May 20, 202612 min readDevOps

Kubernetes Pod Startup Critical Path Analysis

Break down cold-start latency from scheduler queue to readiness probe success, with specific tactics for reducing startup p99 in multi-tenant clusters.

Startup delay is a chain problem: image pull, cgroup setup, sidecar init, app boot, readiness checks. Fixing one link rarely solves p99.

Instrument the Full Timeline

Scheduler enqueue to bind.
Node pull and unpack.
Container create and entrypoint.
Probe pass and endpoint publish.

startupProbe:
  httpGet:
    path: /healthz
    port: 8080
  periodSeconds: 2
  failureThreshold: 40

Treat startup latency as a product metric, not just platform noise.