{{ message }}
Repository navigation
Expand file tree
/
Copy pathparallel-examples.sql
More file actions
393 lines (328 loc) · 10.7 KB
/
Copy pathparallel-examples.sql
File metadata and controls
393 lines (328 loc) · 10.7 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
-- ============================================================================
-- RuVector Parallel Query Execution Examples
-- ============================================================================
--
-- This file demonstrates how to use RuVector's parallel query execution
-- for high-performance vector similarity search in PostgreSQL.
-- ============================================================================
-- Setup
-- ============================================================================
-- Load the RuVector extension
CREATE EXTENSION IF NOT EXISTS ruvector;
-- Configure PostgreSQL for parallel execution
SET max_parallel_workers_per_gather = 4;
SET parallel_setup_cost = 1000;
SET parallel_tuple_cost = 0.1;
SET min_parallel_table_scan_size = '8MB';
-- Create a sample table with vector embeddings
CREATE TABLE embeddings (
id SERIAL PRIMARY KEY,
content TEXT,
embedding vector(768),
created_at TIMESTAMP DEFAULT NOW()
);
-- Insert sample data (simulating 100K embeddings)
-- In production, you would load real embeddings
INSERT INTO embeddings (content, embedding)
SELECT
'Document ' || i,
-- Generate random 768-dimensional vector
array_to_string(array_agg(random()::real), ',')::vector(768)
FROM generate_series(1, 100000) i,
generate_series(1, 768) j
GROUP BY i;
-- ============================================================================
-- Index Creation with Parallel-Safe Support
-- ============================================================================
-- Create HNSW index for L2 distance
CREATE INDEX embeddings_hnsw_l2_idx
ON embeddings
USING ruhnsw (embedding vector_l2_ops)
WITH (
m = 16, -- Connections per node
ef_construction = 64 -- Build-time quality
);
-- Create HNSW index for cosine distance
CREATE INDEX embeddings_hnsw_cosine_idx
ON embeddings
USING ruhnsw (embedding vector_cosine_ops)
WITH (
m = 16,
ef_construction = 64
);
-- ============================================================================
-- Basic Parallel Query Examples
-- ============================================================================
-- Example 1: Simple k-NN search with automatic parallelization
-- The query planner will automatically use parallel workers if beneficial
EXPLAIN (ANALYZE, BUFFERS, VERBOSE)
SELECT
id,
content,
embedding <-> '[0.1, 0.2, ...]'::vector(768) AS distance
FROM embeddings
ORDER BY distance
LIMIT 10;
-- Example 2: Larger k with parallel execution
SELECT
id,
content,
embedding <-> '[0.1, 0.2, ...]'::vector(768) AS distance
FROM embeddings
ORDER BY distance
LIMIT 100;
-- Example 3: Cosine distance search
SELECT
id,
content,
embedding <=> '[0.1, 0.2, ...]'::vector(768) AS distance
FROM embeddings
ORDER BY distance
LIMIT 50;
-- ============================================================================
-- Monitoring and Diagnostics
-- ============================================================================
-- Check parallel query capabilities
SELECT * FROM ruvector_parallel_info();
-- Estimate workers for a specific query
SELECT ruvector_estimate_workers(
pg_relation_size('embeddings_hnsw_l2_idx') / 8192, -- pages
(SELECT count(*) FROM embeddings), -- tuples
100, -- k
100 -- ef_search
) AS recommended_workers;
-- Explain how query will be parallelized
SELECT * FROM ruvector_explain_parallel(
'embeddings_hnsw_l2_idx',
100, -- k
100, -- ef_search
768 -- dimensions
);
-- Get parallel execution statistics
SELECT * FROM ruvector_parallel_stats();
-- ============================================================================
-- Performance Benchmarking
-- ============================================================================
-- Benchmark parallel vs sequential execution
SELECT * FROM ruvector_benchmark_parallel(
'embeddings',
'embedding',
'[0.1, 0.2, ...]'::vector(768),
100
);
-- Compare different worker counts
DO $$
DECLARE
workers INT;
start_time TIMESTAMP;
end_time TIMESTAMP;
duration INTERVAL;
BEGIN
CREATE TEMP TABLE benchmark_results (
workers INT,
duration_ms FLOAT
);
FOR workers IN 1..8 LOOP
-- Set worker count
EXECUTE 'SET max_parallel_workers_per_gather = ' || workers;
-- Run query and measure time
start_time := clock_timestamp();
PERFORM id
FROM embeddings
ORDER BY embedding <-> '[0.1, 0.2, ...]'::vector(768)
LIMIT 100;
end_time := clock_timestamp();
duration := end_time - start_time;
-- Record result
INSERT INTO benchmark_results
VALUES (workers, EXTRACT(EPOCH FROM duration) * 1000);
RAISE NOTICE 'Workers: %, Duration: %ms', workers, EXTRACT(EPOCH FROM duration) * 1000;
END LOOP;
-- Show results
SELECT * FROM benchmark_results ORDER BY workers;
END $$;
-- ============================================================================
-- Advanced Query Patterns
-- ============================================================================
-- Example 4: Filter + k-NN with parallel execution
EXPLAIN (ANALYZE)
SELECT
id,
content,
created_at,
embedding <-> '[0.1, 0.2, ...]'::vector(768) AS distance
FROM embeddings
WHERE created_at > NOW() - INTERVAL '7 days'
ORDER BY distance
LIMIT 50;
-- Example 5: Join with parallel execution
CREATE TABLE categories (
id SERIAL PRIMARY KEY,
name TEXT,
embedding vector(768)
);
-- Find similar documents across categories
SELECT
e.id,
e.content,
c.name AS category,
e.embedding <-> c.embedding AS distance
FROM embeddings e
CROSS JOIN LATERAL (
SELECT name, embedding
FROM categories
ORDER BY categories.embedding <-> e.embedding
LIMIT 1
) c
ORDER BY distance
LIMIT 100;
-- Example 6: Aggregate queries with parallel execution
SELECT
bucket,
count(*) AS doc_count,
avg(distance) AS avg_distance
FROM (
SELECT
width_bucket(
embedding <-> '[0.1, 0.2, ...]'::vector(768),
0, 2, 10
) AS bucket,
embedding <-> '[0.1, 0.2, ...]'::vector(768) AS distance
FROM embeddings
) sub
GROUP BY bucket
ORDER BY bucket;
-- ============================================================================
-- Background Worker Management
-- ============================================================================
-- Start background maintenance worker
SELECT ruvector_bgworker_start();
-- Check background worker status
SELECT * FROM ruvector_bgworker_status();
-- Configure background worker
SELECT ruvector_bgworker_config(
maintenance_interval_secs := 300, -- 5 minutes
auto_optimize := true,
collect_stats := true,
auto_vacuum := true
);
-- Stop background worker
-- SELECT ruvector_bgworker_stop();
-- ============================================================================
-- Configuration Tuning
-- ============================================================================
-- Configure parallel execution behavior
SELECT ruvector_set_parallel_config(
enable := true,
min_tuples_for_parallel := 10000,
min_pages_for_parallel := 100
);
-- Adjust HNSW search parameters
SET ruvector.ef_search = 100; -- Higher = better recall, slower
-- Adjust PostgreSQL parallel query costs
SET parallel_setup_cost = 500; -- Lower = more likely to parallelize
SET parallel_tuple_cost = 0.05; -- Lower = favor parallel execution
-- ============================================================================
-- Query Plan Analysis
-- ============================================================================
-- Analyze query plan with parallel workers
EXPLAIN (ANALYZE, BUFFERS, VERBOSE, COSTS, TIMING)
SELECT
id,
embedding <-> '[0.1, 0.2, ...]'::vector(768) AS distance
FROM embeddings
ORDER BY distance
LIMIT 100;
-- Compare with forced sequential execution
SET max_parallel_workers_per_gather = 0;
EXPLAIN (ANALYZE)
SELECT
id,
embedding <-> '[0.1, 0.2, ...]'::vector(768) AS distance
FROM embeddings
ORDER BY distance
LIMIT 100;
-- Reset to parallel
SET max_parallel_workers_per_gather = 4;
-- ============================================================================
-- Production Best Practices
-- ============================================================================
-- 1. Create indexes with appropriate parameters
CREATE INDEX CONCURRENTLY embeddings_hnsw_idx
ON embeddings
USING ruhnsw (embedding vector_l2_ops)
WITH (
m = 16,
ef_construction = 64
);
-- 2. Analyze table statistics
ANALYZE embeddings;
-- 3. Monitor query performance
CREATE EXTENSION IF NOT EXISTS pg_stat_statements;
SELECT
query,
calls,
mean_exec_time,
total_exec_time,
rows
FROM pg_stat_statements
WHERE query LIKE '%<->%'
ORDER BY mean_exec_time DESC
LIMIT 10;
-- 4. Check index usage
SELECT
schemaname,
tablename,
indexname,
idx_scan,
idx_tup_read,
idx_tup_fetch
FROM pg_stat_user_indexes
WHERE indexname LIKE '%hnsw%';
-- 5. Monitor memory usage
SELECT
pid,
backend_type,
pg_size_pretty(pg_backend_memory_contexts()) as memory_context
FROM pg_stat_activity
WHERE backend_type LIKE 'parallel%';
-- ============================================================================
-- Performance Testing Queries
-- ============================================================================
-- Test 1: Small k (should be fast even without parallelism)
\timing on
SELECT id, embedding <-> '[0.1, 0.2, ...]'::vector(768) AS distance
FROM embeddings
ORDER BY distance
LIMIT 10;
-- Test 2: Medium k (benefits from parallelism)
SELECT id, embedding <-> '[0.1, 0.2, ...]'::vector(768) AS distance
FROM embeddings
ORDER BY distance
LIMIT 100;
-- Test 3: Large k (maximum benefit from parallelism)
SELECT id, embedding <-> '[0.1, 0.2, ...]'::vector(768) AS distance
FROM embeddings
ORDER BY distance
LIMIT 1000;
\timing off
-- ============================================================================
-- Cleanup
-- ============================================================================
-- Drop temporary tables
DROP TABLE IF EXISTS benchmark_results;
-- Optionally drop the sample table
-- DROP TABLE IF EXISTS embeddings CASCADE;
-- DROP TABLE IF EXISTS categories CASCADE;
-- ============================================================================
-- Additional Functions
-- ============================================================================
-- Get RuVector version and capabilities
SELECT ruvector_version();
SELECT ruvector_simd_info();
-- Get memory statistics
SELECT * FROM ruvector_memory_stats();
-- Get index information
SELECT * FROM ruhnsw_index_info('embeddings_hnsw_l2_idx');
-- Perform manual index maintenance
SELECT ruvector_index_maintenance('embeddings_hnsw_l2_idx');
You can’t perform that action at this time.