1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
|
# Copyright (C) 2022-present MongoDB, Inc.
#
# This program is free software: you can redistribute it and/or modify
# it under the terms of the Server Side Public License, version 1,
# as published by MongoDB, Inc.
#
# This program is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
# Server Side Public License for more details.
#
# You should have received a copy of the Server Side Public License
# along with this program. If not, see
# <http://www.mongodb.com/licensing/server-side-public-license>.
#
# As a special exception, the copyright holders give permission to link the
# code of portions of this program with the OpenSSL library under certain
# conditions as described in each individual source file and distribute
# linked combinations including the program with the OpenSSL library. You
# must comply with the Server Side Public License in all respects for
# all of the code used other than as permitted herein. If you modify file(s)
# with this exception, you may extend this exception to your version of the
# file(s), but you are not obligated to do so. If you do not wish to do so,
# delete this exception statement from your version. If you delete this
# exception statement from all source files in the program, then also delete
# it in the license file.
#
"""A/B performance test to compare plans produced by different sets of cost model coefficients."""
from __future__ import annotations
import asyncio
import logging
from dataclasses import asdict, dataclass
from typing import Sequence
from scipy import stats
import bson.json_util as json
from config import BenchmarkConfig
from database_instance import DatabaseInstance, Pipeline, get_database_parameter
import physical_tree
import execution_tree
@dataclass
class CostModelCoefficients:
"""Represent MongoDB's Cost Model Coefficients initially defined in the cost_model.idl file."""
scan_incremental_cost: float | None = None
scan_startup_cost: float | None = None
index_scan_incremental_cost: float | None = None
index_scan_startup_cost: float | None = None
seek_cost: float | None = None
seek_startup_cost: float | None = None
filter_incremental_cost: float | None = None
filter_startup_cost: float | None = None
eval_incremental_cost: float | None = None
eval_startup_cost: float | None = None
group_by_incremental_cost: float | None = None
group_by_startup_cost: float | None = None
unwind_incremental_cost: float | None = None
unwind_startup_cost: float | None = None
nested_loop_join_incremental_cost: float | None = None
nested_loop_join_startup_cost: float | None = None
hash_join_incremental_cost: float | None = None
hash_join_startup_cost: float | None = None
merge_join_incremental_cost: float | None = None
merge_join_startup_cost: float | None = None
unique_incremental_cost: float | None = None
unique_startup_cost: float | None = None
collation_incremental_cost: float | None = None
collation_startup_cost: float | None = None
collation_with_limit_incremental_cost: float | None = None
collation_with_limit_startup_cost: float | None = None
union_incremental_cost: float | None = None
union_startup_cost: float | None = None
exchange_incremental_cost: float | None = None
exchange_startup_cost: float | None = None
limit_skip_incremental_cost: float | None = None
limit_skip_startup_cost: float | None = None
def to_dict(self):
"""Convert to a dictionary with keys defined according to cost_model.idl."""
fields = asdict(self)
return {to_camel_case(k): v for k, v in fields.items() if v is not None}
def to_camel_case(string):
"""Convert a snake_case string to camelCase one."""
words = string.split('_')
return words[0] + ''.join(w.capitalize() for w in words[1:])
@dataclass
class BenchmarkTask:
"""Define an A/B performance testing task."""
collection_name: str
pipeline: Pipeline
# Statistical significance threshold, usually 0.05 or 0.05.
threshold: float
# Cost Model Coefficients overrides of Variant A.
cost_model_a: CostModelCoefficients
# Cost Model Coefficients overrides of Variant B.
cost_model_b: CostModelCoefficients
def print(self):
"""Prints the task."""
print('Cost Model Coefficients Overrides:')
print(f'\tA: {self.cost_model_a.to_dict()}')
print(f'\tB: {self.cost_model_b.to_dict()}')
print(f'threshold: {self.threshold}')
print(f'collection: {self.collection_name}')
print(f'pipeline: {self.pipeline}')
@dataclass
class BenchmarkResult:
"""Represent the A/B performance testing results."""
task: BenchmarkTask
variant_a: ExperimentResult
variant_b: ExperimentResult
# If the p-value is less than the task threshold value,
# there is no significant difference between the variant of the Cost Model Coefficients.
pvalue: float
def print(self):
"""Print the results."""
print('## Benchmark Task')
self.task.print()
print('\n## Result')
print(f'Means: A: {self.variant_a.mean:,.2f}, B: {self.variant_b.mean:,.2f}.')
print(f"t-test's p-value: {self.pvalue}.")
if self.pvalue < self.task.threshold:
print('The means are significantly different.')
else:
print('The means are not significantly different.')
print('\n### A\n')
self.variant_a.print()
print('\n### B\n')
self.variant_b.print()
@dataclass
class ExperimentResult:
"""Represent one variant results of an A/B performance test."""
explain: Sequence[dict[str, any]]
physical_tree: Sequence[physical_tree.Node]
execution_tree: Sequence[execution_tree.Node]
mean: float
def print(self, index: int = None):
"""Print the results."""
if index is None:
index = len(self.explain) // 2
print('ABT Physical Tree')
self.physical_tree[index].print()
print('\nSBE Execution Tree')
self.execution_tree[index].print()
async def benchmark(config: BenchmarkConfig, database: DatabaseInstance, task: BenchmarkTask):
"""Run the A/B performance task.
It executes the given pipeline for both overrides of Cost Model Coefficients,
then runs Student's t-test to prove a Null hypothesis that the means of the execution times
of both overrides are equal. If the p-value produced by the t-test is less than the threshold
value (usually 0.05 or 0.01) we can say that the Null hypothesis is proven and there is
no significant difference in the execution times.
"""
async with get_database_parameter(database, 'internalCostModelCoefficients') as db_param:
await db_param.set(json.dumps(task.cost_model_a.to_dict()))
result_a = await run(config, database, task.collection_name, task.pipeline)
await db_param.set(json.dumps(task.cost_model_b.to_dict()))
result_b = await run(config, database, task.collection_name, task.pipeline)
variant_a = make_variant(result_a)
variant_b = make_variant(result_b)
execution_times_a = [et.total_execution_time for et in variant_a.execution_tree]
execution_times_b = [et.total_execution_time for et in variant_b.execution_tree]
ttest_result = stats.ttest_ind(execution_times_a, execution_times_b, equal_var=False)
return BenchmarkResult(task=task, variant_a=variant_a, variant_b=variant_b,
pvalue=ttest_result.pvalue)
def make_variant(explain: Sequence[dict[str, any]]) -> ExperimentResult:
"""Make one variant of the A/B test."""
pt = [physical_tree.build(e['queryPlanner']['winningPlan']['optimizerPlan']) for e in explain]
et = [execution_tree.build_execution_tree(e['executionStats']) for e in explain]
mean = sum(et.total_execution_time for et in et) / len(et)
return ExperimentResult(explain=explain, physical_tree=pt, execution_tree=et, mean=mean)
async def run(config: BenchmarkConfig, database: DatabaseInstance, collection: str,
pipeline: Pipeline):
"""Run one variant of the A/B test."""
# warmup
for _ in range(config.warmup_runs):
await database.explain(collection, pipeline)
result = []
for _ in range(config.runs):
explain = await database.explain(collection, pipeline)
if explain['ok'] == 1:
result.append(explain)
else:
logging.warn('Query execution failed: %s', explain)
return result
async def smoke_test():
"""Smoke test and usage example function."""
from calibration_settings import main_config
config = BenchmarkConfig(warmup_runs=3, runs=20)
database = DatabaseInstance(main_config.database)
await database.enable_cascades(True)
cost_model_a = CostModelCoefficients(index_scan_incremental_cost=0.0001)
cost_model_b = CostModelCoefficients(index_scan_incremental_cost=0.9)
task = BenchmarkTask(collection_name='c_str_05_45000',
pipeline=[{'$match': {'choice1': 'hello', 'choice2': 'gaussian'}}],
cost_model_a=cost_model_a, cost_model_b=cost_model_b, threshold=0.05)
res = await benchmark(config, database, task)
res.print()
if __name__ == '__main__':
loop = asyncio.get_event_loop()
loop.run_until_complete(smoke_test())
|
