So a job is nonsense, a pod can't be constructed, or Kubernetes won't accept it. Currently, we build a whole 'nother pod containing an agent, a command container, an init container, etc using the default agent image, and tell it to...
echo "the error message" && exit 1
which smacks of cracking a walnut with the Mesta 50000.
This PR changes it so that the scheduler uses the new agent functionality introduced in buildkite/agent#2915 - core.Controller and core.JobController - to acquire and fail the job directly within the agent-stack-k8s controller.
So a job is nonsense, a pod can't be constructed, or Kubernetes won't accept it. Currently, we build a whole 'nother pod containing an agent, a command container, an init container, etc using the default agent image, and tell it to...
which smacks of cracking a walnut with the Mesta 50000.
This PR changes it so that the scheduler uses the new agent functionality introduced in buildkite/agent#2915 -
core.Controllerandcore.JobController- to acquire and fail the job directly within the agent-stack-k8s controller.Fixes #273