WIP: SlapObject: when destroying a partition, no residues from user activity should remain
When users connect to a slappart by ssh, they can run or plan tasks running in the background. As these aren't run by supervisord, they are not stopped when the slappart is destroyed, and they remain even when the slappart is attributed to another user.
In the consequence of this, instanciation errors can appear later if the services run in background are using some ports that the new user wants also to use, or even worst.
This patch proposes to kill all the processes belonging to the slapuser when its slappart is destroyed, and clean the user's crontabs and at jobs.
-
@Rafael, @alain.takoudjou : what do you think of this merge request ?
I also need your advices on how can I test this feature ?
-
Also, despite the use of
x.pid != current_pid, I'm afraid this commit as such would kill a testnode. -
Changed title: WIP: SlapObject: when destroying a partition, no processes should be remaining → WIP: SlapObject: when destroying a partition, no residues from user activity should remain
Toggle commit list -
726 732 except IOError as exc: 727 733 raise IOError("I/O error while freeing partition (%s): %s" % (self.instance_path, exc)) 728 734 735 # Kill remaining processes 736 try: 737 current_pid = os.getpid() 738 [x.terminate() for x in psutil.process_iter() if uid in x.uids() 739 and x.pid != current_pid] 740 except (psutil.NoSuchProcess, psutil.AccessDenied) as e: 741 self.logger.info("Process finished before being " -
Maybe self.logger.warn() here is better I think
-
-
726 732 except IOError as exc: 727 733 raise IOError("I/O error while freeing partition (%s): %s" % (self.instance_path, exc)) 728 734 735 # Kill remaining processes 736 try: 737 current_pid = os.getpid() 738 [x.terminate() for x in psutil.process_iter() if uid in x.uids() -
terminate (kill with SIGTERM) will not work at 100%, maybe the use of kill() will be required if the process is still alive, as you have the list of all process (parents and children).
-
-
In general is ok the idea, but I think the kill should be more aggressive probably.
If it fails to kill (for whatever reason) the destroy should be consider unsuccessfull, so next run of slapos node report should retry try it again on the next run or something.
-
Please register or sign in to post a comment