* Use containerd cgroups pkg to read cgroups stats

* Add slurm_job to metric name to be consistent across collectors

* Add cpu and memory PSI metrics

* Walk through pids of cgroup only to get job details

Signed-off-by: Mahendra Paipuri <[email protected]>

* Test when slurm prolog files are not present

* This tests if we are able to read procfs correctly and get env vars

* Update test fixtures

Signed-off-by: Mahendra Paipuri <[email protected]>

Signed-off-by: Mahendra Paipuri <[email protected]>

Signed-off-by: Mahendra Paipuri <[email protected]>

* Merge GPU jobID map with slurm collector. This is more logical organization instead of having a separate collector

* Dont report swap and PSI metrics by default. They can be enabled  using CLI flag

* Add a hidden flag to force cgroups version for testing

* Refacorting of certain receivers to be more clean

* Add more unit tests to cover more scenarios

* Add test fixtures to be able to unit test

* Add more e2e test scenarios

Signed-off-by: Mahendra Paipuri <[email protected]>

Signed-off-by: Mahendra Paipuri <[email protected]>

* Iteration over map is undefined in go and not reproducible

* To ensure we always have same behaviour we use int as map index and iterate over range

* This is done to avoid unit test failures as order in slice gpuOrdinals is important in cmp

Signed-off-by: Mahendra Paipuri <[email protected]>