Virtualization · Cluster
Platform Engineering
Internal platforms and golden paths.
- 01 What a Golden Path Guarantees: The Platform Contract as a Versioned Interface A golden path is a contract only when it declares its inputs, the runtime facts it provides, and the platform versions it tracks. What that declaration contains, and what breaks when a piece of it is missing.
- 02 Tenant Isolation in an Internal Platform: Which Boundary Does the Work Namespaces, service accounts, resource quotas, and network policy each stop a different class of interference. Choosing the wrong one leaves a gap that looks like isolation but is only naming.
- 03 Reconciliation Loops: Why Level-Triggered Control Beats Event Handling A platform that compares desired state to observed state on every pass recovers from missed events and partial failures. The mechanism, its cost in API traffic, and the edge cases where a loop oscillates.
- 04 Admission Control as the Enforcement Point for a Golden Path Validating and mutating admission sit between the tenant and the stored object. How they differ, why mutation order is undefined, and what a rejection should tell the tenant.
- 05 Propagating Template Changes Across an Instantiated Fleet Once a golden path has been copied a hundred times, a change to the template is a migration. Three propagation models, the merge conflicts each produces, and how to make a breaking template change survivable.
- 06 Quota Arithmetic: Why a Shared Pool Cannot Promise Every Tenant Its Share Requests, limits, and overcommit interact through the scheduler and the cgroup hierarchy. How to compute whether a set of quotas fits a pool, and where the arithmetic stops being safe.
- 07 Ephemeral Preview Environments: Naming, Lifetime, and Cost Control A per-change environment needs a deterministic name, a lifecycle bound to something that can be observed, and a cost ceiling. Where each of those decisions fails, and why cleanup is the hard part.
- 08 Cluster Segmentation: Choosing a Blast Radius Instead of Accepting One Control plane failure, node failure, and tenant misbehaviour propagate differently. How to decide what shares a cluster, what must be separated, and what separation costs in operational surface.
- 09 Service Level Objectives for a Platform, Not for a Service A platform SLO measures whether tenants can get work done, not whether a component is up. How to define the request journey, choose indicators that tenants feel, and handle the shared-dependency problem.
- 10 Sunsetting a Golden Path Without Breaking the Tenants on It Retiring a path means migrating every instantiated copy. How to inventory the fleet, freeze adoption, migrate in waves, and decide when a path is retired rather than repaired.