MBG wiki | RecentChanges | Blog | 2024-04-20 | 2024-04-19

OOM killer enabled

We have had some problems with nodes dying under OOM (out-of-memory) conditions (nicelly reproduced by Georgoulia via VMD runs with 1.5 GBytes DCD files). The linux kernel normally deals with such problems by killing the processes that are responsible for the OOM condition. But Georgoulia kept on reproducibly crashing nodes, which got me to /usr/src/linux-2.4.22-openmosix3/mm/oom_kill.c :

......

#ifdef CONFIG_MOSIX_NO_OOM

/* We don't need no stinking OOM in openMosix
 * Rather, we just send a message to syslog that the system
 * is out of memory and let fate take over the system  - Moshe Bar  */

        printk("openMosix: this node is seriously out of memory due to VM overcommit. I think I will pass out now.\n");
  
        return;
#endif 

......

So, it seems that under oM the OOM killer is disabled, which perfectly explains the crashes. A new kernel has been compiled with a configuration flag set to include the killer. Preliminary tests with Georgoulias' VMD files showed that the new kernel correctly kills the offending VMD job.