Check in SHOW command that backups in RUNNING status

Author: Artur Zakirov

backup.c

@@ -432,7 +432,7 @@ do_backup(bool smooth_checkpoint)
 	elog(LOG, "----------------------------------------");
 
 	/* get exclusive lock of backup catalog */
-	catalog_lock(true);
+	catalog_lock(true, NULL);
 
 	/* initialize backup result */
 	current.status = BACKUP_STATUS_RUNNING;

catalog.c

@@ -42,8 +42,8 @@ unlink_lock_atexit(void)
 /*
  * Create a lockfile.
  */
-int
-catalog_lock(bool check_catalog)
+void
+catalog_lock(bool check_catalog, pid_t *run_pid)
 {
 	int			fd;
 	char		buffer[MAXPGPATH * 2 + 256];
@@ -53,6 +53,9 @@ catalog_lock(bool check_catalog)
 	pid_t		my_pid,
 				my_p_pid;
 
+	if (run_pid)
+		*run_pid = 0;
+
 	join_path_components(lock_file, backup_path, BACKUP_CATALOG_PID);
 
 	/*
@@ -149,7 +152,16 @@ catalog_lock(bool check_catalog)
 		{
 			if (kill(encoded_pid, 0) == 0 ||
 				(errno != ESRCH && errno != EPERM))
-				elog(ERROR, "lock file \"%s\" already exists", lock_file);
+			{
+				/* If run_pid was specified just return encoded_pid */
+				if (run_pid)
+				{
+					*run_pid = encoded_pid;
+					return;
+				}
+				else
+					elog(ERROR, "lock file \"%s\" already exists", lock_file);
+			}
 		}
 
 		/*
@@ -220,8 +232,6 @@ catalog_lock(bool check_catalog)
 			elog(ERROR, "Backup directory was initialized for system id = %ld, but target system id = %ld",
 				 system_identifier, id);
 	}
-
-	return 0;
 }
 
 /*

delete.c

@@ -31,7 +31,7 @@ do_delete(time_t backup_id)
 		elog(ERROR, "required backup ID not specified");
 
 	/* Lock backup catalog */
-	catalog_lock(false);
+	catalog_lock(false, NULL);
 
 	/* Get complete list of backups */
 	backup_list = catalog_get_backup_list(0);
@@ -98,7 +98,7 @@ do_deletewal(time_t backup_id, bool strict, bool need_catalog_lock)
 
 	/* Lock backup catalog */
 	if (need_catalog_lock)
-		catalog_lock(false);
+		catalog_lock(false, NULL);
 
 	/* Find oldest LSN, used by backups */
 	backup_list = catalog_get_backup_list(0);
@@ -154,7 +154,7 @@ do_retention_purge(void)
 		elog(ERROR, "retention policy is not set");
 
 	/* Lock backup catalog */
-	catalog_lock(false);
+	catalog_lock(false, NULL);
 
 	/* Get a complete list of backups. */
 	backup_list = catalog_get_backup_list(0);

pg_probackup.h

@@ -300,7 +300,7 @@ extern parray *catalog_get_backup_list(time_t backup_id);
 extern pgBackup *catalog_get_last_data_backup(parray *backup_list,
 											  TimeLineID tli);
 
-extern int catalog_lock(bool check_catalog);
+extern void catalog_lock(bool check_catalog, pid_t *run_pid);
 
 extern void pgBackupWriteConfigSection(FILE *out, pgBackup *backup);
 extern void pgBackupWriteResultSection(FILE *out, pgBackup *backup);

restore.c

@@ -106,7 +106,7 @@ do_restore(time_t backup_id,
 	elog(LOG, "restore start");
 
 	/* get exclusive lock of backup catalog */
-	catalog_lock(false);
+	catalog_lock(false, NULL);
 
 	/* confirm the PostgreSQL server is not running */
 	if (is_pg_running())

show.c

@@ -28,7 +28,8 @@ do_show(time_t backup_id)
 	 */
 	if (backup_id != 0)
 	{
-		pgBackup *backup;
+		pgBackup   *backup;
+		pid_t		run_pid;
 
 		backup = read_backup(backup_id);
 		if (backup == NULL)
@@ -40,6 +41,18 @@ do_show(time_t backup_id)
 			/* This is not error case */
 			return 0;
 		}
+
+		/* Fix backup status */
+		if (backup->status == BACKUP_STATUS_RUNNING)
+		{
+			catalog_lock(false, &run_pid);
+			if (run_pid == 0)
+			{
+				backup->status = BACKUP_STATUS_ERROR;
+				pgBackupWriteIni(backup);
+			}
+		}
+
 		show_backup_detail(stdout, backup);
 
 		/* cleanup */
@@ -184,7 +197,8 @@ get_parent_tli(TimeLineID child_tli)
 static void
 show_backup_list(FILE *out, parray *backup_list)
 {
-	int i;
+	int			i;
+	pid_t		run_pid = -1;
 
 	/* show header */
 	fputs("=========================================================================================\n", out);
@@ -193,14 +207,25 @@ show_backup_list(FILE *out, parray *backup_list)
 
 	for (i = 0; i < parray_num(backup_list); i++)
 	{
-		pgBackup *backup;
-		const char *modes[] = { "", "PAGE", "PTRACK", "FULL", "", "PAGE+STREAM", "PTRACK+STREAM", "FULL+STREAM"};
-		TimeLineID  parent_tli;
-		char timestamp[20] = "----";
-		char duration[20] = "----";
-		char data_bytes_str[10] = "----";
-
-		backup = parray_get(backup_list, i);
+		pgBackup   *backup = parray_get(backup_list, i);
+		const char *modes[] = {"", "PAGE", "PTRACK", "FULL", "", "PAGE+STREAM", "PTRACK+STREAM", "FULL+STREAM"};
+		TimeLineID	parent_tli;
+		char		timestamp[20] = "----";
+		char		duration[20] = "----";
+		char		data_bytes_str[10] = "----";
+
+		/* Fix backup status */
+		if (backup->status == BACKUP_STATUS_RUNNING)
+		{
+			if (run_pid == -1)
+				catalog_lock(false, &run_pid);
+
+			if (run_pid == 0 || i + 1 < parray_num(backup_list))
+				backup->status = BACKUP_STATUS_ERROR;
+
+			if (run_pid == 0)
+				pgBackupWriteIni(backup);
+		}
 
 		if (backup->recovery_time != (time_t) 0)
 			time2iso(timestamp, lengthof(timestamp), backup->recovery_time);

tests/restore_test.py

@@ -33,9 +33,14 @@ def test_restore_to_latest_1(self):
 		node.stop({"-m": "immediate"})
 		node.cleanup()
 
+		# 1 - Test recovery from latest
 		self.assertIn(six.b("INFO: restore complete"),
 			self.restore_pb(node, options=["-j", "4", "--verbose"]))
 
+		# 2 - Test that recovery.conf was created
+		recovery_conf = path.join(node.data_dir, "recovery.conf")
+		self.assertEqual(path.isfile(recovery_conf), True)
+
 		node.start({"-t": "600"})
 
 		after = node.execute("postgres", "SELECT * FROM pgbench_branches")
@@ -556,3 +561,60 @@ def test_restore_with_tablespace_mapping_12(self):
 		self.assertEqual(id[0][0], 2)
 
 		node.stop()
+
+	def test_restore_with_tablespace_mapping_13(self):
+		"""recovery using tablespace-mapping option and page backup"""
+		node = self.make_bnode('restore_with_tablespace_mapping_13',
+			base_dir="tmp_dirs/restore/restore_with_tablespace_mapping_13")
+		node.start()
+		self.assertEqual(self.init_pb(node), six.b(""))
+
+		# Full backup
+		self.backup_pb(node)
+		self.assertEqual(self.show_pb(node)[0].status, six.b("OK"))
+
+		# Create tablespace
+		tblspc_path = path.join(node.base_dir, "tblspc")
+		os.makedirs(tblspc_path)
+		with node.connect("postgres") as con:
+			con.connection.autocommit = True
+			con.execute("CREATE TABLESPACE tblspc LOCATION '%s'" % tblspc_path)
+			con.connection.autocommit = False
+			con.execute("CREATE TABLE tbl AS SELECT * FROM generate_series(0,3) AS integer")
+			con.commit()
+
+		# First page backup
+		self.backup_pb(node, backup_type="page")
+		self.assertEqual(self.show_pb(node)[1].status, six.b("OK"))
+
+		# Create tablespace table
+		with node.connect("postgres") as con:
+			con.connection.autocommit = True
+			con.execute("CHECKPOINT")
+			con.connection.autocommit = False
+			con.execute("CREATE TABLE tbl1 (a int) TABLESPACE tblspc")
+			con.execute("INSERT INTO tbl1 SELECT * FROM generate_series(0,3) AS integer")
+			con.commit()
+
+		# Second page backup
+		self.backup_pb(node, backup_type="page")
+		self.assertEqual(self.show_pb(node)[2].status, six.b("OK"))
+
+		node.stop()
+		node.cleanup()
+
+		tblspc_path_new = path.join(node.base_dir, "tblspc_new")
+		self.assertIn(six.b("INFO: restore complete."),
+			self.restore_pb(node,
+				options=["-T", "%s=%s" % (tblspc_path, tblspc_path_new)]))
+
+		# Check tables
+		node.start()
+
+		count = node.execute("postgres", "SELECT count(*) FROM tbl")
+		self.assertEqual(count[0][0], 4)
+
+		count = node.execute("postgres", "SELECT count(*) FROM tbl1")
+		self.assertEqual(count[0][0], 4)
+
+		node.stop()

validate.c

@@ -41,7 +41,7 @@ do_validate(time_t backup_id,
 	bool		success_validate,
 				need_validate_wal = true;
 
-	catalog_lock(false);
+	catalog_lock(false, NULL);
 
 	rt = checkIfCreateRecoveryConf(target_time, target_xid, target_inclusive);
 	if (rt == NULL)