Thread (7 messages) 7 messages, 2 authors, 2016-06-03

Re: [PATCH] Force processes to non-realtime before mm_exit

From: Brian Silverman <hidden>
Date: 2016-06-03 23:34:04
Also in: lkml

Sebastian had some questions about this patch when I first sent it to rt-users.

On Wed, May 25, 2016 at 12:54 PM, Sebastian Andrzej Siewior
[off-list ref] wrote:
On 05/25/2016 08:00 PM, Brian Silverman wrote:
quoted
quoted
Why can't the application drop the RT priority before its exit? Wouldn't
that be appropriate?
If it crashes or gets killed, it doesn't have a chance to drop priority.
That is correct. The task with the highest priority is usually one of
the most important ones. Usually if that task crashes or gets killed by
the OOM killer while in production you have usually bigger problems
than this.
I sometimes use high priority for low latency rather than because a
process is important. Processes like that are designed to always run
quickly, but they need to run with low latency, so they're high
priority. However, they should never cause high latencies for other
processes, even if they crash.

Also, when something bad does happen and an important process crashes,
it's nice to have it not cause problems for other processes too.
Dumping core or freeing memory pages is never more important than
continuing to run the rest of the system.
I'm neither pro nor against this patch. This patch can go actually
upstream if accepted since it is not RT specific. If you have a good
use case please submit please post it upstream and CC me. Once accepted
I would pull it in -RT as well.
Done

Brian
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help