
868
|
|
|
logging.getLogger().debug(status) |
|
869
|
|
|
logging.getLogger().debug(result) |
|
870
|
|
|
distance_0 = numpy.inner(numpy.array(query_vecs[0]), numpy.array(vectors[0])) |
|
871
|
|
|
distance_1 = numpy.inner(numpy.array(query_vecs[0]), numpy.array(vectors[1])) |
|
872
|
|
|
assert abs(result[0][0].distance - max(distance_0, distance_1)) <= gen_inaccuracy(result[0][0].distance) |
|
873
|
|
|
|
|
874
|
|
|
# def test_search_concurrent(self, connect, collection): |
|
875
|
|
|
# vectors, ids = self.init_data(connect, collection, nb=5000) |
|
876
|
|
|
# thread_num = 50 |
|
877
|
|
|
# nq = 1 |
|
878
|
|
|
# top_k = 2 |
|
879
|
|
|
# threads = [] |
|
880
|
|
|
# query_vecs = vectors[:nq] |
|
881
|
|
|
# def search(thread_number): |
|
882
|
|
|
# for i in range(1000000): |
|
883
|
|
|
# status, result = connect.search(collection, top_k, query_vecs, timeout=2) |
|
884
|
|
|
# assert len(result) == len(query_vecs) |
|
885
|
|
|
# assert status.OK() |
|
886
|
|
|
# if i % 1000 == 0: |
|
887
|
|
|
# logging.getLogger().info("In %d, %d" % (thread_number, i)) |
|
888
|
|
|
# logging.getLogger().info("%d finished" % thread_number) |
|
889
|
|
|
# # with concurrent.futures.ThreadPoolExecutor(max_workers=thread_num) as executor: |
|
890
|
|
|
# # future_results = {executor.submit( |
|
891
|
|
|
# # search): i for i in range(1000000)} |
|
892
|
|
|
# # for future in concurrent.futures.as_completed(future_results): |
|
893
|
|
|
# # future.result() |
|
894
|
|
|
# for i in range(thread_num): |
|
895
|
|
|
# t = threading.Thread(target=search, args=(i, )) |
|
896
|
|
|
# threads.append(t) |
|
897
|
|
|
# t.start() |
|
898
|
|
|
# for t in threads: |
|
899
|
|
|
# t.join() |
|
900
|
|
|
|
|
901
|
|
View Code Duplication |
@pytest.mark.level(2) |
|
|
|
|
|
|
902
|
|
|
@pytest.mark.timeout(30) |
|
903
|
|
|
def test_search_concurrent_multithreads(self, args): |
|
904
|
|
|
''' |
|
905
|
|
|
target: test concurrent search with multiprocessess |
|
906
|
|
|
method: search with 10 processes, each process uses dependent connection |
|
907
|
|
|
expected: status ok and the returned vectors should be query_records |
|
908
|
|
|
''' |
|
909
|
|
|
nb = 100 |
|
910
|
|
|
top_k = 10 |
|
911
|
|
|
threads_num = 4 |
|
912
|
|
|
threads = [] |
|
913
|
|
|
collection = gen_unique_str("test_search_concurrent_multiprocessing") |
|
914
|
|
|
uri = "tcp://%s:%s" % (args["ip"], args["port"]) |
|
915
|
|
|
param = {'collection_name': collection, |
|
916
|
|
|
'dimension': dim, |
|
917
|
|
|
'index_type': IndexType.FLAT, |
|
918
|
|
|
'store_raw_vector': False} |
|
919
|
|
|
# create collection |
|
920
|
|
|
milvus = get_milvus(args["ip"], args["port"], handler=args["handler"]) |
|
921
|
|
|
milvus.create_collection(param) |
|
922
|
|
|
vectors, ids = self.init_data(milvus, collection, nb=nb) |
|
923
|
|
|
query_vecs = vectors[nb // 2:nb] |
|
924
|
|
|
|
|
925
|
|
|
def search(milvus): |
|
926
|
|
|
status, result = milvus.search(collection, top_k, query_vecs) |
|
927
|
|
|
assert len(result) == len(query_vecs) |
|
928
|
|
|
for i in range(len(query_vecs)): |
|
929
|
|
|
assert result[i][0].id in ids |
|
930
|
|
|
assert result[i][0].distance == 0.0 |
|
931
|
|
|
|
|
932
|
|
|
for i in range(threads_num): |
|
933
|
|
|
milvus = get_milvus(args["ip"], args["port"], handler=args["handler"]) |
|
934
|
|
|
t = threading.Thread(target=search, args=(milvus,)) |
|
935
|
|
|
threads.append(t) |
|
936
|
|
|
t.start() |
|
937
|
|
|
time.sleep(0.2) |
|
938
|
|
|
for t in threads: |
|
939
|
|
|
t.join() |
|
940
|
|
|
|
|
941
|
|
|
# TODO: enable |
|
942
|
|
View Code Duplication |
@pytest.mark.timeout(30) |
|
|
|
|
|
|
943
|
|
|
def _test_search_concurrent_multiprocessing(self, args): |
|
944
|
|
|
''' |
|
945
|
|
|
target: test concurrent search with multiprocessess |
|
946
|
|
|
method: search with 10 processes, each process uses dependent connection |
|
947
|
|
|
expected: status ok and the returned vectors should be query_records |
|
948
|
|
|
''' |
|
949
|
|
|
nb = 100 |
|
950
|
|
|
top_k = 10 |
|
951
|
|
|
process_num = 4 |
|
952
|
|
|
processes = [] |
|
953
|
|
|
collection = gen_unique_str("test_search_concurrent_multiprocessing") |
|
954
|
|
|
uri = "tcp://%s:%s" % (args["ip"], args["port"]) |
|
955
|
|
|
param = {'collection_name': collection, |
|
956
|
|
|
'dimension': dim, |
|
957
|
|
|
'index_type': IndexType.FLAT, |
|
958
|
|
|
'store_raw_vector': False} |
|
959
|
|
|
# create collection |
|
960
|
|
|
milvus = get_milvus(args["ip"], args["port"], handler=args["handler"]) |
|
961
|
|
|
milvus.create_collection(param) |
|
962
|
|
|
vectors, ids = self.init_data(milvus, collection, nb=nb) |
|
963
|
|
|
query_vecs = vectors[nb // 2:nb] |
|
964
|
|
|
|
|
965
|
|
|
def search(milvus): |
|
966
|
|
|
status, result = milvus.search(collection, top_k, query_vecs) |
|
967
|
|
|
assert len(result) == len(query_vecs) |
|
968
|
|
|
for i in range(len(query_vecs)): |
|
969
|
|
|
assert result[i][0].id in ids |
|
970
|
|
|
assert result[i][0].distance == 0.0 |
|
971
|
|
|
|
|
972
|
|
|
for i in range(process_num): |
|
973
|
|
|
milvus = get_milvus(args["ip"], args["port"], handler=args["handler"]) |
|
974
|
|
|
p = Process(target=search, args=(milvus,)) |
|
975
|
|
|
processes.append(p) |
|
976
|
|
|
p.start() |
|
977
|
|
|
time.sleep(0.2) |
|
978
|
|
|
for p in processes: |
|
979
|
|
|
p.join() |
|
980
|
|
|
|
|
981
|
|
View Code Duplication |
def test_search_multi_collection_L2(search, args): |
|
|
|
|
|
|
982
|
|
|
''' |
|
983
|
|
|
target: test search multi collections of L2 |
|
984
|
|
|
method: add vectors into 10 collections, and search |
|
985
|
|
|
expected: search status ok, the length of result |
|
986
|
|
|
''' |
|
987
|
|
|
num = 10 |
|
988
|
|
|
top_k = 10 |
|
989
|
|
|
collections = [] |
|
990
|
|
|
idx = [] |
|
991
|
|
|
for i in range(num): |
|
992
|
|
|
collection = gen_unique_str("test_add_multicollection_%d" % i) |
|
993
|
|
|
uri = "tcp://%s:%s" % (args["ip"], args["port"]) |
|
994
|
|
|
param = {'collection_name': collection, |
|
995
|
|
|
'dimension': dim, |
|
996
|
|
|
'index_file_size': 10, |
|
997
|
|
|
'metric_type': MetricType.L2} |
|
998
|
|
|
# create collection |
|
999
|
|
|
milvus = get_milvus(args["ip"], args["port"], handler=args["handler"]) |
|
1000
|
|
|
milvus.create_collection(param) |
|
1001
|
|
|
status, ids = milvus.insert(collection, vectors) |
|
1002
|
|
|
assert status.OK() |
|
1003
|
|
|
assert len(ids) == len(vectors) |
|
1004
|
|
|
collections.append(collection) |
|
1005
|
|
|
idx.append(ids[0]) |
|
1006
|
|
|
idx.append(ids[10]) |
|
1007
|
|
|
idx.append(ids[20]) |
|
1008
|
|
|
milvus.flush([collection]) |
|
1009
|
|
|
query_vecs = [vectors[0], vectors[10], vectors[20]] |
|
1010
|
|
|
# start query from random collection |
|
1011
|
|
|
for i in range(num): |
|
1012
|
|
|
collection = collections[i] |
|
1013
|
|
|
status, result = milvus.search(collection, top_k, query_vecs) |
|
|
|
|
|
|
1014
|
|
|
assert status.OK() |
|
1015
|
|
|
assert len(result) == len(query_vecs) |
|
1016
|
|
|
for j in range(len(query_vecs)): |
|
1017
|
|
|
assert len(result[j]) == top_k |
|
1018
|
|
|
for j in range(len(query_vecs)): |
|
1019
|
|
|
assert check_result(result[j], idx[3 * i + j]) |
|
1020
|
|
|
|
|
1021
|
|
View Code Duplication |
def test_search_multi_collection_IP(search, args): |
|
|
|
|
|
|
1022
|
|
|
''' |
|
1023
|
|
|
target: test search multi collections of IP |
|
1024
|
|
|
method: add vectors into 10 collections, and search |
|
1025
|
|
|
expected: search status ok, the length of result |
|
1026
|
|
|
''' |
|
1027
|
|
|
num = 10 |
|
1028
|
|
|
top_k = 10 |
|
1029
|
|
|
collections = [] |
|
1030
|
|
|
idx = [] |
|
1031
|
|
|
for i in range(num): |
|
1032
|
|
|
collection = gen_unique_str("test_add_multicollection_%d" % i) |
|
1033
|
|
|
uri = "tcp://%s:%s" % (args["ip"], args["port"]) |
|
1034
|
|
|
param = {'collection_name': collection, |
|
1035
|
|
|
'dimension': dim, |
|
1036
|
|
|
'index_file_size': 10, |
|
1037
|
|
|
'metric_type': MetricType.L2} |
|
1038
|
|
|
# create collection |
|
1039
|
|
|
milvus = get_milvus(args["ip"], args["port"], handler=args["handler"]) |
|
1040
|
|
|
milvus.create_collection(param) |
|
1041
|
|
|
status, ids = milvus.insert(collection, vectors) |
|
1042
|
|
|
assert status.OK() |
|
1043
|
|
|
assert len(ids) == len(vectors) |
|
1044
|
|
|
collections.append(collection) |
|
1045
|
|
|
idx.append(ids[0]) |
|
1046
|
|
|
idx.append(ids[10]) |
|
1047
|
|
|
idx.append(ids[20]) |
|
1048
|
|
|
milvus.flush([collection]) |
|
1049
|
|
|
query_vecs = [vectors[0], vectors[10], vectors[20]] |
|
1050
|
|
|
# start query from random collection |
|
1051
|
|
|
for i in range(num): |
|
1052
|
|
|
collection = collections[i] |
|
1053
|
|
|
status, result = milvus.search(collection, top_k, query_vecs) |
|
|
|
|
|
|
1054
|
|
|
assert status.OK() |
|
1055
|
|
|
assert len(result) == len(query_vecs) |
|
1056
|
|
|
for j in range(len(query_vecs)): |
|
1057
|
|
|
assert len(result[j]) == top_k |
|
1058
|
|
|
for j in range(len(query_vecs)): |
|
1059
|
|
|
assert check_result(result[j], idx[3 * i + j]) |
|
1060
|
|
|
|
|
1061
|
|
|
@pytest.fixture(params=MetricType) |
|
1062
|
|
|
def get_binary_metric_types(self, request): |
|
1063
|
|
|
if request.param == MetricType.INVALID: |
|
1064
|
|
|
pytest.skip(("metric type invalid")) |
|
1065
|
|
|
if request.param in [MetricType.L2, MetricType.IP]: |
|
1066
|
|
|
pytest.skip(("L2 and IP not support in binary")) |
|
1067
|
|
|
return request.param |
|
1068
|
|
|
|
|
1069
|
|
|
# 4678 and # 4683 |
|
1070
|
|
|
def test_search_binary_dim_not_power_of_2(self, connect, get_binary_metric_types): |
|
1071
|
|
|
metric = get_binary_metric_types |
|
1072
|
|
|
collection = gen_unique_str(collection_id) |
|
1073
|
|
|
dim = 200 |
|
1074
|
|
|
top_k = 1 |
|
1075
|
|
|
param = {'collection_name': collection, |
|
1076
|
|
|
'dimension': dim, |
|
1077
|
|
|
'index_file_size': 10, |
|
1078
|
|
|
'metric_type': metric} |
|
1079
|
|
|
status = connect.create_collection(param) |
|
1080
|
|
|
assert status.OK() |
|
1081
|
|
|
int_vectors, vectors, ids = self.init_binary_data(connect, collection, nb=1000, dim=dim) |
|
1082
|
|
|
search_param = get_search_param(IndexType.FLAT) |
|
1083
|
|
|
status, result = connect.search(collection, top_k, vectors[:1], params=search_param) |
|
1084
|
|
|
assert status.OK() |
|
1085
|
|
|
assert result[0][0].id in ids |
|
1086
|
|
|
assert result[0][0].distance == 0.0 |
|
1087
|
|
|
|
|
1088
|
|
|
@pytest.fixture(params=MetricType) |
|
1089
|
|
|
def get_metric_types(self, request): |
|
1090
|
|
|
if request.param == MetricType.INVALID: |
|
1091
|
|
|
pytest.skip(("metric type invalid")) |
|
1092
|
|
|
if request.param not in [MetricType.L2, MetricType.IP]: |
|
1093
|
|
|
pytest.skip(("L2 and IP not support in binary")) |
|
1094
|
|
|
return request.param |
|
1095
|
|
|
|
|
1096
|
|
|
def test_search_float_dim_not_power_of_2(self, connect, get_metric_types): |
|
1097
|
|
|
metric = get_metric_types |
|
1098
|
|
|
collection = gen_unique_str(collection_id) |
|
1099
|
|
|
dim = 200 |
|
1100
|
|
|
top_k = 1 |
|
1101
|
|
|
param = {'collection_name': collection, |
|
1102
|
|
|
'dimension': dim, |
|
1103
|
|
|
'index_file_size': 10, |
|
1104
|
|
|
'metric_type': metric} |
|
1105
|
|
|
status = connect.create_collection(param) |
|
1106
|
|
|
assert status.OK() |
|
1107
|
|
|
vectors, ids = self.init_data(connect, collection, nb=1000, dim=dim) |
|
1108
|
|
|
search_param = get_search_param(IndexType.FLAT) |
|
1109
|
|
|
status, result = connect.search(collection, top_k, vectors[:1], params=search_param) |
|
1110
|
|
|
assert status.OK() |
|
1111
|
|
|
assert result[0][0].id in ids |
|
1112
|
|
|
|
|
1113
|
|
|
""" |
|
1114
|
|
|
****************************************************************** |
|
1115
|
|
|
# The following cases are used to test `search_vectors` function |
|
1116
|
|
|
# with invalid collection_name top-k / nprobe / query_range |
|
1117
|
|
|
****************************************************************** |
|
1118
|
|
|
""" |
|
1119
|
|
|
|
|
1120
|
|
|
|
|
1121
|
|
|
class TestSearchParamsInvalid(object): |
|
1122
|
|
|
nlist = 16384 |
|
1123
|
|
|
index_type = IndexType.IVF_SQ8 |
|
1124
|
|
|
index_param = {"nlist": nlist} |
|
1125
|
|
|
logging.getLogger().info(index_param) |
|
1126
|
|
|
|
|
1127
|
|
|
def init_data(self, connect, collection, nb=6000): |
|
1128
|
|
|
''' |
|
1129
|
|
|
Generate vectors and add it in collection, before search vectors |
|
1130
|
|
|
''' |
|
1131
|
|
|
global vectors |
|
1132
|
|
|
if nb == 6000: |
|
1133
|
|
|
insert = vectors |
|
1134
|
|
|
else: |
|
1135
|
|
|
insert = gen_vectors(nb, dim) |
|
1136
|
|
|
status, ids = connect.insert(collection, insert) |
|
1137
|
|
|
connect.flush([collection]) |
|
1138
|
|
|
return insert, ids |
|
1139
|
|
|
|
|
1140
|
|
|
""" |
|
1141
|
|
|
Test search collection with invalid collection names |
|
1142
|
|
|
""" |
|
1143
|
|
|
|
|
1144
|
|
|
@pytest.fixture( |
|
1145
|
|
|
scope="function", |
|
1146
|
|
|
params=gen_invalid_collection_names() |
|
1147
|
|
|
) |
|
1148
|
|
|
def get_collection_name(self, request): |
|
1149
|
|
|
yield request.param |
|
1150
|
|
|
|
|
1151
|
|
|
@pytest.mark.level(2) |
|
1152
|
|
|
def test_search_with_invalid_collectionname(self, connect, get_collection_name): |
|
1153
|
|
|
collection_name = get_collection_name |
|
1154
|
|
|
logging.getLogger().info(collection_name) |
|
1155
|
|
|
nprobe = 1 |
|
1156
|
|
|
query_vecs = gen_vectors(1, dim) |
|
1157
|
|
|
status, result = connect.search(collection_name, top_k, query_vecs) |
|
1158
|
|
|
assert not status.OK() |
|
1159
|
|
|
|
|
1160
|
|
|
@pytest.mark.level(1) |
|
1161
|
|
|
def test_search_with_invalid_tag_format(self, connect, collection): |
|
1162
|
|
|
nprobe = 1 |
|
1163
|
|
|
query_vecs = gen_vectors(1, dim) |
|
1164
|
|
|
with pytest.raises(Exception) as e: |
|
1165
|
|
|
status, result = connect.search(collection, top_k, query_vecs, partition_tags="tag") |
|
1166
|
|
|
logging.getLogger().debug(result) |
|
1167
|
|
|
|
|
1168
|
|
|
@pytest.mark.level(1) |
|
1169
|
|
|
def test_search_with_tag_not_existed(self, connect, collection): |
|
1170
|
|
|
nprobe = 1 |
|
1171
|
|
|
query_vecs = gen_vectors(1, dim) |
|
1172
|
|
|
status, result = connect.search(collection, top_k, query_vecs, partition_tags=["tag"]) |
|
1173
|
|
|
logging.getLogger().info(result) |
|
1174
|
|
|
assert not status.OK() |
|
1175
|
|
|
|
|
1176
|
|
|
""" |
|
1177
|
|
|
Test search collection with invalid top-k |
|
1178
|
|
|
""" |
|
1179
|
|
|
|
|
1180
|
|
|
@pytest.fixture( |
|
1181
|
|
|
scope="function", |
|
1182
|
|
|
params=gen_invalid_top_ks() |
|
1183
|
|
|
) |
|
1184
|
|
|
def get_top_k(self, request): |
|
1185
|
|
|
yield request.param |
|
1186
|
|
|
|
|
1187
|
|
View Code Duplication |
@pytest.mark.level(1) |
|
|
|
|
|
|
1188
|
|
|
def test_search_with_invalid_top_k(self, connect, collection, get_top_k): |
|
1189
|
|
|
''' |
|
1190
|
|
|
target: test search fuction, with the wrong top_k |
|
1191
|
|
|
method: search with top_k |
|
1192
|
|
|
expected: raise an error, and the connection is normal |
|
1193
|
|
|
''' |
|
1194
|
|
|
top_k = get_top_k |
|
1195
|
|
|
logging.getLogger().info(top_k) |
|
1196
|
|
|
nprobe = 1 |
|
1197
|
|
|
query_vecs = gen_vectors(1, dim) |
|
1198
|
|
|
if isinstance(top_k, int): |
|
1199
|
|
|
status, result = connect.search(collection, top_k, query_vecs) |
|
1200
|
|
|
assert not status.OK() |
|
1201
|
|
|
else: |
|
1202
|
|
|
with pytest.raises(Exception) as e: |
|
1203
|
|
|
status, result = connect.search(collection, top_k, query_vecs) |
|
1204
|
|
|
|
|
1205
|
|
View Code Duplication |
@pytest.mark.level(2) |
|
|
|
|
|
|
1206
|
|
|
def test_search_with_invalid_top_k_ip(self, connect, ip_collection, get_top_k): |
|
1207
|
|
|
''' |
|
1208
|
|
|
target: test search fuction, with the wrong top_k |
|
1209
|
|
|
method: search with top_k |
|
1210
|
|
|
expected: raise an error, and the connection is normal |
|
1211
|
|
|
''' |
|
1212
|
|
|
top_k = get_top_k |
|
1213
|
|
|
logging.getLogger().info(top_k) |
|
1214
|
|
|
nprobe = 1 |
|
1215
|
|
|
query_vecs = gen_vectors(1, dim) |
|
1216
|
|
|
if isinstance(top_k, int): |
|
1217
|
|
|
status, result = connect.search(ip_collection, top_k, query_vecs) |
|
1218
|
|
|
assert not status.OK() |
|
1219
|
|
|
else: |
|
1220
|
|
|
with pytest.raises(Exception) as e: |
|
1221
|
|
|
status, result = connect.search(ip_collection, top_k, query_vecs) |
|
1222
|
|
|
|
|
1223
|
|
|
""" |
|
1224
|
|
|
Test search collection with invalid nprobe |
|
1225
|
|
|
""" |
|
1226
|
|
|
|
|
1227
|
|
|
@pytest.fixture( |
|
1228
|
|
|
scope="function", |
|
1229
|
|
|
params=gen_invalid_nprobes() |
|
1230
|
|
|
) |
|
1231
|
|
|
def get_nprobes(self, request): |
|
1232
|
|
|
yield request.param |
|
1233
|
|
|
|
|
1234
|
|
View Code Duplication |
@pytest.mark.level(1) |
|
|
|
|
|
|
1235
|
|
|
def test_search_with_invalid_nprobe(self, connect, collection, get_nprobes): |
|
1236
|
|
|
''' |
|
1237
|
|
|
target: test search fuction, with the wrong nprobe |
|
1238
|
|
|
method: search with nprobe |
|
1239
|
|
|
expected: raise an error, and the connection is normal |
|
1240
|
|
|
''' |
|
1241
|
|
|
index_type = IndexType.IVF_SQ8 |
|
1242
|
|
|
index_param = {"nlist": 16384} |
|
1243
|
|
|
connect.create_index(collection, index_type, index_param) |
|
1244
|
|
|
nprobe = get_nprobes |
|
1245
|
|
|
search_param = {"nprobe": nprobe} |
|
1246
|
|
|
logging.getLogger().info(nprobe) |
|
1247
|
|
|
query_vecs = gen_vectors(1, dim) |
|
1248
|
|
|
# if isinstance(nprobe, int): |
|
1249
|
|
|
status, result = connect.search(collection, top_k, query_vecs, params=search_param) |
|
1250
|
|
|
assert not status.OK() |
|
1251
|
|
|
# else: |
|
1252
|
|
|
# with pytest.raises(Exception) as e: |
|
1253
|
|
|
# status, result = connect.search(collection, top_k, query_vecs, params=search_param) |
|
1254
|
|
|
|
|
1255
|
|
View Code Duplication |
@pytest.mark.level(2) |
|
|
|
|
|
|
1256
|
|
|
def test_search_with_invalid_nprobe_ip(self, connect, ip_collection, get_nprobes): |
|
1257
|
|
|
''' |
|
1258
|
|
|
target: test search fuction, with the wrong top_k |
|
1259
|
|
|
method: search with top_k |
|
1260
|
|
|
expected: raise an error, and the connection is normal |
|
1261
|
|
|
''' |
|
1262
|
|
|
index_type = IndexType.IVF_SQ8 |
|
1263
|
|
|
index_param = {"nlist": 16384} |
|
1264
|
|
|
connect.create_index(ip_collection, index_type, index_param) |
|
1265
|
|
|
nprobe = get_nprobes |
|
1266
|
|
|
search_param = {"nprobe": nprobe} |
|
1267
|
|
|
logging.getLogger().info(nprobe) |
|
1268
|
|
|
query_vecs = gen_vectors(1, dim) |
|
1269
|
|
|
|
|
1270
|
|
|
# if isinstance(nprobe, int): |
|
1271
|
|
|
status, result = connect.search(ip_collection, top_k, query_vecs, params=search_param) |
|
1272
|
|
|
assert not status.OK() |
|
1273
|
|
|
# else: |
|
1274
|
|
|
# with pytest.raises(Exception) as e: |
|
1275
|
|
|
# status, result = connect.search(ip_collection, top_k, query_vecs, params=search_param) |
|
1276
|
|
|
|
|
1277
|
|
|
def test_search_with_2049_nprobe(self, connect, collection): |
|
1278
|
|
|
''' |
|
1279
|
|
|
target: test search function, with 2049 nprobe in GPU mode |
|
1280
|
|
|
method: search with nprobe |
|
1281
|
|
|
expected: status not ok |
|
1282
|
|
|
''' |
|
1283
|
|
|
if str(connect._cmd("mode")[1]) == "CPU": |
|
1284
|
|
|
pytest.skip("Only support GPU mode") |
|
1285
|
|
|
for index in gen_simple_index(): |
|
1286
|
|
|
if index["index_type"] in [IndexType.IVF_PQ, IndexType.IVFLAT, IndexType.IVF_SQ8, IndexType.IVF_SQ8H]: |
|
1287
|
|
|
index_type = index["index_type"] |
|
1288
|
|
|
index_param = index["index_param"] |
|
1289
|
|
|
self.init_data(connect, collection) |
|
1290
|
|
|
connect.create_index(collection, index_type, index_param) |
|
1291
|
|
|
nprobe = 2049 |
|
1292
|
|
|
search_param = {"nprobe": nprobe} |
|
1293
|
|
|
query_vecs = gen_vectors(nprobe, dim) |
|
1294
|
|
|
status, result = connect.search(collection, top_k, query_vecs, params=search_param) |
|
1295
|
|
|
assert status.OK() |
|
1296
|
|
|
|
|
1297
|
|
View Code Duplication |
@pytest.fixture( |
|
|
|
|
|
|
1298
|
|
|
scope="function", |
|
1299
|
|
|
params=gen_simple_index() |
|
1300
|
|
|
) |
|
1301
|
|
|
def get_simple_index(self, request, connect): |
|
1302
|
|
|
if str(connect._cmd("mode")[1]) == "CPU": |
|
1303
|
|
|
if request.param["index_type"] == IndexType.IVF_SQ8H: |
|
1304
|
|
|
pytest.skip("sq8h not support in CPU mode") |
|
1305
|
|
|
if str(connect._cmd("mode")[1]) == "GPU": |
|
1306
|
|
|
if request.param["index_type"] == IndexType.IVF_PQ: |
|
1307
|
|
|
pytest.skip("ivfpq not support in GPU mode") |
|
1308
|
|
|
return request.param |
|
1309
|
|
|
|
|
1310
|
|
|
def test_search_with_empty_params(self, connect, collection, args, get_simple_index): |
|
1311
|
|
|
''' |
|
1312
|
|
|
target: test search fuction, with empty search params |
|
1313
|
|
|
method: search with params |
|
1314
|
|
|
expected: search status not ok, and the connection is normal |
|
1315
|
|
|
''' |
|
1316
|
|
|
if args["handler"] == "HTTP": |
|
1317
|
|
|
pytest.skip("skip in http mode") |
|
1318
|
|
|
index_type = get_simple_index["index_type"] |
|
1319
|
|
|
index_param = get_simple_index["index_param"] |
|
1320
|
|
|
connect.create_index(collection, index_type, index_param) |
|
1321
|
|
|
query_vecs = gen_vectors(1, dim) |
|
1322
|
|
|
status, result = connect.search(collection, top_k, query_vecs, params={}) |
|
1323
|
|
|
|
|
1324
|
|
|
if index_type == IndexType.FLAT: |
|
1325
|
|
|
assert status.OK() |
|
1326
|
|
|
else: |
|
1327
|
|
|
assert not status.OK() |
|
1328
|
|
|
|
|
1329
|
|
|
@pytest.fixture( |
|
1330
|
|
|
scope="function", |
|
1331
|
|
|
params=gen_invaild_search_params() |
|
1332
|
|
|
) |
|
1333
|
|
|
def get_invalid_search_param(self, request, connect): |
|
1334
|
|
|
if str(connect._cmd("mode")[1]) == "CPU": |
|
1335
|
|
|
if request.param["index_type"] == IndexType.IVF_SQ8H: |
|
1336
|
|
|
pytest.skip("sq8h not support in CPU mode") |
|
1337
|
|
|
if str(connect._cmd("mode")[1]) == "GPU": |
|
1338
|
|
|
if request.param["index_type"] == IndexType.IVF_PQ: |
|
1339
|
|
|
pytest.skip("ivfpq not support in GPU mode") |
|
1340
|
|
|
return request.param |
|
1341
|
|
|
|
|
1342
|
|
|
def test_search_with_invalid_params(self, connect, collection, get_invalid_search_param): |
|
1343
|
|
|
''' |
|
1344
|
|
|
target: test search fuction, with invalid search params |
|
1345
|
|
|
method: search with params |
|
1346
|
|
|
expected: search status not ok, and the connection is normal |
|
1347
|
|
|
''' |
|
1348
|
|
|
index_type = get_invalid_search_param["index_type"] |
|
1349
|
|
|
search_param = get_invalid_search_param["search_param"] |
|
1350
|
|
|
for index in gen_simple_index(): |
|
1351
|
|
|
if index_type == index["index_type"]: |
|
1352
|
|
|
connect.create_index(collection, index_type, index["index_param"]) |
|
1353
|
|
|
query_vecs = gen_vectors(1, dim) |
|
1354
|
|
|
status, result = connect.search(collection, top_k, query_vecs, params=search_param) |
|
1355
|
|
|
assert not status.OK() |
|
1356
|
|
|
|
|
1357
|
|
|
|
|
1358
|
|
|
def check_result(result, id): |
|
1359
|
|
|
if len(result) >= 5: |
|
1360
|
|
|
return id in [result[0].id, result[1].id, result[2].id, result[3].id, result[4].id] |
|
1361
|
|
|
else: |
|
1362
|
|
|
return id in (i.id for i in result) |
|
|
|
|
|
|
1363
|
|
|
|
milvus-io /
milvus
Loading history...
