Imperative vs Declarative

Kubernetes is declarative: i.e. it focuses on the what is the desired result, rather than how to make the desired result.

Imperative Command

• Make 2 containers: depend on the current state, and thus the final result is affected by the initial state.

Declarative Command

• There should be 2 containers: only depend on the declarative statement, regardless of the inital/current state.

Reconciliation

Reconciliation: Managing the state of containers by consistently monitoring the bridge - the difference between the desired and current state - and adjusting the configuration.

Value for developers

With reconciliation, developers gain value in 4 aspects:

• Declarative: All actions are performed declaratively.

• Idempotency: Applying the same definition multiple times always yields the same result.

• Self-Healing: Automatically recovering from unexpected inconsistencies in the current state.

• Consistency: Aligning the state with the final specification regardless of the current state or operations in Kubernetes.

Reconciliation Example: Container Version Update

Assume the following situation:

• desired state: container built from image version 1.0.1
• current state: container is built from image version 1.0.0

If my kubernetes cluster do not have an image with version 1.0.1, what would happen?

Because a bridge is monitored during reconciliation, the kubernetes cluster will try to create a new contiainer with image version 1.0.1. However, because there is no such image, kubernetes will fail to build the container, so the current container has image version as1.0.0. The bridge is still monitored, so kubernetes eventually enters an infinite loop.

Reconciliation Example: Restarting the container

Restarting containers is a bit ambiguous in Kubernetes because the concept of restarting is not easily defined in a declarative manner. While Kubernetes supports imperative commands that allow you to achieve restarts, it's not inherently designed for straightforward restarts.

If you want to restart a container, one approach is to remove the container, which effectively creates a bridge between the current state and the desired state. Alternatively, you can set the container count to 0 and then increase it again to achieve a restart.

Strength of Self-Healing

• Automatically corrects container states if they deviate from the desired state.
• Replace containers with new ones if processes terminate or containers become unhealthy.
• When utilized in an operational environment without user intervention, these features become incredibly powerful.

What should developers do?

Therefore, what developers should do is:

• Build applications that are easy for Kubernetes to observe.
• Define the desired final state accurately.
• Implement systematic version control or automated build and deployment processes.
• Develop applications that can handle sudden termination or restart without issues.