Use cgroup to calculate available memory

[ntovas]

Trying to fix #49317

It seems that on GCToOSInterface::GetMemoryStatus the GetPhysicalMemoryLimit works by checking first cgroup limits, but GetAvailablePhysicalMemory uses /proc/meminfo that in a containerized environment it will report the available memory of the worker and not of the pod.
This has as a result the check if (total > available) will be in a lot of cases false (when the worker available memory is more than the container's limit) and to report back to gc inconsistent data.

This is more a trigger for a discussion than a ready PR, since I have no experience in c++, neither in gc. We stumble upon this problem in the organization I am working, and this is an attempt to solve a gc related performance degradation.

[ghost]

Tagging subscribers to this area: @dotnet/gc
See info in area-owners.md if you want to be subscribed.

Issue Details

---

Trying to fix #49317

It seems that on GCToOSInterface::GetMemoryStatus the GetPhysicalMemoryLimit works by checking first cgroup limits, but GetAvailablePhysicalMemory uses /proc/meminfo that in a containerized environment it will report the available memory of the worker and not of the pod.
This has as a result the check if (total > available) will be in a lot of cases false (when the worker available memory is more than the container's limit) and to report back to gc inconsistent data.

This is more a trigger for a discussion than a ready PR, since I have no experience in c++, neither in gc. We stumble upon this problem in the organization I am working, and this is an attempt to solve a gc related performance degradation.

Author:	ntovas
Assignees:	-
Labels:	area-GC-coreclr
Milestone:	-

[mangod9]

@janvorli to review.