Platform and delivery
I work with applications in both models: running on a Windows server with IIS and running in a container on Kubernetes. Knowing my way around both is the skill, because while both are live, picking a side is not an option.
What I do
- Package .NET applications into images, with a private registry, versioning and tag immutability
- Take to containers applications that were not born for it
- Declare workers as scheduled tasks of the cluster, with the schedule outside the application code
- Diagnose a delivery that failed by reading pipeline, registry and sync state
- Use ephemeral containers in integration tests, to prove behavior against a real database
- Bring up local development dependencies with compose
Two things that have already bitten me
Immutable image tags. Rewriting different content onto the same tag breaks synchronization silently, and the cluster thinks it is up to date. Refusing the rewrite is what saves you from spending an afternoon looking for why the new version did not go up.
The desired state is a commit. Rollback is a revert, not a command applied by someone, and that changes how you investigate an incident: the history of what was live stays readable.
Where I stop
I do not administer clusters. I do not write base charts, I do not configure networking or security policy, and when the problem is in the platform I am not the one who solves it.
What I do is understand enough to take the application there, separate what is an application problem from what is an infrastructure problem, and give the right context to whoever looks after it. It is a comfortable position to hold and worth stating, because a developer who does not know where their own application runs turns every incident into a ticket for another team.
I also have no experience with service mesh or with distributed observability through tracing.