Fix: test parallel_schedule with greenplum_schedule

src/backend/utils/activity/pgstat.c

@@ -1772,6 +1772,7 @@ assign_stats_fetch_consistency(int newval, void *extra)
 void
 pgstat_send_qd_tabstats(void)
 {
+	int					nest_level;
 	PgStat_SubXactStatus *xact_state;
 	StringInfoData		buf;
 	PgStatTabRecordFromQE *records;
@@ -1788,10 +1789,14 @@ pgstat_send_qd_tabstats(void)
 	 * current nesting level's per-table insert/update/delete counts.
 	 */
 	xact_state = pgstat_get_current_xact_stack();
-
 	if (xact_state == NULL)
 		return;
 
+	nest_level = GetCurrentTransactionNestLevel();
+
+	if (nest_level != xact_state->nest_level)
+		return;
+
 	records = (PgStatTabRecordFromQE *)
 		palloc(capacity * sizeof(PgStatTabRecordFromQE));
 

src/bin/pg_basebackup/bbstreamer_file.c

@@ -323,12 +323,15 @@ extract_directory(const char *filename, mode_t mode)
 		 * pg_xlog we assume is deleted at the start of
 		 * pg_basebackup. We cannot delete pg_xlog because if
 		 * streammode was used then it may have already copied
-		 * new xlog files into pg_xlog directory.
+		 * new xlog files into pg_xlog directory. Similarly,
+		 * archive_status may have .done files created by the
+		 * WAL streamer child process via mark_file_as_archived().
 		 */
 		if (pg_str_endswith(filename, "/pg_log") ||
 			pg_str_endswith(filename, "/log") ||
 			pg_str_endswith(filename, "/pg_wal") ||
-			pg_str_endswith(filename, "/pg_xlog"))
+			pg_str_endswith(filename, "/pg_xlog") ||
+			pg_str_endswith(filename, "/archive_status"))
 			return;
 
 		rmtree(filename, true);

src/test/isolation2/Makefile

@@ -29,7 +29,7 @@ test_parallel_retrieve_cursor_extended_query_error: test_parallel_retrieve_curso
 	$(CC) $(CPPFLAGS) $(rpath) -I$(top_builddir)/src/interfaces/libpq -L$(GPHOME)/lib -L$(top_builddir)/src/interfaces/libpq  -o $@ $< -lpq
 
 pg_regress.o:
-	$(MAKE) -C $(top_builddir)/src/test/regress pg_regress.o
+	$(MAKE) -C $(top_builddir)/src/test/regress install
 	rm -f $@ && $(LN_S) $(top_builddir)/src/test/regress/pg_regress.o .
 
 gpstringsubs.pl:

src/test/isolation2/expected/bitmap_index_crash.out

@@ -111,7 +111,7 @@ select count(*) from test_bitmap where type = 520;
 (1 row)
 
 -- start_ignore
-! gpstop -rai;
+! gpstop -raf;
 -- end_ignore
 
 0: select count(*) from test_bitmap where type = 520;

src/test/isolation2/expected/gdd/end.out

@@ -6,7 +6,7 @@ ALTER
 -- Use utility session on seg 0 to restart master. This way avoids the
 -- situation where session issuing the restart doesn't disappear
 -- itself.
-1U:SELECT pg_ctl(dir, 'restart') from datadir;
+1U:SELECT pg_ctl(dir, 'restart', 'fast') from datadir;
  pg_ctl 
 --------
  OK     

src/test/isolation2/expected/gdd/prepare.out

@@ -75,7 +75,7 @@ ALTER
 -- Use utility session on seg 0 to restart master. This way avoids the
 -- situation where session issuing the restart doesn't disappear
 -- itself.
-1U:SELECT pg_ctl(dir, 'restart') from datadir;
+1U:SELECT pg_ctl(dir, 'restart', 'fast') from datadir;
  pg_ctl 
 --------
  OK     

src/test/isolation2/expected/instr_in_shmem_setup.out

@@ -1,3 +1,3 @@
 -- start_ignore
-! gpconfig -c gp_enable_query_metrics -v on; ! gpstop -rai;
+! gpconfig -c gp_enable_query_metrics -v on; ! gpstop -raf;
 -- end_ignore

src/test/isolation2/expected/pg_rewind_fail_missing_xlog.out

@@ -19,7 +19,7 @@ INSERT 3
 
 -- end_ignore
 (exited with code 0)
-!\retcode gpstop -ari;
+!\retcode gpstop -arf;
 (exited with code 0)
 
 -- Test 1: Ensure that pg_rewind doesn't fail due to checkpoints inadvertently
@@ -543,14 +543,40 @@ server closed the connection unexpectedly
 
 -- end_ignore
 (exited with code 0)
-!\retcode gpstop -ari;
+!\retcode gpstop -arf;
 -- start_ignore
 -- end_ignore
 (exited with code 0)
 
 -- Create an unlogged table on the primary that remembers replication slot's last restart_lsn and number of WAL files.
 1U: CREATE UNLOGGED TABLE unlogged_wal_retention_test(restart_lsn_before pg_lsn, wal_count_before int);
 CREATE
+-- Force WAL segment switches to ensure there are old WAL files that can be removed after checkpoint.
+1U: SELECT pg_switch_wal() IS NOT NULL;
+ ?column? 
+----------
+ t        
+(1 row)
+1U: SELECT pg_switch_wal() IS NOT NULL;
+ ?column? 
+----------
+ t        
+(1 row)
+1U: SELECT pg_switch_wal() IS NOT NULL;
+ ?column? 
+----------
+ t        
+(1 row)
+1U: SELECT pg_switch_wal() IS NOT NULL;
+ ?column? 
+----------
+ t        
+(1 row)
+1U: SELECT pg_switch_wal() IS NOT NULL;
+ ?column? 
+----------
+ t        
+(1 row)
 1U: INSERT INTO unlogged_wal_retention_test SELECT (select restart_lsn FROM pg_replication_slots WHERE slot_name = 'internal_wal_replication_slot') as restart_lsn_before, (select count(*) from pg_ls_waldir()) as wal_count_before;
 INSERT 1
 5: CHECKPOINT;
@@ -663,7 +689,7 @@ DROP
 
 -- end_ignore
 (exited with code 0)
-!\retcode gpstop -ari;
+!\retcode gpstop -arf;
 -- start_ignore
 -- end_ignore
 (exited with code 0)

src/test/isolation2/expected/prepare_limit.out

@@ -14,7 +14,7 @@
 -- Enable gp_autostats_mode to make sure the single value insert triger 2pc
 !\retcode gpconfig -c gp_autostats_mode -v 'on_no_stats' --skipvalidation;
 (exited with code 0)
-!\retcode gpstop -ari;
+!\retcode gpstop -arf;
 (exited with code 0)
 
 5: create table prepare_limit1 (a int);
@@ -119,5 +119,5 @@ DROP
 (exited with code 0)
 !\retcode gpconfig -c gp_autostats_mode -v 'none' --skipvalidation;
 (exited with code 0)
-!\retcode gpstop -ari;
+!\retcode gpstop -arf;
 (exited with code 0)

src/test/isolation2/expected/resgroup/resgroup_auxiliary_tools_v1.out

@@ -27,7 +27,7 @@ ERROR:  language "plpython3u" already exists
 
 ! gpconfig -c runaway_detector_activation_percent -v 100;
 
-! gpstop -rai;
+! gpstop -raf;
 -- end_ignore
 
 -- after the restart we need a new connection to run the queries

src/test/isolation2/expected/resgroup/resgroup_auxiliary_tools_v2.out

@@ -27,7 +27,7 @@ ERROR:  language "plpython3u" already exists
 
 ! gpconfig -c runaway_detector_activation_percent -v 100;
 
-! gpstop -rai;
+! gpstop -raf;
 -- end_ignore
 
 -- after the restart we need a new connection to run the queries

src/test/isolation2/expected/resgroup/resgroup_cpuset_empty_default.out

@@ -30,7 +30,7 @@ ALTER RESOURCE GROUP
 
 
 -- start_ignore
-! gpstop -rai;
+! gpstop -raf;
 -- end_ignore
 
 -- Cleanup in a new connection as the default one is disconnected by gpstop

src/test/isolation2/expected/resgroup/resgroup_disable_resgroup.out

@@ -2,7 +2,7 @@
 -- start_ignore
 ! gpconfig -r gp_resource_manager;
 
-! gpstop -rai;
+! gpstop -ra;
 
 SHOW gp_resource_manager;
  gp_resource_manager 

src/test/isolation2/expected/resgroup/resgroup_syntax.out

@@ -563,7 +563,7 @@ DROP
 -- start_ignore
 -- end_ignore
 (exited with code 0)
-!\retcode gpstop -ari;
+!\retcode gpstop -arf;
 -- start_ignore
 -- end_ignore
 (exited with code 0)
@@ -579,7 +579,7 @@ DROP
 -- start_ignore
 -- end_ignore
 (exited with code 0)
-!\retcode gpstop -ari;
+!\retcode gpstop -arf;
 -- start_ignore
 -- end_ignore
 (exited with code 0)

src/test/isolation2/init_file_isolation2

@@ -47,8 +47,8 @@ m/.*Process \d+ waits for ExclusiveLock on resource queue \d+; blocked by proces
 s/.*Process \d+ waits for ExclusiveLock on resource queue \d+; blocked by process \d+./Process PID waits for ExclusiveLock on resource queue OID; blocked by process PID./
 
 # For resource queue self-deadlock detail
-m/resource queue id: \d+, portal id: \d+/
-s/resource queue id: \d+, portal id: \d+/resource queue id: OID, portal id: NUM/
+m/resource queue id: (\d+|XXXX), portal id: \d+/
+s/resource queue id: (\d+|XXXX), portal id: \d+/resource queue id: OID, portal id: NUM/
 
 m/^Process \d+ waits for ShareLock on transaction \d+; blocked by process \d+./
 s/^Process \d+ waits for ShareLock on transaction \d+; blocked by process \d+./Process PID waits for ShareLock on transaction XID; blocked by process PID./

src/test/isolation2/input/uao/ao_unique_index_vacuum.source

@@ -51,9 +51,22 @@ CREATE TABLE unique_index_vacuum_@amname@(i int UNIQUE) USING @amname@
 INSERT INTO unique_index_vacuum_@amname@ SELECT generate_series(1, 5);
 DELETE FROM unique_index_vacuum_@amname@ WHERE i = 5;
 
+-- Enable debug logging to diagnose flaky compaction behavior.
+SET debug_appendonly_print_compaction = on;
+SET debug_appendonly_print_segfile_choice = on;
+SET debug_appendonly_print_visimap = on;
+
 select gp_inject_fault_infinite('vacuum_ao_after_index_delete', 'skip', dbid)
 	from gp_segment_configuration where role = 'p' AND content <> -1;
 
+-- The first VACUUM performs COMPACT (marks old segfile AWAITING_DROP) and
+-- POST_CLEANUP. Due to a race condition, POST_CLEANUP's cutoff XID check in
+-- AppendOptimizedCollectDeadSegments() may find the AWAITING_DROP segment's
+-- xmin too new, resulting in an empty dead_segs bitmap and scan_index() being
+-- called instead of vacuum_appendonly_index() (which has the fault injector).
+-- The second VACUUM guarantees the xid is old enough for the dead segment to
+-- be collected, ensuring the fault is triggered.
+VACUUM unique_index_vacuum_@amname@;
 VACUUM unique_index_vacuum_@amname@;
 
 -- Wait until all old index entries have been deleted by the VACUUM.
@@ -64,4 +77,8 @@ SELECT gp_wait_until_triggered_fault('vacuum_ao_after_index_delete', 1, dbid)
 SELECT gp_inject_fault_infinite('vacuum_ao_after_index_delete', 'reset', dbid)
     FROM gp_segment_configuration WHERE role = 'p' AND content <> -1;
 
+RESET debug_appendonly_print_compaction;
+RESET debug_appendonly_print_segfile_choice;
+RESET debug_appendonly_print_visimap;
+
 DROP TABLE unique_index_vacuum_@amname@;

src/test/isolation2/input/workfile_mgr_test.source

@@ -57,7 +57,7 @@ language plpgsql volatile execute on all segments;
 
 -- start_ignore
 !\retcode gpconfig -c gp_workfile_max_entries -v 32 --skipvalidation;
-!\retcode gpstop -ari;
+!\retcode gpstop -arf;
 -- end_ignore
 
 -- setup for workfile made in temp tablespace test
@@ -80,7 +80,7 @@ language plpgsql volatile execute on all segments;
 
 -- start_ignore
 !\retcode gpconfig -r gp_workfile_max_entries --skipvalidation;
-!\retcode gpstop -ari;
+!\retcode gpstop -arf;
 -- end_ignore
 
 -- test workset cleanup

src/test/isolation2/output/uao/ao_unique_index_vacuum.source

@@ -83,6 +83,14 @@ INSERT 5
 DELETE FROM unique_index_vacuum_@amname@ WHERE i = 5;
 DELETE 1
 
+-- Enable debug logging to diagnose flaky compaction behavior.
+SET debug_appendonly_print_compaction = on;
+SET
+SET debug_appendonly_print_segfile_choice = on;
+SET
+SET debug_appendonly_print_visimap = on;
+SET
+
 select gp_inject_fault_infinite('vacuum_ao_after_index_delete', 'skip', dbid) from gp_segment_configuration where role = 'p' AND content <> -1;
  gp_inject_fault_infinite 
 --------------------------
@@ -91,6 +99,15 @@ select gp_inject_fault_infinite('vacuum_ao_after_index_delete', 'skip', dbid) fr
  Success:                 
 (3 rows)
 
+-- The first VACUUM performs COMPACT (marks old segfile AWAITING_DROP) and
+-- POST_CLEANUP. Due to a race condition, POST_CLEANUP's cutoff XID check in
+-- AppendOptimizedCollectDeadSegments() may find the AWAITING_DROP segment's
+-- xmin too new, resulting in an empty dead_segs bitmap and scan_index() being
+-- called instead of vacuum_appendonly_index() (which has the fault injector).
+-- The second VACUUM guarantees the xid is old enough for the dead segment to
+-- be collected, ensuring the fault is triggered.
+VACUUM unique_index_vacuum_@amname@;
+VACUUM
 VACUUM unique_index_vacuum_@amname@;
 VACUUM
 
@@ -112,5 +129,12 @@ SELECT gp_inject_fault_infinite('vacuum_ao_after_index_delete', 'reset', dbid) F
  Success:                 
 (3 rows)
 
+RESET debug_appendonly_print_compaction;
+RESET
+RESET debug_appendonly_print_segfile_choice;
+RESET
+RESET debug_appendonly_print_visimap;
+RESET
+
 DROP TABLE unique_index_vacuum_@amname@;
 DROP

src/test/isolation2/output/workfile_mgr_test.source

@@ -30,7 +30,7 @@ CREATE
 
 -- end_ignore
 (exited with code 0)
-!\retcode gpstop -ari;
+!\retcode gpstop -arf;
 -- start_ignore
 20200923:12:05:57:014232 gpstop:mdw:gpadmin-[INFO]:-Starting gpstop with args: -ari
 20200923:12:05:57:014232 gpstop:mdw:gpadmin-[INFO]:-Gathering information and validating the environment...
@@ -132,7 +132,7 @@ DROP
 
 -- end_ignore
 (exited with code 0)
-!\retcode gpstop -ari;
+!\retcode gpstop -arf;
 -- start_ignore
 20200923:12:08:05:016154 gpstop:mdw:gpadmin-[INFO]:-Starting gpstop with args: -ari
 20200923:12:08:05:016154 gpstop:mdw:gpadmin-[INFO]:-Gathering information and validating the environment...

src/test/isolation2/sql/bitmap_index_crash.sql

@@ -49,7 +49,7 @@ create index on test_bitmap using bitmap(type);
 select count(*) from test_bitmap where type = 520;
 
 -- start_ignore
-! gpstop -rai;
+! gpstop -raf;
 -- end_ignore
 
 0: select count(*) from test_bitmap where type = 520;

src/test/isolation2/sql/gdd/end.sql

@@ -4,7 +4,7 @@ ALTER SYSTEM RESET gp_global_deadlock_detector_period;
 -- Use utility session on seg 0 to restart master. This way avoids the
 -- situation where session issuing the restart doesn't disappear
 -- itself.
-1U:SELECT pg_ctl(dir, 'restart') from datadir;
+1U:SELECT pg_ctl(dir, 'restart', 'fast') from datadir;
 -- Start new session on master to make sure it has fully completed
 -- recovery and up and running again.
 1: SHOW gp_enable_global_deadlock_detector;

src/test/isolation2/sql/gdd/prepare.sql

@@ -59,7 +59,7 @@ ALTER SYSTEM SET gp_global_deadlock_detector_period TO 5;
 -- Use utility session on seg 0 to restart master. This way avoids the
 -- situation where session issuing the restart doesn't disappear
 -- itself.
-1U:SELECT pg_ctl(dir, 'restart') from datadir;
+1U:SELECT pg_ctl(dir, 'restart', 'fast') from datadir;
 -- Start new session on master to make sure it has fully completed
 -- recovery and up and running again.
 1: SHOW gp_enable_global_deadlock_detector;

src/test/isolation2/sql/instr_in_shmem_setup.sql

@@ -1,4 +1,4 @@
 -- start_ignore
 ! gpconfig -c gp_enable_query_metrics -v on; 
-! gpstop -rai;
+! gpstop -raf;
 -- end_ignore