@@ -330,23 +330,43 @@ def version(self):
330330 """
331331return self ._pg_version
332332
333- def _try_shutdown (self ,max_attempts ):
333+ def _try_shutdown (self ,max_attempts , with_force = False ):
334334attempts = 0
335+ node_pid = self .pid
335336
336337# try stopping server N times
337338while attempts < max_attempts :
338339try :
339- self .stop ()
340+ self .stop (wait = False )
340341break # OK
341342except ExecUtilException :
342343pass # one more time
343344except Exception :
344- # TODO: probably should kill stray instance
345345eprint ('cannot stop node {}' .format (self .name ))
346346break
347347
348348attempts += 1
349349
350+ # If force stopping is enabled and PID is valid
351+ if with_force and node_pid != 0 :
352+ # If we couldn't stop the node
353+ p_status_output = self .os_ops .exec_command (cmd = f'ps -p{ node_pid } ' ,shell = True ).decode ('utf-8' )
354+ if self .status ()!= NodeStatus .Stopped and p_status_output and str (node_pid )in p_status_output :
355+ try :
356+ eprint (f'Force stopping node{ self .name } with PID{ node_pid } ' )
357+ self .os_ops .kill (node_pid ,signal .SIGKILL ,expect_error = False )
358+ except Exception :
359+ # The node has already stopped
360+ pass
361+
362+ # Check that node stopped
363+ p_status_output = self .os_ops .exec_command (f'ps -p{ node_pid } ' ,shell = True ,expect_error = True ).decode ('utf-8' )
364+ if p_status_output and str (node_pid )in p_status_output :
365+ eprint (f'Failed to stop node{ self .name } .' )
366+ else :
367+ eprint (f'Node{ self .name } has been stopped successfully.' )
368+
369+
350370def _assign_master (self ,master ):
351371"""NOTE: this is a private method!"""
352372
@@ -935,6 +955,11 @@ def cleanup(self, max_attempts=3, full=False):
935955rm_dir = self .data_dir # just data, save logs
936956
937957self .os_ops .rmdirs (rm_dir ,ignore_errors = False )
958+ is_not_removed = self .os_ops .isdir (rm_dir )
959+ while is_not_removed :
960+ time .sleep (1 )
961+ self .os_ops .rmdirs (rm_dir ,ignore_errors = False )
962+ is_not_removed = self .os_ops .isdir (rm_dir )
938963
939964return self
940965