Fix segementation fault during INSERT into compressed hypertable.

INSERT into compressed hypertable with number of open chunks greater
than ts_guc_max_open_chunks_per_insert causes segementation fault.
New row which needs to be inserted into compressed chunk has to be
compressed. Memory required as part of compressing a row is allocated
from RowCompressor::per_row_ctx memory context. Once row is compressed,
ExecInsert() is called, where memory from same context is used to
allocate and free it instead of using "Executor State". This causes
a corruption in memory.

Fixes: timescale#4778

CHANGELOG.md

@@ -24,6 +24,7 @@ argument or resolve the type ambiguity by casting to the intended type.
 * #4696 Report warning when enabling compression on hypertable
 * #4745 Fix FK constraint violation error while insert into hypertable which references partitioned table
 * #4756 Improve compression job IO performance
+* #4807 Fix segmentation fault during INSERT into compressed hypertable.
 
 **Thanks**
 * @jvanns for reporting hypertable FK reference to vanilla PostgreSQL partitioned table doesn't seem to work

tsl/src/compression/compression.c

@@ -707,10 +707,12 @@ static void
 row_compressor_update_group(RowCompressor *row_compressor, TupleTableSlot *row)
 {
 	int col;
-
+	/* save original memory context */
+	const MemoryContext oldcontext = CurrentMemoryContext;
 	Assert(row_compressor->rows_compressed_into_current_value == 0);
 	Assert(row_compressor->n_input_columns <= row->tts_nvalid);
 
+	MemoryContextSwitchTo(row_compressor->per_row_ctx->parent);
 	for (col = 0; col < row_compressor->n_input_columns; col++)
 	{
 		PerColumn *column = &row_compressor->per_column[col];
@@ -722,13 +724,13 @@ row_compressor_update_group(RowCompressor *row_compressor, TupleTableSlot *row)
 
 		Assert(column->compressor == NULL);
 
-		MemoryContextSwitchTo(row_compressor->per_row_ctx->parent);
 		/* Performance Improvment: We should just use array access here; everything is guaranteed to
 		   be fetched */
 		val = slot_getattr(row, AttrOffsetGetAttrNumber(col), &is_null);
 		segment_info_update(column->segment_info, val, is_null);
-		MemoryContextSwitchTo(row_compressor->per_row_ctx);
 	}
+	/* switch to original memory context */
+	MemoryContextSwitchTo(oldcontext);
 }
 
 static bool

tsl/test/shared/expected/compression_dml.out

@@ -24,3 +24,42 @@ EXECUTE p2('2021-02-22T08:00:00+00');
 DEALLOCATE p1;
 DEALLOCATE p2;
 DROP TABLE i3719;
+-- github issue 4778
+CREATE TABLE metric_5m (
+    time TIMESTAMPTZ NOT NULL,
+    value DOUBLE PRECISION NOT NULL,
+    series_id BIGINT NOT NULL
+);
+SELECT table_name FROM create_hypertable(
+                            'metric_5m'::regclass,
+                            'time'::name, chunk_time_interval=>interval '5m',
+                            create_default_indexes=> false);
+ table_name 
+ metric_5m
+(1 row)
+
+-- enable compression
+ALTER TABLE metric_5m SET (
+    timescaledb.compress,
+    timescaledb.compress_segmentby = 'series_id',
+    timescaledb.compress_orderby = 'time, value'
+);
+SET work_mem TO '64kB';
+SELECT '2022-10-10 14:33:44.1234+05:30' as start_date \gset
+-- populate hypertable
+INSERT INTO metric_5m (time, series_id, value)
+    SELECT t, s,1 from generate_series(:'start_date'::timestamptz, :'start_date'::timestamptz + interval '1 day', '10s') t cross join generate_series(1,10, 1) s;
+-- manually compress all chunks
+SELECT count(compress_chunk(c.schema_name|| '.' || c.table_name))
+    FROM _timescaledb_catalog.chunk c, _timescaledb_catalog.hypertable ht where
+    c.hypertable_id = ht.id and ht.table_name = 'metric_5m' and c.compressed_chunk_id IS NULL;
+ count 
+   289
+(1 row)
+
+-- populate into compressed hypertable, this should not crash
+INSERT INTO metric_5m (time, series_id, value)
+    SELECT t, s,1 from generate_series(:'start_date'::timestamptz, :'start_date'::timestamptz + interval '1 day', '10s') t cross join generate_series(1,10, 1) s;
+-- clean up
+RESET work_mem;
+DROP TABLE metric_5m;

tsl/test/shared/sql/compression_dml.sql

@@ -21,3 +21,36 @@ DEALLOCATE p1;
 DEALLOCATE p2;
 
 DROP TABLE i3719;
+
+-- github issue 4778
+CREATE TABLE metric_5m (
+    time TIMESTAMPTZ NOT NULL,
+    value DOUBLE PRECISION NOT NULL,
+    series_id BIGINT NOT NULL
+);
+SELECT table_name FROM create_hypertable(
+                            'metric_5m'::regclass,
+                            'time'::name, chunk_time_interval=>interval '5m',
+                            create_default_indexes=> false);
+-- enable compression
+ALTER TABLE metric_5m SET (
+    timescaledb.compress,
+    timescaledb.compress_segmentby = 'series_id',
+    timescaledb.compress_orderby = 'time, value'
+);
+SET work_mem TO '64kB';
+SELECT '2022-10-10 14:33:44.1234+05:30' as start_date \gset
+-- populate hypertable
+INSERT INTO metric_5m (time, series_id, value)
+    SELECT t, s,1 from generate_series(:'start_date'::timestamptz, :'start_date'::timestamptz + interval '1 day', '10s') t cross join generate_series(1,10, 1) s;
+-- manually compress all chunks
+SELECT count(compress_chunk(c.schema_name|| '.' || c.table_name))
+    FROM _timescaledb_catalog.chunk c, _timescaledb_catalog.hypertable ht where
+    c.hypertable_id = ht.id and ht.table_name = 'metric_5m' and c.compressed_chunk_id IS NULL;
+
+-- populate into compressed hypertable, this should not crash
+INSERT INTO metric_5m (time, series_id, value)
+    SELECT t, s,1 from generate_series(:'start_date'::timestamptz, :'start_date'::timestamptz + interval '1 day', '10s') t cross join generate_series(1,10, 1) s;
+-- clean up
+RESET work_mem;
+DROP TABLE metric_5m;