On Apr 10, 2005 10:50 PM, JM Ibanez <[EMAIL PROTECTED]> wrote: > On Apr 8, 2005 2:16 PM, ian sison (mailing list) <[EMAIL PROTECTED]> wrote: > > Secondly, there are certain problems associated with deploying LTSP > > solutions: > > > > - for one reason or another, there are cases when the remote session > > disconnects from the server (power failure? network problems?) So the > > desktop is rebooted, but alas, the programs the user was running are > > still running on the server. Now the sysad has to kill each program > > one by one... There is no 'watchdog' program that takes care of this > > yet, at least not to my knowledge. GNOME is also notorious for > > leaving behind some programs even though the user has already logged > > out. Some say this is may be intentional, I say it's a bug. > > Hmm... would be simple to write a shell script which executes at > logout (i.e. ~/.bash_logout or similar) that kills all tasks running > as that user. But then again, I'm not well-versed with LTSP... ;)
The thing is, the user doesn't "log out" at all, because the session is left stale. For example, if the client PC is disconnected for reasons that Ian mentioned (e.g. power interruption), then the server still maintains the session, thereby leaving a lot of "zombie" processes. Debian-edu has 2 scripts worth checking out: nightkill.sh and logoutkill.sh. Not exactly the watchdog feature Ian mentioned, but they're worthwhile janitorial scripts. > > - in the course of running LTSP based systems, it may be that one > > user, possibly not intentionally, can take up *a lot* of resources, > > both CPU and memory, thus slowing everyone else down. Sometimes, > > untrained users (or those who are truly impatient) contribute to this > > by repeatedly clicking an icon, thereby launching multiple instances > > of a program - bringing the entire network to a crawl.There should be > > a way to limit the amount of resources a user can consume, so that > > there will be no possibility of an accidental (or intentional) DoS. > > Isn't this a tunable on all Linux systems? IIRC, can't remember > exactly off-hand, something to do with the maximum number of processes > you can spawn or what not. Or maybe someone could write a kernel patch > to limit the number of processes a user can spawn within a certain > time limit (i.e. 10 processes per second at the most, after that the > kernel just kills processes?) > > Of course, disclaimer: I'm not well-versed with LTSP nor have I deployed LTSP. The overall desired effect is to limit a user's use of certain resources -- CPU, memory, disk, etc. It's not easy to "jail" a user in this manner. Meaning, we can't say that, if there are 10 users logged on, give these users about 10% of CPU resources. A user can easily bring down the system by doing something CPU intensive -- e.g. using OO.org to edit a complex document or visiting a Flash-heavy site. I've tried an approach involving renicing a user's processes based on resource usage, but this didn't play too well in a large scale setting, mainly because the turnaround/response times from the user's POV became unaccaptable (users in general don't like waiting too long). gino ledesma -- Philippine Linux Users' Group (PLUG) Mailing List [email protected] (#PLUG @ irc.free.net.ph) Official Website: http://plug.linux.org.ph Searchable Archives: http://marc.free.net.ph . To leave, go to http://lists.q-linux.com/mailman/listinfo/plug . Are you a Linux newbie? To join the newbie list, go to http://lists.q-linux.com/mailman/listinfo/ph-linux-newbie
