RFR: 8331560: Refactor Hotspot container detection code so that subsystem delegates to controllers [v3]

[Thomas Stuefe]

On Wed, 29 May 2024 17:42:13 GMT, Severin Gehwolf <sgehwolf at openjdk.org> wrote:

>> Please review this preparatory PR which is an enabler for some bugfix/alignment work (e.g. [JDK-8322420](https://bugs.openjdk.org/browse/JDK-8322420)). The idea is to delegate limit lookup to controllers (`memory`, `cpu`, etc.) from the overarching `CgroupSubsystem` class. This way we can - once the cg type has been detected - "adjust" a controller's path to the limit files when the controller is being created (on init) and then left alone. In this case, the two preparatory entry points are `CgroupUtil::processor_count()` taking a version-agnostic `CgroupCpuController` to do the actual look-up and `CgroupMemoryController::read_memory_limit_in_bytes()` for the same in terms of memory limits.
>> 
>> This enables setting the contoller's path to the interface files (wherever it might be in the hierarchy of it's original cgroup path), look up the limit and "freeze" the path once it found a - lower - limit.
>> 
>> It also ensures that both cgroup versions return `-1` or `-2` (`OSCONTAINER_ERROR`) - in both cases negative - for some notion of unlimited. I.e. it makes the upper bound by the host's physical memory apparent in the version agnostic classes.
>> 
>> Testing:
>> - [x] GHA
>> - [x] gtest:cgroupTest tests
>> - [x] container tests on Linux cgv1 and cgv2 on x86_64
>> 
>> Thoughts?
>
> Severin Gehwolf has updated the pull request with a new target base due to a merge or a rebase. The pull request now contains 30 commits:
> 
>  - Review feedback
>  - Clean up
>  - Appropriately handle version specific printing
>  - Fix inheritance hierarchy
>  - Fixup after merge
>  - Merge branch 'master' into jdk-8331560-cgroup-controller-delegation
>  - Resolve ambiguity with static_cast
>  - Remove limit_from_str from util class
>  - Merge branch 'jdk-8302744-cleanup-getcontainer-info' into jdk-8331560-cgroup-controller-delegation
>  - Add proper comments for parsing utility functions
>  - ... and 20 more: https://git.openjdk.org/jdk/compare/9b64ece5...e177a086

src/hotspot/os/linux/cgroupV1Subsystem_linux.hpp line 52:

> 50:     char *subsystem_path() { return _path; }
> 51: };
> 52: 

Drive by comment. I don't understand the aim behind the complexity, but multiple inheritance, even with pure abstract side classes, is rarely a good idea. It is also against the hotspot style guide:


Restrict inheritance to the "is-a" case; use composition rather than
non-is-a related inheritance.

Avoid multiple inheritance.  Never use virtual inheritance.


Most cases I have seen of multiple inheritance are driven by the wish to reuse code, but the added complexity and confusion (e.g. typical C++ IDEs cannot follow call chains) makes life for maintainers actually harder. Composition is usually a better way to do things.

-------------

PR Review Comment: https://git.openjdk.org/jdk/pull/19085#discussion_r1636962412