topology-aware: exclude isolated CPUs from policy-picked reserved cpusets. #474
Conversation
klihub
changed the title
fmuyassarov
changed the title
klihub
force-pushed
the
fixes/avoid-isolated-reserved
branch
from
0edeb25 to
caea3ee
Compare
There was a problem hiding this comment.
Looks good.
One small nit: would it make sense to include this logging commit caea3ee in a related commit (598ec9c) instead of a separate one? Just thinking about keeping the commit history clean, but I'm okay to keep as it is as well if there is a reason behind.
Just to clarify for myself, you are not allowing to have more than one isolated CPU for the reserved cpuset because isolated CPUs don't participate in load balancing right?
I wonder if we should do the similar check in Balloons as well...
The reason was mostly that even if I left one of those other commits out, I'd want to log the effective reserved CPU set on startup/reconfiguration, so based on that it did not seem to fit that clearly into either of those. But I can squash it, but I it think it fits better to the latter one, not the former.
For policy-allocated reserved cpuset, I think we should have similar restrictions for isolated CPUs. They try to avoid setups which are problematic from a scheduler/load-balancing point of view. |
Accept a single hand-picked isolated CPU as the reserved cpuset, but give a warning about it being isolated. Reject the reserved cpuset if it mixes isolated with normal CPUs or if it contains more than a single isolated CPU. Signed-off-by: Krisztian Litkey <[email protected]>
Exclude isolated CPUs if we pick reserved CPUs by CPU (resource) quantity. Log effective reserved cpuset. Signed-off-by: Krisztian Litkey <[email protected]>
klihub
force-pushed
the
fixes/avoid-isolated-reserved
branch
from
caea3ee to
954d61c
Compare
This PR updates the reserved cpuset setup code to