Whenever there's a significant amount of pages swapped, where significant seems to be merely 100 megabytes or so, shutting down the system properly becomes a matter of luck. I've read before on the mailing lists that swapping pages back in is rather, lets say, icky (resulting in the VM 2.0 initiative), so I'm blaming it, especially since the problems only appear if there was swapping. Anyway, it seems that swapping things back into memory during a parallel service shutdown by SMF causes funky behaviour.
Just earlier today, the system was caught spamming "Cannot start console-login, requesting maintenance mode" over my display until I've hard resetted it (I suppose maintenance mode requires console-login, catch-22). More often than not, it's usually the GDM service failing and requesting maintenance mode, requiring hands on to make the system shut down or reboot. So as such, is there a way to push all contents in swap back into memory? If not, there should be. If this was possible, we'd be able to do it somewhere in the service shutdown order, right before the core services but after the non-vital ones. So when it shuts down, it could make sure that all paged code is back in memory and prevent the service shutdown from breaking. Regards, -mg This message posted from opensolaris.org _______________________________________________ perf-discuss mailing list perf-discuss@opensolaris.org
