Self hosted, v 3.4.25, Kubernetes worker Concurrency limit set to 1 at deployment level, policy set to cancel new Common behavior: duplicate runs start, several seconds apart, one gets cancelled -- Why, has anyone seen this? Example yaml deployment config:

- name: other
  description: XXX
  entrypoint: XXXX:build_other
  schedules:
      - cron: "0 6 * * *"
        timezone: "Europe/Prague"
        active: true
  parameters: {}
  work_pool: *transformations
  pull: *work_dir
  concurrency_limit:
    limit: 1
    collision_strategy: "CANCEL_NEW"