Support metrics-based workload discovery #59
Description
Is your feature request related to a problem? Please describe.
In some cases, we want to run the KRR program locally. But for the security consideration, the API server of the Kubernetes cluster cannot be accessed outside of the cluster.
So we can use the Prometheus-based workload discovery if kube-state-metrics is installed.
Describe the solution you'd like
We can do the workload based discovery with the following steps,
- List Deployments together with their ReplicaSets,
replicasets = await self.metrics_loader.loader.query("count by (namespace, owner_name, replicaset) (kube_replicaset_owner{"
f'namespace=~"{ns}", '
'owner_kind="Deployment"})')- List Pods from a group of ReplicaSets
# owner_name is ReplicaSet names
pods = await self.metrics_loader.loader.query("count by (owner_name, replicaset, pod) (kube_pod_owner{"
f'namespace="{namespace}", '
f'owner_name=~"{owner_name}", '
'owner_kind="ReplicaSet"})')- List containers from Pods got from step (2)
containers = await self.metrics_loader.loader.query("count by (container) (kube_pod_container_info{"
f'namespace="{namespace}", '
f'pod=~"{pod_selector}"'
"})")- Build K8sObjectData for containers got from step (3)
async def __build_from_owner(self, namespace: str, app_name: str, containers: List[str], pod_names: List[str]) -> List[K8sObjectData]:
return [
K8sObjectData(
cluster=None,
namespace=namespace,
name=app_name,
kind="Deployment",
container=container_name,
allocations=await self.__parse_allocation(namespace, "|".join(pod_names), container_name), # find
pods=[PodData(name=pod_name, deleted=False) for pod_name in pod_names], # list pods
)
for container_name in containers
]Describe alternatives you've considered
A clear and concise description of any alternative solutions or features you've considered.
Additional context
Add any other context or screenshots about the feature request here.