Stablize regression tests of dispatcher.

Signed-off-by: NPengzhou Tang <ptang@pivotal.io>

src/test/regress/greenplum_schedule

@@ -25,7 +25,7 @@ test: leastsquares
 test: opr_sanity_gp decode_expr bitmapscan bitmapscan_ao case_gp limit_gp notin percentile naivebayes join_gp union_gp gpcopy gp_create_table
 test: filter gpctas gpdist matrix toast sublink table_functions olap_setup complex opclass_ddl information_schema guc_env_var
 test: bitmap_index gp_dump_query_oids
-test: indexjoin as_alias regex_gp gpparams with_clause transient_types gang_mgmt
+test: indexjoin as_alias regex_gp gpparams with_clause transient_types
 # dispatch should always run seperately from other cases.
 test: dispatch
 

src/test/regress/input/dispatch.source

@@ -83,14 +83,14 @@ CREATE OR REPLACE FUNCTION cleanupAllGangs() RETURNS BOOL
 AS '@abs_builddir@/regress@DLSUFFIX@', 'cleanupAllGangs' LANGUAGE C;
 
 -- check if segments has backend running
-CREATE OR REPLACE FUNCTION numBackendsOnSegment() RETURNS INTEGER
+CREATE OR REPLACE FUNCTION numBackendsOnSegment(int) RETURNS INTEGER
 AS '@abs_builddir@/regress@DLSUFFIX@', 'numBackendsOnSegment' LANGUAGE C;
 
 -- check if QD has reusable gangs
 CREATE OR REPLACE FUNCTION hasGangsExist() RETURNS BOOL
 AS '@abs_builddir@/regress@DLSUFFIX@', 'hasGangsExist' LANGUAGE C;
 
-CREATE VIEW v_hasBackendsOnSegment as select sum(numBackendsOnSegment()) > 0 from
+CREATE VIEW v_hasBackendsOnSegment as select sum(numBackendsOnSegment(30)) > 0 from
 gp_dist_random('gp_id');
 
 -- disable debug linger to get immediate feedback from FATAL errors.
@@ -156,12 +156,6 @@ select 'wait recovery finish' from gp_dist_random('gp_id');
 -- cleanup all reusable gangs 
 select cleanupAllGangs();
 
--- sleep 1 second to wait all QEs to exit, since QE exit in this situation is an
--- asynchronous procedure, 1 second may not guarantee they have exited in
--- extreme poor performant machines, but we have no better solution now, and this
--- test case is pretty necessary, so...
-select pg_sleep(1);
-
 -- expect no zombie backends left on segments 
 select * from v_hasBackendsOnSegment;
 
@@ -185,12 +179,6 @@ where dispatch_test_t1.c2 = dispatch_test_t2.c2 and dispatch_test_t2.c3 = dispat
 -- expect no resuable gang exist
 select * from hasGangsExist();
 
--- sleep 1 second to wait all QEs to exit, since QE exit in this situation is an
--- asynchronous procedure, 1 second may not guarantee they have exited in
--- extreme poor performant machines, but we have no better solution now, and this
--- test case is pretty necessary, so...
-select pg_sleep(1);
-
 -- expect no zombie backends left on segments.
 select * from v_hasBackendsOnSegment;
 
@@ -208,12 +196,6 @@ where dispatch_test_t1.c2 = dispatch_test_t2.c2 and dispatch_test_t2.c3 = dispat
 -- expect resuable gang exist
 select * from hasGangsExist();
 
--- sleep 1 second to wait all QEs to exit, since QE exit in this situation is an
--- asynchronous procedure, 1 second may not guarantee they have exited in
--- extreme poor performant machines, but we have no better solution now, and this
--- test case is pretty necessary, so...
-select pg_sleep(1);
-
 -- expect QEs exist.
 select * from v_hasBackendsOnSegment;
 
@@ -260,3 +242,22 @@ select * from dispatch_test_t1, dispatch_test_t2, dispatch_test_t3
 where dispatch_test_t1.c2 = dispatch_test_t2.c2 and dispatch_test_t2.c3 = dispatch_test_t3.c3;
 
 \! gpfaultinjector -q -f after_one_slice_dispatched -y reset --seg_dbid 1
+
+-- test logging of gang management
+SET gp_log_gang = 'debug';
+
+-- test INFO raised from segments with various kinds of fields
+CREATE OR REPLACE FUNCTION udf_raise_info() RETURNS BOOL
+AS '@abs_builddir@/regress@DLSUFFIX@', 'gangRaiseInfo' LANGUAGE C;
+
+SELECT udf_raise_info() FROM gp_dist_random('gp_id') WHERE gp_segment_id = 0;
+
+-- terminate connections to all segments
+\c
+
+-- test brutal exit of session can correctly clean up all QEs
+-- start and finish a new session
+\! psql regression -f @abs_builddir@/sql/dispatch_test_helper.sql
+
+-- verify all QEs of the exited session have gone, expect to return false
+select * from v_hasBackendsOnSegment;

src/test/regress/input/gang_mgmt.source

--- test logging of gang management
-SET gp_log_gang = 'debug';
-
--- test INFO raised from segments with various kinds of fields
-CREATE OR REPLACE FUNCTION udf_raise_info() RETURNS BOOL
-AS '@abs_builddir@/regress@DLSUFFIX@', 'gangRaiseInfo' LANGUAGE C;
-
-SELECT udf_raise_info() FROM gp_dist_random('gp_id') WHERE gp_segment_id = 0;
-
--- terminate connections to all segments
-\c
-
--- test brutal exit of session can correctly clean up all QEs
--- start and finish a new session
-\! psql regression -f @abs_builddir@/sql/gang_mgmt_helper.sql
-
--- sleep 1 second to wait all QEs to exit, since QE exit in this situation is an
--- asynchronous procedure, 1 second may not guarantee they have exited in
--- extreme poor performant machines, but we have no better solution now, and this
--- test case is pretty necessary, so...
-select pg_sleep(1);
-
--- verify all QEs of the exited session have gone, expect to return nothing
-SELECT sess_id FROM tmp_sess_id, (SELECT get_segment_qe_session_ids() FROM gp_dist_random('gp_id')) sessions WHERE tmp_sess_id.sess_id = sessions.get_segment_qe_session_ids;

src/test/regress/output/dispatch.source

@@ -105,12 +105,12 @@ insert into dispatch_test_t3 values (3,1,2);
 CREATE OR REPLACE FUNCTION cleanupAllGangs() RETURNS BOOL
 AS '@abs_builddir@/regress@DLSUFFIX@', 'cleanupAllGangs' LANGUAGE C;
 -- check if segments has backend running
-CREATE OR REPLACE FUNCTION numBackendsOnSegment() RETURNS INTEGER
+CREATE OR REPLACE FUNCTION numBackendsOnSegment(int) RETURNS INTEGER
 AS '@abs_builddir@/regress@DLSUFFIX@', 'numBackendsOnSegment' LANGUAGE C;
 -- check if QD has reusable gangs
 CREATE OR REPLACE FUNCTION hasGangsExist() RETURNS BOOL
 AS '@abs_builddir@/regress@DLSUFFIX@', 'hasGangsExist' LANGUAGE C;
-CREATE VIEW v_hasBackendsOnSegment as select sum(numBackendsOnSegment()) > 0 from
+CREATE VIEW v_hasBackendsOnSegment as select sum(numBackendsOnSegment(30)) > 0 from
 gp_dist_random('gp_id');
 -- disable debug linger to get immediate feedback from FATAL errors.
 set gp_debug_linger to 0;
@@ -200,16 +200,6 @@ select cleanupAllGangs();
  t
 (1 row)
 
--- sleep 1 second to wait all QEs to exit, since QE exit in this situation is an
--- asynchronous procedure, 1 second may not guarantee they have exited in
--- extreme poor performant machines, but we have no better solution now, and this
--- test case is pretty necessary, so...
-select pg_sleep(1);
- pg_sleep 
-----------
- 
-(1 row)
-
 -- expect no zombie backends left on segments 
 select * from v_hasBackendsOnSegment;
  ?column? 
@@ -250,16 +240,6 @@ select * from hasGangsExist();
  f
 (1 row)
 
--- sleep 1 second to wait all QEs to exit, since QE exit in this situation is an
--- asynchronous procedure, 1 second may not guarantee they have exited in
--- extreme poor performant machines, but we have no better solution now, and this
--- test case is pretty necessary, so...
-select pg_sleep(1);
- pg_sleep 
-----------
- 
-(1 row)
-
 -- expect no zombie backends left on segments.
 select * from v_hasBackendsOnSegment;
  ?column? 
@@ -290,16 +270,6 @@ select * from hasGangsExist();
  t
 (1 row)
 
--- sleep 1 second to wait all QEs to exit, since QE exit in this situation is an
--- asynchronous procedure, 1 second may not guarantee they have exited in
--- extreme poor performant machines, but we have no better solution now, and this
--- test case is pretty necessary, so...
-select pg_sleep(1);
- pg_sleep 
-----------
- 
-(1 row)
-
 -- expect QEs exist.
 select * from v_hasBackendsOnSegment;
  ?column? 
@@ -356,3 +326,38 @@ select * from dispatch_test_t1, dispatch_test_t2, dispatch_test_t3
 where dispatch_test_t1.c2 = dispatch_test_t2.c2 and dispatch_test_t2.c3 = dispatch_test_t3.c3;
 ERROR:  canceling statement due to user request
 \! gpfaultinjector -q -f after_one_slice_dispatched -y reset --seg_dbid 1
+-- test logging of gang management
+SET gp_log_gang = 'debug';
+-- test INFO raised from segments with various kinds of fields
+CREATE OR REPLACE FUNCTION udf_raise_info() RETURNS BOOL
+AS '@abs_builddir@/regress@DLSUFFIX@', 'gangRaiseInfo' LANGUAGE C;
+SELECT udf_raise_info() FROM gp_dist_random('gp_id') WHERE gp_segment_id = 0;
+INFO:  testing hook function MPPnoticeReceiver  (seg0 slice1 172.17.0.2:25432 pid=34013)
+DETAIL:  this test aims at covering code paths not hit before
+HINT:  no special hint
+CONTEXT:  PL/C function defined in regress.c
+ udf_raise_info 
+----------------
+ t
+(1 row)
+
+-- terminate connections to all segments
+\c
+-- test brutal exit of session can correctly clean up all QEs
+-- start and finish a new session
+\! psql regression -f @abs_builddir@/sql/dispatch_test_helper.sql
+BEGIN
+DECLARE CURSOR
+SET
+psql:@abs_builddir@/sql/dispatch_test_helper.sql:10: FATAL:  User fault injection raised fatal
+server closed the connection unexpectedly
+	This probably means the server terminated abnormally
+	before or while processing the request.
+psql:@abs_builddir@/sql/dispatch_test_helper.sql:10: connection to server was lost
+-- verify all QEs of the exited session have gone, expect to return false
+select * from v_hasBackendsOnSegment;
+ ?column? 
+----------
+ f
+(1 row)
+

src/test/regress/output/gang_mgmt.source

--- test logging of gang management
-SET gp_log_gang = 'debug';
--- test INFO raised from segments with various kinds of fields
-CREATE OR REPLACE FUNCTION udf_raise_info() RETURNS BOOL
-AS '@abs_srcdir@/regress.so', 'gangRaiseInfo' LANGUAGE C;
-SELECT udf_raise_info() FROM gp_dist_random('gp_id') WHERE gp_segment_id = 0;
-INFO:  testing hook function MPPnoticeReceiver  (seg0 slice1 127.0.0.1:10000 pid=9647)
-DETAIL:  this test aims at covering code paths not hit before
-HINT:  no special hint
-CONTEXT:  PL/C function defined in regress.c
- udf_raise_info 
-----------------
- t
-(1 row)
-
--- terminate connections to all segments
-\c
--- test brutal exit of session can correctly clean up all QEs
--- start and finish a new session
-\! psql regression -f @abs_srcdir@/sql/gang_mgmt_helper.sql
-psql:@abs_srcdir@/sql/gang_mgmt_helper.sql:5: NOTICE:  table "tmp_sess_id" does not exist, skipping
-DROP TABLE
-psql:@abs_srcdir@/sql/gang_mgmt_helper.sql:6: NOTICE:  Table doesn't have 'DISTRIBUTED BY' clause -- Using column(s) named 'sess_id' as the Greenplum Database data distribution key for this table.
-HINT:  The 'DISTRIBUTED BY' clause determines the distribution of data. Make sure column(s) chosen are the optimal data distribution key to minimize skew.
-SELECT 1
-CREATE FUNCTION
-BEGIN
-DECLARE CURSOR
-SET
-psql:@abs_srcdir@/sql/gang_mgmt_helper.sql:18: FATAL:  User fault injection raised fatal
-server closed the connection unexpectedly
-	This probably means the server terminated abnormally
-	before or while processing the request.
-psql:@abs_srcdir@/sql/gang_mgmt_helper.sql:18: connection to server was lost
--- sleep 1 second to wait all QEs to exit, since QE exit in this situation is an
--- asynchronous procedure, 1 second may not guarantee they have exited in
--- extreme poor performant machines, but we have no better solution now, and this
--- test case is pretty necessary, so...
-select pg_sleep(1);
- pg_sleep
- ----------
-
- (1 row)
-
--- verify all QEs of the exited session have gone, expect to return nothing
-SELECT sess_id FROM tmp_sess_id, (SELECT get_segment_qe_session_ids() FROM gp_dist_random('gp_id')) sessions WHERE tmp_sess_id.sess_id = sessions.get_segment_qe_session_ids;
- sess_id 
----------
-(0 rows)
-

src/test/regress/regress.c

@@ -2473,15 +2473,26 @@ PG_FUNCTION_INFO_V1(numBackendsOnSegment);
 Datum
 numBackendsOnSegment(PG_FUNCTION_ARGS)
 {
-	int	beid;
-	int32 result = 0;
+	int beid;
+	int32 result;
+	int timeout = PG_GETARG_INT32(0);
+	if (timeout <= 0)
+		elog(ERROR, "timeout expected to be larger than 0");
 	int pid = getpid();
-	int tot_backends = pgstat_fetch_stat_numbackends();
-	for (beid = 1; beid <= tot_backends; beid++)
+	while (timeout > 0)
 	{
-		PgBackendStatus *beentry = pgstat_fetch_stat_beentry(beid);
-		if (beentry && beentry->st_procpid >0 && beentry->st_procpid != pid)
-			result++;
+		result = 0;
+		int tot_backends = pgstat_fetch_stat_numbackends();
+		for (beid = 1; beid <= tot_backends; beid++)
+		{
+			PgBackendStatus *beentry = pgstat_fetch_stat_beentry(beid);
+			if (beentry && beentry->st_procpid >0 && beentry->st_procpid != pid)
+				result++;
+		}
+		if (result == 0)
+			break;
+		sleep(1); /* 1 second */
+		timeout--;
 	}
 	
 	PG_RETURN_INT32(result);

src/test/regress/sql/.gitignore

@@ -27,6 +27,5 @@ external_table.sql
 largeobject.sql
 qp_gist_indexes2.sql
 qp_regexp.sql
-gang_mgmt.sql
 guc_env_var.sql
 dispatch.sql

src/test/regress/sql/gang_mgmt_helper.sql → src/test/regress/sql/dispatch_test_helper.sql

--- this is a SQL file invoked by gang_mgmt.sql, and should not be executed
+-- this is a SQL file invoked by dispatch.sql, and should not be executed
 -- directly in any test schedule;
 
--- preserve current session id into a table
-DROP TABLE IF EXISTS tmp_sess_id;
-SELECT sess_id INTO tmp_sess_id FROM pg_stat_activity WHERE procpid = pg_backend_pid();
-
--- UDF that returns the set of backend session ids in one segment(duplications
--- included), will be invoked by gang_mgmt.sql to do verification
-CREATE OR REPLACE FUNCTION get_segment_qe_session_ids() RETURNS SETOF int AS $$ SELECT pg_stat_get_backend_session_id(pg_stat_get_backend_idset) FROM pg_stat_get_backend_idset(); $$ LANGUAGE SQL;
-
 BEGIN;
 DECLARE c1 CURSOR FOR SELECT pg_sleep(300) FROM gp_dist_random('gp_id');