[mythtv-users] swappiness

Ian Clark mrrooster at gmail.com
Mon Feb 15 14:51:39 UTC 2016


On 12 February 2016 at 21:01, Eric Sharkey <eric at lisaneric.org> wrote:
> On Fri, Feb 12, 2016 at 11:37 AM, Ian Clark <mrrooster at gmail.com> wrote:
>>>  I thought that the result of that
>>> wpould be a complete lockup of everything, including any dirty pages and the
>>> only way out would be a hard reset, with risk of disk trashing.
>>>
>> I believe on most distros the OOM killer will fire up it's shotgun and
>> start randomly* terminating things. (Which I'll be honest always seems
>> a little crazy to me, but it can be disabled).
>
> No, it's a lot more sophisticated than that.  There's a number of
> heuristics used to score all the processes and the kernel kills the
> process with the highest score.

You are of course correct. (The * was supposed to indicate the
'random' was a little tongue in cheek), however I still am not a big
fan. I don't like the non-determanistic (or determanistic if you can
examine all the heuristics at the point of memory exhaustion which is
unlikely) nature of it. Mind you this is probably wandering a little
off topic.
>
> If you have a process with a run-away memory leak, the OOM killer
> almost always targets that process.
>
'almost' :)

>
>> Assuming you're using a fs with journaling you should be fairly safe
>> from FS corruption.
>
> You don't even need journaling for that.  The kernel maintains the FS
> and it won't kill itself.  Journaling only helps for kernel panics or
> power losses.
>
Yes, those were the situations I was referring to. (The 'it might
crash horribly if I run out of memory' worry in the quoted post). :)

Ian


More information about the mythtv-users mailing list