Optimize recompression for non-segmentby chunks (timescale#7632)

Enables the segmentwise recompression flow to be used for chunks without
segmentby columns.

This should be more performant than doing a full recompression.

.unreleased/pr_7632

@@ -0,0 +1 @@
+Implements: #7632 Optimize recompression for chunks without segmentby

tsl/src/compression/api.c

@@ -1066,22 +1066,9 @@ get_compressed_chunk_index_for_recompression(Chunk *uncompressed_chunk)
 
 	CompressionSettings *settings = ts_compression_settings_get(compressed_chunk->table_id);
 
-	// For chunks with no segmentby, we don't want to do segmentwise recompression as it is less
-	// performant than a full recompression. This is temporary; once we optimize recompression
-	// code for chunks with no segments we should remove this check.
-	int num_segmentby = ts_array_length(settings->fd.segmentby);
-
-	if (num_segmentby == 0)
-	{
-		table_close(compressed_chunk_rel, NoLock);
-		table_close(uncompressed_chunk_rel, NoLock);
-		return InvalidOid;
-	}
-
 	CatalogIndexState indstate = CatalogOpenIndexes(compressed_chunk_rel);
 	Oid index_oid = get_compressed_chunk_index(indstate, settings);
 	CatalogCloseIndexes(indstate);
-
 	table_close(compressed_chunk_rel, NoLock);
 	table_close(uncompressed_chunk_rel, NoLock);
 

tsl/src/compression/recompress.c

@@ -8,6 +8,9 @@
 #include <parser/parse_coerce.h>
 #include <parser/parse_relation.h>
 #include <utils/inval.h>
+#include <utils/lsyscache.h>
+#include <utils/rel.h>
+#include <utils/relcache.h>
 #include <utils/snapmgr.h>
 #include <utils/syscache.h>
 #include <utils/typcache.h>
@@ -210,6 +213,13 @@ recompress_chunk_segmentwise_impl(Chunk *uncompressed_chunk)
 						true /*need_bistate*/,
 						0 /*insert options*/);
 
+	/* For chunks with no segmentby settings, we can still do segmentwise recompression
+	 * The entire chunk is treated as a single segment
+	 */
+	elog(ts_guc_debug_compression_path_info ? INFO : DEBUG1,
+		 "Using index \"%s\" for recompression",
+		 get_rel_name(row_compressor.index_oid));
+
 	Relation index_rel = index_open(row_compressor.index_oid, ExclusiveLock);
 	ereport(DEBUG1,
 			(errmsg("locks acquired for recompression: \"%s.%s\"",

tsl/test/expected/compression.out

@@ -2794,12 +2794,12 @@ COPY compressed_table (time,a,b,c) FROM stdin;
 SELECT compress_chunk(i, if_not_compressed => true) FROM show_chunks('compressed_table') i;
               compress_chunk               
 -------------------------------------------
- _timescaledb_internal._hyper_49_108_chunk
+ _timescaledb_internal._hyper_49_107_chunk
 (1 row)
 
 \set ON_ERROR_STOP 0
 COPY compressed_table (time,a,b,c) FROM stdin;
-ERROR:  duplicate key value violates unique constraint "_hyper_49_108_chunk_compressed_table_index"
+ERROR:  duplicate key value violates unique constraint "_hyper_49_107_chunk_compressed_table_index"
 \set ON_ERROR_STOP 1
 COPY compressed_table (time,a,b,c) FROM stdin;
 SELECT * FROM compressed_table;
@@ -2813,7 +2813,7 @@ SELECT * FROM compressed_table;
 SELECT compress_chunk(i, if_not_compressed => true) FROM show_chunks('compressed_table') i;
               compress_chunk               
 -------------------------------------------
- _timescaledb_internal._hyper_49_108_chunk
+ _timescaledb_internal._hyper_49_107_chunk
 (1 row)
 
 -- Check DML decompression limit
@@ -2837,15 +2837,15 @@ NOTICE:  default order by for hypertable "hyper_84" is set to ""time" DESC"
 SELECT compress_chunk(ch) FROM show_chunks('hyper_84') ch;
               compress_chunk               
 -------------------------------------------
- _timescaledb_internal._hyper_51_110_chunk
+ _timescaledb_internal._hyper_51_109_chunk
 (1 row)
 
 -- indexscan for decompression: UPDATE
 UPDATE hyper_84 SET temp = 100 where device = 1;
 SELECT compress_chunk(ch) FROM show_chunks('hyper_84') ch;
               compress_chunk               
 -------------------------------------------
- _timescaledb_internal._hyper_51_110_chunk
+ _timescaledb_internal._hyper_51_109_chunk
 (1 row)
 
 -- indexscan for decompression: DELETE

tsl/test/expected/compression_ddl.out

@@ -2227,15 +2227,15 @@ EXPLAIN (COSTS OFF) SELECT * FROM space_part ORDER BY time;
  Custom Scan (ChunkAppend) on space_part
    Order: space_part."time"
    ->  Custom Scan (DecompressChunk) on _hyper_35_133_chunk
-         ->  Index Scan Backward using compress_hyper_36_139_chunk__ts_meta_min_1__ts_meta_max_1_idx on compress_hyper_36_139_chunk
+         ->  Index Scan Backward using compress_hyper_36_135_chunk__ts_meta_min_1__ts_meta_max_1_idx on compress_hyper_36_135_chunk
    ->  Custom Scan (DecompressChunk) on _hyper_35_134_chunk
          ->  Index Scan Backward using compress_hyper_36_136_chunk__ts_meta_min_1__ts_meta_max_1_idx on compress_hyper_36_136_chunk
    ->  Merge Append
          Sort Key: _hyper_35_137_chunk."time"
          ->  Custom Scan (DecompressChunk) on _hyper_35_137_chunk
-               ->  Index Scan Backward using compress_hyper_36_140_chunk__ts_meta_min_1__ts_meta_max_1_idx on compress_hyper_36_140_chunk
+               ->  Index Scan Backward using compress_hyper_36_139_chunk__ts_meta_min_1__ts_meta_max_1_idx on compress_hyper_36_139_chunk
          ->  Custom Scan (DecompressChunk) on _hyper_35_138_chunk
-               ->  Index Scan Backward using compress_hyper_36_141_chunk__ts_meta_min_1__ts_meta_max_1_idx on compress_hyper_36_141_chunk
+               ->  Index Scan Backward using compress_hyper_36_140_chunk__ts_meta_min_1__ts_meta_max_1_idx on compress_hyper_36_140_chunk
 (12 rows)
 
 -- make second one of them partial
@@ -2248,15 +2248,15 @@ EXPLAIN (COSTS OFF) SELECT * FROM space_part ORDER BY time;
  Custom Scan (ChunkAppend) on space_part
    Order: space_part."time"
    ->  Custom Scan (DecompressChunk) on _hyper_35_133_chunk
-         ->  Index Scan Backward using compress_hyper_36_139_chunk__ts_meta_min_1__ts_meta_max_1_idx on compress_hyper_36_139_chunk
+         ->  Index Scan Backward using compress_hyper_36_135_chunk__ts_meta_min_1__ts_meta_max_1_idx on compress_hyper_36_135_chunk
    ->  Custom Scan (DecompressChunk) on _hyper_35_134_chunk
          ->  Index Scan Backward using compress_hyper_36_136_chunk__ts_meta_min_1__ts_meta_max_1_idx on compress_hyper_36_136_chunk
    ->  Merge Append
          Sort Key: _hyper_35_137_chunk."time"
          ->  Custom Scan (DecompressChunk) on _hyper_35_137_chunk
-               ->  Index Scan Backward using compress_hyper_36_140_chunk__ts_meta_min_1__ts_meta_max_1_idx on compress_hyper_36_140_chunk
+               ->  Index Scan Backward using compress_hyper_36_139_chunk__ts_meta_min_1__ts_meta_max_1_idx on compress_hyper_36_139_chunk
          ->  Custom Scan (DecompressChunk) on _hyper_35_138_chunk
-               ->  Index Scan Backward using compress_hyper_36_141_chunk__ts_meta_min_1__ts_meta_max_1_idx on compress_hyper_36_141_chunk
+               ->  Index Scan Backward using compress_hyper_36_140_chunk__ts_meta_min_1__ts_meta_max_1_idx on compress_hyper_36_140_chunk
          ->  Sort
                Sort Key: _hyper_35_138_chunk."time"
                ->  Seq Scan on _hyper_35_138_chunk
@@ -2271,18 +2271,18 @@ EXPLAIN (COSTS OFF) SELECT * FROM space_part ORDER BY time;
  Custom Scan (ChunkAppend) on space_part
    Order: space_part."time"
    ->  Custom Scan (DecompressChunk) on _hyper_35_133_chunk
-         ->  Index Scan Backward using compress_hyper_36_139_chunk__ts_meta_min_1__ts_meta_max_1_idx on compress_hyper_36_139_chunk
+         ->  Index Scan Backward using compress_hyper_36_135_chunk__ts_meta_min_1__ts_meta_max_1_idx on compress_hyper_36_135_chunk
    ->  Custom Scan (DecompressChunk) on _hyper_35_134_chunk
          ->  Index Scan Backward using compress_hyper_36_136_chunk__ts_meta_min_1__ts_meta_max_1_idx on compress_hyper_36_136_chunk
    ->  Merge Append
          Sort Key: _hyper_35_137_chunk."time"
          ->  Custom Scan (DecompressChunk) on _hyper_35_137_chunk
-               ->  Index Scan Backward using compress_hyper_36_140_chunk__ts_meta_min_1__ts_meta_max_1_idx on compress_hyper_36_140_chunk
+               ->  Index Scan Backward using compress_hyper_36_139_chunk__ts_meta_min_1__ts_meta_max_1_idx on compress_hyper_36_139_chunk
          ->  Sort
                Sort Key: _hyper_35_137_chunk."time"
                ->  Seq Scan on _hyper_35_137_chunk
          ->  Custom Scan (DecompressChunk) on _hyper_35_138_chunk
-               ->  Index Scan Backward using compress_hyper_36_141_chunk__ts_meta_min_1__ts_meta_max_1_idx on compress_hyper_36_141_chunk
+               ->  Index Scan Backward using compress_hyper_36_140_chunk__ts_meta_min_1__ts_meta_max_1_idx on compress_hyper_36_140_chunk
          ->  Sort
                Sort Key: _hyper_35_138_chunk."time"
                ->  Seq Scan on _hyper_35_138_chunk
@@ -2315,14 +2315,14 @@ values ('meter1', 1, 2.3, '2022-01-01'::timestamptz, '2022-01-01'::timestamptz),
 select compress_chunk(show_chunks('mytab'));
               compress_chunk               
 -------------------------------------------
- _timescaledb_internal._hyper_37_142_chunk
+ _timescaledb_internal._hyper_37_141_chunk
 (1 row)
 
 REINDEX TABLE mytab; -- should update index
 select decompress_chunk(show_chunks('mytab'));
              decompress_chunk              
 -------------------------------------------
- _timescaledb_internal._hyper_37_142_chunk
+ _timescaledb_internal._hyper_37_141_chunk
 (1 row)
 
 \set EXPLAIN 'EXPLAIN (costs off,timing off,summary off)'
@@ -2333,7 +2333,7 @@ set enable_indexscan = on;
 :EXPLAIN_ANALYZE select * from mytab where lower(col1::text) = 'meter1';
                                             QUERY PLAN                                            
 --------------------------------------------------------------------------------------------------
- Index Scan using _hyper_37_142_chunk_myidx_unique on _hyper_37_142_chunk (actual rows=3 loops=1)
+ Index Scan using _hyper_37_141_chunk_myidx_unique on _hyper_37_141_chunk (actual rows=3 loops=1)
    Index Cond: (lower((col1)::text) = 'meter1'::text)
 (2 rows)
 
@@ -2351,19 +2351,19 @@ WHERE (value > 2.4 AND value < 3);
 select compress_chunk(show_chunks('mytab'));
               compress_chunk               
 -------------------------------------------
- _timescaledb_internal._hyper_37_142_chunk
+ _timescaledb_internal._hyper_37_141_chunk
 (1 row)
 
 select decompress_chunk(show_chunks('mytab'));
              decompress_chunk              
 -------------------------------------------
- _timescaledb_internal._hyper_37_142_chunk
+ _timescaledb_internal._hyper_37_141_chunk
 (1 row)
 
 :EXPLAIN_ANALYZE SELECT * FROM mytab WHERE value BETWEEN 2.4 AND 2.8;
                                       QUERY PLAN                                       
 ---------------------------------------------------------------------------------------
- Seq Scan on _hyper_37_142_chunk (actual rows=1 loops=1)
+ Seq Scan on _hyper_37_141_chunk (actual rows=1 loops=1)
    Filter: ((value >= '2.4'::double precision) AND (value <= '2.8'::double precision))
    Rows Removed by Filter: 2
 (3 rows)
@@ -2414,28 +2414,28 @@ NOTICE:  default order by for hypertable "hyper_unique_deferred" is set to ""tim
 select compress_chunk(show_chunks('hyper_unique_deferred')); -- also worked fine before 2.11.0
               compress_chunk               
 -------------------------------------------
- _timescaledb_internal._hyper_40_146_chunk
+ _timescaledb_internal._hyper_40_145_chunk
 (1 row)
 
 select decompress_chunk(show_chunks('hyper_unique_deferred'));
              decompress_chunk              
 -------------------------------------------
- _timescaledb_internal._hyper_40_146_chunk
+ _timescaledb_internal._hyper_40_145_chunk
 (1 row)
 
 \set ON_ERROR_STOP 0
 begin; insert INTO hyper_unique_deferred values (1257987700000000000, 'dev1', 1); abort;
-ERROR:  new row for relation "_hyper_40_146_chunk" violates check constraint "hyper_unique_deferred_sensor_1_check"
+ERROR:  new row for relation "_hyper_40_145_chunk" violates check constraint "hyper_unique_deferred_sensor_1_check"
 \set ON_ERROR_STOP 1
 select compress_chunk(show_chunks('hyper_unique_deferred'));
               compress_chunk               
 -------------------------------------------
- _timescaledb_internal._hyper_40_146_chunk
+ _timescaledb_internal._hyper_40_145_chunk
 (1 row)
 
 \set ON_ERROR_STOP 0
 begin; insert INTO hyper_unique_deferred values (1257987700000000000, 'dev1', 1); abort;
-ERROR:  duplicate key value violates unique constraint "146_2_hyper_unique_deferred_time_key"
+ERROR:  duplicate key value violates unique constraint "145_2_hyper_unique_deferred_time_key"
 \set ON_ERROR_STOP 1
 -- tests chunks being compressed using different segmentby settings
 -- github issue #7102
@@ -2465,7 +2465,7 @@ FROM timescaledb_information.chunks
 WHERE hypertable_name = 'compression_drop' AND NOT is_compressed;
                 CHUNK_NAME                 
 -------------------------------------------
- _timescaledb_internal._hyper_42_151_chunk
+ _timescaledb_internal._hyper_42_150_chunk
 (1 row)
 
 -- try dropping column v0, should fail
@@ -2495,7 +2495,7 @@ ALTER TABLE test2 SET (
 );
 \set ON_ERROR_STOP 0
 INSERT INTO test2(ts,b,t) VALUES ('2024-11-18 18:04:51',99,'magic');
-ERROR:  null value in column "i" of relation "_hyper_44_180_chunk" violates not-null constraint
+ERROR:  null value in column "i" of relation "_hyper_44_179_chunk" violates not-null constraint
 \set ON_ERROR_STOP 1
 ALTER TABLE test2 ALTER COLUMN i DROP NOT NULL;
 INSERT INTO test2(ts,b,t) VALUES ('2024-11-18 18:04:51',99,'magic');
@@ -2543,7 +2543,7 @@ SELECT count(*) FROM test2 WHERE i IS NULL;
 
 \set ON_ERROR_STOP 0
 ALTER TABLE test2 ALTER COLUMN i SET NOT NULL;
-ERROR:  column "i" of relation "_hyper_44_181_chunk" contains null values
+ERROR:  column "i" of relation "_hyper_44_180_chunk" contains null values
 DELETE FROM test2 WHERE i IS NULL;
 SELECT count(*) FROM test2 WHERE i IS NULL;
  count 
@@ -2567,46 +2567,46 @@ INSERT INTO test_notnull VALUES ('2025-01-01',NULL,NULL);
 -- should fail since we have NULL value
 \set ON_ERROR_STOP 0
 ALTER TABLE test_notnull ALTER COLUMN value SET NOT NULL;
-ERROR:  column "value" of relation "_hyper_46_238_chunk" contains null values
+ERROR:  column "value" of relation "_hyper_46_237_chunk" contains null values
 ALTER TABLE test_notnull ALTER COLUMN device SET NOT NULL;
-ERROR:  column "device" of relation "_hyper_46_238_chunk" contains null values
+ERROR:  column "device" of relation "_hyper_46_237_chunk" contains null values
 \set ON_ERROR_STOP 1
 SELECT compress_chunk(show_chunks('test_notnull'));
               compress_chunk               
 -------------------------------------------
- _timescaledb_internal._hyper_46_238_chunk
+ _timescaledb_internal._hyper_46_237_chunk
 (1 row)
 
 -- should fail since we have NULL value
 \set ON_ERROR_STOP 0
 ALTER TABLE test_notnull ALTER COLUMN value SET NOT NULL;
-ERROR:  column "value" of relation "_hyper_46_238_chunk" contains null values
+ERROR:  column "value" of relation "_hyper_46_237_chunk" contains null values
 ALTER TABLE test_notnull ALTER COLUMN device SET NOT NULL;
-ERROR:  column "device" of relation "_hyper_46_238_chunk" contains null values
+ERROR:  column "device" of relation "_hyper_46_237_chunk" contains null values
 \set ON_ERROR_STOP 1
 UPDATE test_notnull SET value = 1;
 ALTER TABLE test_notnull ALTER COLUMN value SET NOT NULL;
 ALTER TABLE test_notnull ALTER COLUMN value DROP NOT NULL;
 SELECT compress_chunk(show_chunks('test_notnull'));
               compress_chunk               
 -------------------------------------------
- _timescaledb_internal._hyper_46_238_chunk
+ _timescaledb_internal._hyper_46_237_chunk
 (1 row)
 
 ALTER TABLE test_notnull ALTER COLUMN value SET NOT NULL;
 ALTER TABLE test_notnull ALTER COLUMN value DROP NOT NULL;
 -- device still has NULL
 \set ON_ERROR_STOP 0
 ALTER TABLE test_notnull ALTER COLUMN device SET NOT NULL;
-ERROR:  column "device" of relation "_hyper_46_238_chunk" contains null values
+ERROR:  column "device" of relation "_hyper_46_237_chunk" contains null values
 \set ON_ERROR_STOP 1
 UPDATE test_notnull SET device = 'd1';
 ALTER TABLE test_notnull ALTER COLUMN device SET NOT NULL;
 ALTER TABLE test_notnull ALTER COLUMN device DROP NOT NULL;
 SELECT compress_chunk(show_chunks('test_notnull'));
               compress_chunk               
 -------------------------------------------
- _timescaledb_internal._hyper_46_238_chunk
+ _timescaledb_internal._hyper_46_237_chunk
 (1 row)
 
 ALTER TABLE test_notnull ALTER COLUMN device SET NOT NULL;
@@ -2616,25 +2616,25 @@ ALTER TABLE test_notnull ALTER COLUMN device DROP NOT NULL;
 INSERT INTO test_notnull VALUES ('2025-01-01',NULL,NULL);
 \set ON_ERROR_STOP 0
 ALTER TABLE test_notnull ALTER COLUMN device SET NOT NULL;
-ERROR:  column "device" of relation "_hyper_46_238_chunk" contains null values
+ERROR:  column "device" of relation "_hyper_46_237_chunk" contains null values
 \set ON_ERROR_STOP 1
 -- NULL in compressed part only
 SELECT compress_chunk(show_chunks('test_notnull'));
               compress_chunk               
 -------------------------------------------
- _timescaledb_internal._hyper_46_238_chunk
+ _timescaledb_internal._hyper_46_237_chunk
 (1 row)
 
 INSERT INTO test_notnull VALUES ('2025-01-01','d1',2);
 \set ON_ERROR_STOP 0
 ALTER TABLE test_notnull ALTER COLUMN device SET NOT NULL;
-ERROR:  column "device" of relation "_hyper_46_238_chunk" contains null values
+ERROR:  column "device" of relation "_hyper_46_237_chunk" contains null values
 \set ON_ERROR_STOP 1
 -- test added columns and defaults
 ALTER TABLE test_notnull ADD COLUMN c1 int;
 \set ON_ERROR_STOP 0
 ALTER TABLE test_notnull ALTER COLUMN c1 SET NOT NULL;
-ERROR:  column "c1" of relation "_hyper_46_238_chunk" contains null values
+ERROR:  column "c1" of relation "_hyper_46_237_chunk" contains null values
 \set ON_ERROR_STOP 1
 ALTER TABLE test_notnull ADD COLUMN c2 int DEFAULT 42;
 ALTER TABLE test_notnull ALTER COLUMN c2 SET NOT NULL;
@@ -2643,12 +2643,12 @@ ALTER TABLE test_notnull ALTER COLUMN c2 DROP NOT NULL;
 UPDATE test_notnull SET c2 = NULL;
 \set ON_ERROR_STOP 0
 ALTER TABLE test_notnull ALTER COLUMN c2 SET NOT NULL;
-ERROR:  column "c2" of relation "_hyper_46_238_chunk" contains null values
+ERROR:  column "c2" of relation "_hyper_46_237_chunk" contains null values
 \set ON_ERROR_STOP 1
 SELECT compress_chunk(show_chunks('test_notnull'));
               compress_chunk               
 -------------------------------------------
- _timescaledb_internal._hyper_46_238_chunk
+ _timescaledb_internal._hyper_46_237_chunk
 (1 row)
 
 -- broken atm due to bug in default handling in compression