Account for CPU tuple cost in DecompressChunk

It was always zero due to a typo.
timescale · Dec 19, 2024 · 025dc44 · 025dc44
1 parent 155efa1
commit 025dc44
Show file tree

Hide file tree

Showing 13 changed files with 350 additions and 240 deletions.
diff --git a/tsl/src/nodes/decompress_chunk/decompress_chunk.c b/tsl/src/nodes/decompress_chunk/decompress_chunk.c
@@ -430,12 +430,14 @@ static void
 cost_decompress_chunk(PlannerInfo *root, Path *path, Path *compressed_path)
 {
 	/* startup_cost is cost before fetching first tuple */
-	if (compressed_path->rows > 0)
-		path->startup_cost = compressed_path->total_cost / compressed_path->rows;
+	const double compressed_rows = Max(1, compressed_path->rows);
+	path->startup_cost =
+		compressed_path->startup_cost +
+		(compressed_path->total_cost - compressed_path->startup_cost) / compressed_rows;
 
 	/* total_cost is cost for fetching all tuples */
+	path->rows = compressed_rows * TARGET_COMPRESSED_BATCH_SIZE;
 	path->total_cost = compressed_path->total_cost + path->rows * cpu_tuple_cost;
-	path->rows = compressed_path->rows * TARGET_COMPRESSED_BATCH_SIZE;
 }
 
 /* Smoothstep function S1 (the h01 cubic Hermite spline). */

diff --git a/tsl/test/expected/compress_auto_sparse_index.out b/tsl/test/expected/compress_auto_sparse_index.out
@@ -23,12 +23,12 @@ select count(compress_chunk(x)) from show_chunks('sparse') x;
      1
 (1 row)
 
-explain select * from sparse where value = 1;
+explain (costs off) select * from sparse where value = 1;
                                                        QUERY PLAN                                                        
 -------------------------------------------------------------------------------------------------------------------------
- Custom Scan (DecompressChunk) on _hyper_1_1_chunk  (cost=0.27..20.20 rows=76000 width=12)
+ Custom Scan (DecompressChunk) on _hyper_1_1_chunk
    Vectorized Filter: (value = '1'::double precision)
-   ->  Seq Scan on compress_hyper_2_2_chunk  (cost=0.00..20.20 rows=76 width=76)
+   ->  Seq Scan on compress_hyper_2_2_chunk
          Filter: ((_ts_meta_v2_min_value <= '1'::double precision) AND (_ts_meta_v2_max_value >= '1'::double precision))
 (4 rows)
 
@@ -40,12 +40,12 @@ select count(compress_chunk(decompress_chunk(x))) from show_chunks('sparse') x;
      1
 (1 row)
 
-explain select * from sparse where value = 1;
-                                         QUERY PLAN                                         
---------------------------------------------------------------------------------------------
- Custom Scan (DecompressChunk) on _hyper_1_1_chunk  (cost=0.02..17.80 rows=780000 width=12)
+explain (costs off) select * from sparse where value = 1;
+                      QUERY PLAN                      
+------------------------------------------------------
+ Custom Scan (DecompressChunk) on _hyper_1_1_chunk
    Vectorized Filter: (value = '1'::double precision)
-   ->  Seq Scan on compress_hyper_2_3_chunk  (cost=0.00..17.80 rows=780 width=76)
+   ->  Seq Scan on compress_hyper_2_3_chunk
 (3 rows)
 
 reset timescaledb.auto_sparse_indexes;
@@ -55,33 +55,33 @@ select count(compress_chunk(decompress_chunk(x))) from show_chunks('sparse') x;
      1
 (1 row)
 
-explain select * from sparse where value = 1;
+explain (costs off) select * from sparse where value = 1;
                                                        QUERY PLAN                                                        
 -------------------------------------------------------------------------------------------------------------------------
- Custom Scan (DecompressChunk) on _hyper_1_1_chunk  (cost=0.27..20.20 rows=76000 width=12)
+ Custom Scan (DecompressChunk) on _hyper_1_1_chunk
    Vectorized Filter: (value = '1'::double precision)
-   ->  Seq Scan on compress_hyper_2_4_chunk  (cost=0.00..20.20 rows=76 width=76)
+   ->  Seq Scan on compress_hyper_2_4_chunk
          Filter: ((_ts_meta_v2_min_value <= '1'::double precision) AND (_ts_meta_v2_max_value >= '1'::double precision))
 (4 rows)
 
 -- Should survive renames.
 alter table sparse rename column value to wert;
-explain select * from sparse where wert = 1;
+explain (costs off) select * from sparse where wert = 1;
                                                       QUERY PLAN                                                       
 -----------------------------------------------------------------------------------------------------------------------
- Custom Scan (DecompressChunk) on _hyper_1_1_chunk  (cost=0.27..20.20 rows=76000 width=12)
+ Custom Scan (DecompressChunk) on _hyper_1_1_chunk
    Vectorized Filter: (wert = '1'::double precision)
-   ->  Seq Scan on compress_hyper_2_4_chunk  (cost=0.00..20.20 rows=76 width=76)
+   ->  Seq Scan on compress_hyper_2_4_chunk
          Filter: ((_ts_meta_v2_min_wert <= '1'::double precision) AND (_ts_meta_v2_max_wert >= '1'::double precision))
 (4 rows)
 
 alter table sparse rename column wert to value;
-explain select * from sparse where value = 1;
+explain (costs off) select * from sparse where value = 1;
                                                        QUERY PLAN                                                        
 -------------------------------------------------------------------------------------------------------------------------
- Custom Scan (DecompressChunk) on _hyper_1_1_chunk  (cost=0.27..20.20 rows=76000 width=12)
+ Custom Scan (DecompressChunk) on _hyper_1_1_chunk
    Vectorized Filter: (value = '1'::double precision)
-   ->  Seq Scan on compress_hyper_2_4_chunk  (cost=0.00..20.20 rows=76 width=76)
+   ->  Seq Scan on compress_hyper_2_4_chunk
          Filter: ((_ts_meta_v2_min_value <= '1'::double precision) AND (_ts_meta_v2_max_value >= '1'::double precision))
 (4 rows)
 
@@ -94,12 +94,12 @@ select count(compress_chunk(decompress_chunk(x))) from show_chunks('sparse') x;
      1
 (1 row)
 
-explain select * from sparse where value = 1;
-                                         QUERY PLAN                                         
---------------------------------------------------------------------------------------------
- Custom Scan (DecompressChunk) on _hyper_1_1_chunk  (cost=0.02..17.80 rows=780000 width=12)
+explain (costs off) select * from sparse where value = 1;
+                      QUERY PLAN                      
+------------------------------------------------------
+ Custom Scan (DecompressChunk) on _hyper_1_1_chunk
    Vectorized Filter: (value = '1'::double precision)
-   ->  Seq Scan on compress_hyper_2_5_chunk  (cost=0.00..17.80 rows=780 width=76)
+   ->  Seq Scan on compress_hyper_2_5_chunk
 (3 rows)
 
 -- Not for other index types.
@@ -111,12 +111,12 @@ select count(compress_chunk(decompress_chunk(x))) from show_chunks('sparse') x;
      1
 (1 row)
 
-explain select * from sparse where value = 1;
-                                         QUERY PLAN                                         
---------------------------------------------------------------------------------------------
- Custom Scan (DecompressChunk) on _hyper_1_1_chunk  (cost=0.02..17.80 rows=780000 width=12)
+explain (costs off) select * from sparse where value = 1;
+                      QUERY PLAN                      
+------------------------------------------------------
+ Custom Scan (DecompressChunk) on _hyper_1_1_chunk
    Vectorized Filter: (value = '1'::double precision)
-   ->  Seq Scan on compress_hyper_2_6_chunk  (cost=0.00..17.80 rows=780 width=76)
+   ->  Seq Scan on compress_hyper_2_6_chunk
 (3 rows)
 
 -- When the chunk is recompressed without index, no sparse index is created.
@@ -127,12 +127,12 @@ select count(compress_chunk(decompress_chunk(x))) from show_chunks('sparse') x;
      1
 (1 row)
 
-explain select * from sparse where value = 1;
-                                         QUERY PLAN                                         
---------------------------------------------------------------------------------------------
- Custom Scan (DecompressChunk) on _hyper_1_1_chunk  (cost=0.02..17.80 rows=780000 width=12)
+explain (costs off) select * from sparse where value = 1;
+                      QUERY PLAN                      
+------------------------------------------------------
+ Custom Scan (DecompressChunk) on _hyper_1_1_chunk
    Vectorized Filter: (value = '1'::double precision)
-   ->  Seq Scan on compress_hyper_2_7_chunk  (cost=0.00..17.80 rows=780 width=76)
+   ->  Seq Scan on compress_hyper_2_7_chunk
 (3 rows)
 
 -- Long column names.
@@ -149,12 +149,12 @@ select count(compress_chunk(x)) from show_chunks('sparse') x;
      1
 (1 row)
 
-explain select * from sparse where Abcdef012345678_Bbcdef012345678_Cbcdef012345678_Dbcdef0 = 1;
+explain (costs off) select * from sparse where Abcdef012345678_Bbcdef012345678_Cbcdef012345678_Dbcdef0 = 1;
                                                                           QUERY PLAN                                                                           
 ---------------------------------------------------------------------------------------------------------------------------------------------------------------
- Custom Scan (DecompressChunk) on _hyper_1_1_chunk  (cost=3.48..10.45 rows=3000 width=264)
+ Custom Scan (DecompressChunk) on _hyper_1_1_chunk
    Vectorized Filter: (abcdef012345678_bbcdef012345678_cbcdef012345678_dbcdef0 = 1)
-   ->  Seq Scan on compress_hyper_2_8_chunk  (cost=0.00..10.45 rows=3 width=2092)
+   ->  Seq Scan on compress_hyper_2_8_chunk
          Filter: ((_ts_meta_v2_min_9218_abcdef012345678_bbcdef012345678_cbcdef0 <= 1) AND (_ts_meta_v2_max_9218_abcdef012345678_bbcdef012345678_cbcdef0 >= 1))
 (4 rows)
 
diff --git a/tsl/test/expected/compression.out b/tsl/test/expected/compression.out
@@ -1859,17 +1859,17 @@ SELECT sum(cpu) FROM f_sensor_data;
          Output: (PARTIAL sum(_hyper_37_73_chunk.cpu))
          Workers Planned: 4
          ->  Parallel Append
-               ->  Partial Aggregate
-                     Output: PARTIAL sum(_hyper_37_73_chunk.cpu)
-                     ->  Parallel Seq Scan on _timescaledb_internal._hyper_37_73_chunk
-                           Output: _hyper_37_73_chunk.cpu
                ->  Custom Scan (VectorAgg)
                      Output: (PARTIAL sum(_hyper_37_73_chunk.cpu))
                      Grouping Policy: all compressed batches
                      ->  Custom Scan (DecompressChunk) on _timescaledb_internal._hyper_37_73_chunk
                            Output: _hyper_37_73_chunk.cpu
                            ->  Parallel Seq Scan on _timescaledb_internal.compress_hyper_38_74_chunk
                                  Output: compress_hyper_38_74_chunk._ts_meta_count, compress_hyper_38_74_chunk.sensor_id, compress_hyper_38_74_chunk._ts_meta_min_1, compress_hyper_38_74_chunk._ts_meta_max_1, compress_hyper_38_74_chunk."time", compress_hyper_38_74_chunk.cpu, compress_hyper_38_74_chunk.temperature
+               ->  Partial Aggregate
+                     Output: PARTIAL sum(_hyper_37_73_chunk.cpu)
+                     ->  Parallel Seq Scan on _timescaledb_internal._hyper_37_73_chunk
+                           Output: _hyper_37_73_chunk.cpu
 (17 rows)
 
 :explain
@@ -1880,15 +1880,15 @@ SELECT * FROM f_sensor_data WHERE sensor_id > 100;
    Output: _hyper_37_73_chunk."time", _hyper_37_73_chunk.sensor_id, _hyper_37_73_chunk.cpu, _hyper_37_73_chunk.temperature
    Workers Planned: 3
    ->  Parallel Append
-         ->  Parallel Index Scan using _hyper_37_73_chunk_f_sensor_data_time_sensor_id_idx on _timescaledb_internal._hyper_37_73_chunk
-               Output: _hyper_37_73_chunk."time", _hyper_37_73_chunk.sensor_id, _hyper_37_73_chunk.cpu, _hyper_37_73_chunk.temperature
-               Index Cond: (_hyper_37_73_chunk.sensor_id > 100)
          ->  Custom Scan (DecompressChunk) on _timescaledb_internal._hyper_37_73_chunk
                Output: _hyper_37_73_chunk."time", _hyper_37_73_chunk.sensor_id, _hyper_37_73_chunk.cpu, _hyper_37_73_chunk.temperature
                Filter: (_hyper_37_73_chunk.sensor_id > 100)
                ->  Parallel Index Scan using compress_hyper_38_74_chunk_sensor_id__ts_meta_min_1__ts_met_idx on _timescaledb_internal.compress_hyper_38_74_chunk
                      Output: compress_hyper_38_74_chunk._ts_meta_count, compress_hyper_38_74_chunk.sensor_id, compress_hyper_38_74_chunk._ts_meta_min_1, compress_hyper_38_74_chunk._ts_meta_max_1, compress_hyper_38_74_chunk."time", compress_hyper_38_74_chunk.cpu, compress_hyper_38_74_chunk.temperature
                      Index Cond: (compress_hyper_38_74_chunk.sensor_id > 100)
+         ->  Parallel Index Scan using _hyper_37_73_chunk_f_sensor_data_time_sensor_id_idx on _timescaledb_internal._hyper_37_73_chunk
+               Output: _hyper_37_73_chunk."time", _hyper_37_73_chunk.sensor_id, _hyper_37_73_chunk.cpu, _hyper_37_73_chunk.temperature
+               Index Cond: (_hyper_37_73_chunk.sensor_id > 100)
 (13 rows)
 
 -- Test non-partial paths below append are not executed multiple times
@@ -2802,7 +2802,7 @@ COPY compressed_table (time,a,b,c) FROM stdin;
 ERROR:  duplicate key value violates unique constraint "_hyper_49_108_chunk_compressed_table_index"
 \set ON_ERROR_STOP 1
 COPY compressed_table (time,a,b,c) FROM stdin;
-SELECT * FROM compressed_table;
+SELECT * FROM compressed_table ORDER BY a;
                 time                | a  | b | c 
 ------------------------------------+----+---+---
  Thu Feb 29 01:00:00 2024 PST       |  5 | 1 | 1

diff --git a/tsl/test/expected/compression_qualpushdown.out b/tsl/test/expected/compression_qualpushdown.out
@@ -292,14 +292,19 @@ SELECT compress_chunk(i) FROM show_chunks('deleteme') i;
  _timescaledb_internal._hyper_7_8_chunk
 (1 row)
 
+VACUUM ANALYZE deleteme;
 EXPLAIN (costs off) SELECT sum(data) FROM deleteme WHERE segment::text like '%4%';
-                       QUERY PLAN                        
----------------------------------------------------------
- Aggregate
-   ->  Custom Scan (DecompressChunk) on _hyper_7_8_chunk
-         ->  Seq Scan on compress_hyper_8_9_chunk
-               Filter: ((segment)::text ~~ '%4%'::text)
-(4 rows)
+                                 QUERY PLAN                                  
+-----------------------------------------------------------------------------
+ Finalize Aggregate
+   ->  Gather
+         Workers Planned: 1
+         ->  Partial Aggregate
+               ->  Parallel Append
+                     ->  Custom Scan (DecompressChunk) on _hyper_7_8_chunk
+                           ->  Parallel Seq Scan on compress_hyper_8_9_chunk
+                                 Filter: ((segment)::text ~~ '%4%'::text)
+(8 rows)
 
 EXPLAIN (costs off) SELECT sum(data) FROM deleteme WHERE '4' = segment::text;
                        QUERY PLAN