Add new fault injector to make the transaction management test more stable.

The fault `twophase_transaction_commit_prepared` would cause the segment to
PANIC while writing a `COMMIT PREPARED` record on the segment. The master would
then retry the `COMMIT PREPARED` while the postmaster was still resetting on the
segment causing the master itself to `PANIC`. This patch introduces a new fault
injector `finish_prepared_before_commit`  which will not cause a `PANIC`, but
just error out since the intent of the test is to ensure that the retry works
correctly.

gpMgmt/bin/gppylib/programs/clsInjectFault.py

@@ -424,6 +424,7 @@ class GpInjectFaultProgram:
                   "cursor_qe_reader_after_snapshot (inject fault after QE READER has populated snashot for cursor)" \
 			      "fsync_counter (inject fault to count buffers fsync'ed by checkpoint process), " \
 			      "bg_buffer_sync_default_logic (inject fault to 'skip' in order to flush all buffers in BgBufferSync()), " \
+                  "finish_prepared_after_record_commit_prepared (inject fault in FinishPreparedTransaction() after recording the commit prepared record), " \
 			      "all (affects all faults injected, used for 'status' and 'reset'), ") 
         addTo.add_option("-c", "--ddl_statement", dest="ddlStatement", type="string",
                          metavar="ddlStatement",

src/backend/access/transam/twophase.c

@@ -1646,6 +1646,8 @@ FinishPreparedTransaction(const char *gid, bool isCommit, bool raiseErrorIfNotFo
 
 	END_CRIT_SECTION();
 
+	SIMPLE_FAULT_INJECTOR(FinishPreparedAfterRecordCommitPrepared);
+
 	/* Need to figure out the memory allocation and deallocationfor "buffer". For now, just let it leak. */
 
 	return true;

src/backend/utils/misc/faultinjector.c

@@ -331,6 +331,8 @@ FaultInjectorIdentifierEnumToString[] = {
 		/* inject fault to 'skip' in order to flush all buffers in BgBufferSync() */
 	_("bg_buffer_sync_default_logic"),
 		/* inject fault to count buffers fsync'ed by checkpoint process */
+	_("finish_prepared_after_record_commit_prepared"),
+		/* inject fault in FinishPreparedTransaction() after recording the commit prepared record */
 	_("not recognized"),
 };
 

src/include/utils/faultinjector.h

@@ -221,6 +221,7 @@ typedef enum FaultInjectorIdentifier_e {
 	FsyncCounter,
 	BgBufferSyncDefaultLogic,
 
+	FinishPreparedAfterRecordCommitPrepared,
 	/* INSERT has to be done before that line */
 	FaultInjectorIdMax,
 	

src/test/tinc/Makefile

@@ -65,7 +65,7 @@ non_discover_targets: aoco_compression filerep_end_to_end fts
 # stages for configuration requirements.
 
 storage: storage_vacuum_xidlimits aocoalter_catalog_loaders \
-	storage_queryfinish_and_transactionmanagement \
+	storage_uao_and_transactionmanagement \
 	storage_persistent_accessmethods_and_vacuum \
 	storage_filerep
 

src/test/tinc/tincrepo/mpp/gpdb/tests/storage/transaction_management/mpp23395/test_mpp23395.py

@@ -181,19 +181,17 @@ class mpp23395(MPPTestCase):
         SET debug_dtm_action = "fail_begin_command";
         CREATE TABLE mpp23395(a int);
         '''
-        self.run_sequence(sql, 'twophase_transaction_commit_prepared', 'error', 2, False);
+        self.run_sequence(sql, 'finish_prepared_after_record_commit_prepared', 'error', 2, False);
 
-        # QE panics after writing prepare xlog record.  This should cause
-        # master to broadcast abort but QEs handle the abort in
+        # Scenario 8: QE panics after writing prepare xlog record.  This should
+        # cause master to broadcast abort but QEs handle the abort in
         # DTX_CONTEXT_LOCAL_ONLY context.
         sql = '''
         DROP TABLE IF EXISTS mpp23395;
         CREATE TABLE mpp23395(a int);
         INSERT INTO mpp23395 VALUES(1), (2), (3);
-        BEGIN;
         SET debug_abort_after_segment_prepared = true;
         DELETE FROM mpp23395;
-        COMMIT;
         '''
 
         # No prepared transactions should remain lingering