Containers, Kubernetes and VXLAN

What they ask

Cloud positions (cloud engineer, cloud architect, and the newer cloud software R&D teams that build cloud native products for the global portfolio) get one technical conversation with a member of the team, preceded by a manager conversation that is mostly about salary, benefits and the form of the contract. One candidate who came through a referral reports that the position was offered through an external contracting company rather than a direct contract, so ask early which arrangement applies.

Questions reported from the technical conversation:

• Why run a workload in Docker instead of a VM? What does a container share with the host that a VM does not, and when is a VM still the right call?
• Kubernetes: the difference between a Deployment and a StatefulSet. When does a pod need a stable identity and its own persistent volume, and what happens to each kind on a rolling update and on a node failure?
• What is VXLAN and why is it used? Where does it sit relative to the physical network, what problem does it solve for multi host or multi tenant container networking, and what does the encapsulation cost?
• Scenario questions: a service is unreachable after a deploy, a node shows NotReady, a volume is full. Walk through what you check, in order.

A cloud engineer candidate from an earlier year adds a warning: the interviewers asked about items that were not on the CV and seemed surprised at the gaps, so restate your actual scope in the first two minutes.

What they look for

Operational depth over vocabulary. The Deployment versus StatefulSet answer should reach ordered pod names, headless services and what happens to the PVC, not stop at "stateful apps". The VXLAN answer should land on VNI, the overlay and underlay split and the MTU implication. Scenario answers are judged on the order of checks and on knowing which command or dashboard gives you each fact.