Fix the FATAL error of autovacuum on template0 (#10920)

The autovacuum worker for template0 would FATAL because Gp_session_role is still the dispatcher
role with below message.

2020-09-29 21:20:02.686827 CST,,,p19902,th-881792832,,,,0,,,seg2,,,,sx1,"FATAL","57P03","
connections to primary segments are not allowed","This database instance is running as a
primary segment in a Greenplum cluster and does not permit direct connections.","
To force a connection anyway (dangerous!), use utility mode.",,,,,0,,"postinit.c",1151,

Fixing this by setting Gp_session_role as the utility role.

src/backend/postmaster/autovacuum.c

@@ -1539,6 +1539,7 @@ AutoVacWorkerMain(int argc, char *argv[])
 
 	/* MPP-4990: Autovacuum always runs as utility-mode */
 	Gp_role = GP_ROLE_UTILITY;
+	Gp_session_role = GP_ROLE_UTILITY;
 
 	/* reset MyProcPid */
 	MyProcPid = getpid();

src/backend/utils/init/postinit.c

@@ -1135,8 +1135,7 @@ InitPostgres(const char *in_dbname, Oid dboid, const char *username,
 	 * protect ourselves from normal clients who might be trying to connect to
 	 * the system while we startup.
 	 */
-	if ((Gp_role == GP_ROLE_UTILITY) && (Gp_session_role != GP_ROLE_UTILITY) &&
-		!IsAutoVacuumWorkerProcess())
+	if ((Gp_role == GP_ROLE_UTILITY) && (Gp_session_role != GP_ROLE_UTILITY))
 	{
 		ereport(FATAL,
 				(errcode(ERRCODE_CANNOT_CONNECT_NOW),

src/test/regress/greenplum_schedule

@@ -262,6 +262,7 @@ test: oid_wraparound
 test: fts_recovery_in_progress
 test: mirror_replay
 test: autovacuum-template0
+test: autovacuum-template0-segment
 
 # gpexpand introduce the partial tables, check them if they can run correctly
 test: gangsize gang_reuse

src/test/regress/input/autovacuum-template0-segment.source

+-- start_ignore
+create extension if not exists gp_inject_fault;
+
+create or replace function test_consume_xids(int4) returns void
+as '@abs_srcdir@/regress.so', 'test_consume_xids'
+language C;
+
+\! gpconfig -c debug_burn_xids -v on --skipvalidation
+\! gpstop -au
+-- end_ignore
+
+-- Autovacuum should take care of anti-XID wraparounds of 'template0'. Because
+-- of that, the age of template0 should not go much above
+-- autovacuum_freeze_max_age (we assume the default of 200 million here).
+select age(datfrozenxid) < 200 * 1000000 from gp_dist_random('pg_database') where datname='template0';
+
+-- track that we've updated the row in pg_database for template0
+SELECT gp_inject_fault_infinite('vacuum_update_dat_frozen_xid', 'skip', dbid) from gp_segment_configuration where content = 0 and role = 'p';
+
+-- Suspend the autovacuum worker from vacuuming before
+-- ShmemVariableCache->latestCompletedXid is expected to be updated
+SELECT gp_inject_fault_infinite('auto_vac_worker_before_do_autovacuum', 'suspend', dbid) from gp_segment_configuration where content = 0 and role = 'p';
+
+select test_consume_xids(100 * 1000000) from gp_dist_random('gp_id') where gp_segment_id = 0;
+select test_consume_xids(100 * 1000000) from gp_dist_random('gp_id') where gp_segment_id = 0;
+select test_consume_xids(10 * 1000000) from gp_dist_random('gp_id') where gp_segment_id = 0;
+
+-- Wait until autovacuum is triggered
+SELECT gp_wait_until_triggered_fault('auto_vac_worker_before_do_autovacuum', 1, dbid) from gp_segment_configuration where content = 0 and role = 'p';
+SELECT gp_inject_fault('auto_vac_worker_before_do_autovacuum', 'reset', dbid) from gp_segment_configuration where content = 0 and role = 'p';
+
+-- wait until autovacuum worker updates pg_database
+SELECT gp_wait_until_triggered_fault('vacuum_update_dat_frozen_xid', 1, dbid) from gp_segment_configuration where content = 0 and role = 'p';
+SELECT gp_inject_fault('vacuum_update_dat_frozen_xid', 'reset', dbid) from gp_segment_configuration where content = 0 and role = 'p';
+
+-- template0 should be young
+select age(datfrozenxid) < 200 * 1000000 from gp_dist_random('pg_database') where datname='template0';
+
+-- But autovacuum should not touch other databases. Hence, our database
+-- should be well above the 200 million mark.
+select age(datfrozenxid) > 200 * 1000000 from gp_dist_random('pg_database') where datname=current_database() and gp_segment_id = 0;
+
+-- start_ignore
+\! gpconfig -c debug_burn_xids -v off --skipvalidation
+\! gpstop -au
+-- end_ignore
+

src/test/regress/output/autovacuum-template0-segment.source

+-- start_ignore
+create extension if not exists gp_inject_fault;
+create or replace function test_consume_xids(int4) returns void
+as '@abs_srcdir@/regress.so', 'test_consume_xids'
+language C;
+\! gpconfig -c debug_burn_xids -v on --skipvalidation
+20201009:19:05:11:034329 gpconfig:gpdb6:zhanyi-[INFO]:-completed successfully with parameters '-c debug_burn_xids -v on --skipvalidation'
+\! gpstop -au
+20201009:19:05:12:034666 gpstop:gpdb6:zhanyi-[INFO]:-Starting gpstop with args: -au
+20201009:19:05:12:034666 gpstop:gpdb6:zhanyi-[INFO]:-Gathering information and validating the environment...
+20201009:19:05:12:034666 gpstop:gpdb6:zhanyi-[INFO]:-Obtaining Greenplum Master catalog information
+20201009:19:05:12:034666 gpstop:gpdb6:zhanyi-[INFO]:-Obtaining Segment details from master...
+20201009:19:05:12:034666 gpstop:gpdb6:zhanyi-[INFO]:-Greenplum Version: 'postgres (Greenplum Database) 6.11.1+dev.22.gcf4e9dbf19 build dev'
+20201009:19:05:12:034666 gpstop:gpdb6:zhanyi-[INFO]:-Signalling all postmaster processes to reload
+-- end_ignore
+-- Autovacuum should take care of anti-XID wraparounds of 'template0'. Because
+-- of that, the age of template0 should not go much above
+-- autovacuum_freeze_max_age (we assume the default of 200 million here).
+select age(datfrozenxid) < 200 * 1000000 from gp_dist_random('pg_database') where datname='template0';
+ ?column? 
+----------
+ t
+ t
+ t
+(3 rows)
+
+-- track that we've updated the row in pg_database for template0
+SELECT gp_inject_fault_infinite('vacuum_update_dat_frozen_xid', 'skip', dbid) from gp_segment_configuration where content = 0 and role = 'p';
+ gp_inject_fault_infinite 
+--------------------------
+ Success:
+(1 row)
+
+-- Suspend the autovacuum worker from vacuuming before
+-- ShmemVariableCache->latestCompletedXid is expected to be updated
+SELECT gp_inject_fault_infinite('auto_vac_worker_before_do_autovacuum', 'suspend', dbid) from gp_segment_configuration where content = 0 and role = 'p';
+ gp_inject_fault_infinite 
+--------------------------
+ Success:
+(1 row)
+
+select test_consume_xids(100 * 1000000) from gp_dist_random('gp_id') where gp_segment_id = 0;
+ test_consume_xids 
+-------------------
+ 
+(1 row)
+
+select test_consume_xids(100 * 1000000) from gp_dist_random('gp_id') where gp_segment_id = 0;
+ test_consume_xids 
+-------------------
+ 
+(1 row)
+
+select test_consume_xids(10 * 1000000) from gp_dist_random('gp_id') where gp_segment_id = 0;
+ test_consume_xids 
+-------------------
+ 
+(1 row)
+
+-- Wait until autovacuum is triggered
+SELECT gp_wait_until_triggered_fault('auto_vac_worker_before_do_autovacuum', 1, dbid) from gp_segment_configuration where content = 0 and role = 'p';
+ gp_wait_until_triggered_fault 
+-------------------------------
+ Success:
+(1 row)
+
+SELECT gp_inject_fault('auto_vac_worker_before_do_autovacuum', 'reset', dbid) from gp_segment_configuration where content = 0 and role = 'p';
+ gp_inject_fault 
+-----------------
+ Success:
+(1 row)
+
+-- wait until autovacuum worker updates pg_database
+SELECT gp_wait_until_triggered_fault('vacuum_update_dat_frozen_xid', 1, dbid) from gp_segment_configuration where content = 0 and role = 'p';
+ gp_wait_until_triggered_fault 
+-------------------------------
+ Success:
+(1 row)
+
+SELECT gp_inject_fault('vacuum_update_dat_frozen_xid', 'reset', dbid) from gp_segment_configuration where content = 0 and role = 'p';
+ gp_inject_fault 
+-----------------
+ Success:
+(1 row)
+
+-- template0 should be young
+select age(datfrozenxid) < 200 * 1000000 from gp_dist_random('pg_database') where datname='template0';
+ ?column? 
+----------
+ t
+ t
+ t
+(3 rows)
+
+-- But autovacuum should not touch other databases. Hence, our database
+-- should be well above the 200 million mark.
+select age(datfrozenxid) > 200 * 1000000 from gp_dist_random('pg_database') where datname=current_database() and gp_segment_id = 0;
+ ?column? 
+----------
+ t
+(1 row)
+
+-- start_ignore
+\! gpconfig -c debug_burn_xids -v off --skipvalidation
+20201009:19:06:18:034710 gpconfig:gpdb6:zhanyi-[INFO]:-completed successfully with parameters '-c debug_burn_xids -v off --skipvalidation'
+\! gpstop -au
+20201009:19:06:19:035047 gpstop:gpdb6:zhanyi-[INFO]:-Starting gpstop with args: -au
+20201009:19:06:19:035047 gpstop:gpdb6:zhanyi-[INFO]:-Gathering information and validating the environment...
+20201009:19:06:19:035047 gpstop:gpdb6:zhanyi-[INFO]:-Obtaining Greenplum Master catalog information
+20201009:19:06:19:035047 gpstop:gpdb6:zhanyi-[INFO]:-Obtaining Segment details from master...
+20201009:19:06:19:035047 gpstop:gpdb6:zhanyi-[INFO]:-Greenplum Version: 'postgres (Greenplum Database) 6.11.1+dev.22.gcf4e9dbf19 build dev'
+20201009:19:06:19:035047 gpstop:gpdb6:zhanyi-[INFO]:-Signalling all postmaster processes to reload
+-- end_ignore