---
title: "Designing Kubernetes workloads"
chapter: "06"
---

# Designing Kubernetes workloads

A production workload is more than a Deployment and a Service.

## Stateless service baseline

Use a Deployment with multiple replicas, rolling-update strategy, readiness,
startup and liveness probes, resource requests, sensible limits, topology
spread, Pod disruption budget, security context, dedicated service account,
and controlled configuration.

## Graceful behavior

When Kubernetes terminates a Pod it sends a signal and waits for a grace period.
The application should stop accepting new work, finish or hand off safe work,
close connections, and exit. Align load-balancer draining, readiness, signal
handling, and termination grace.

## Configuration

Keep environment-specific configuration outside images. ConfigMaps are not
secret stores. For sensitive values, prefer Secret Manager integration or
carefully governed Kubernetes Secrets with encryption, access, and rotation.

## Stateful work

Kubernetes can run stateful software, but a managed database often reduces risk.
If state must run in GKE, define stable identity, storage class, replication,
anti-affinity, quorum, backup, upgrades, disruption, and recovery. A StatefulSet
does not create database replication for you.

## Batch and events

Use Jobs and CronJobs for finite work, with concurrency and history limits.
For event-driven scale-to-zero use cases, evaluate Knative-based platforms or
KEDA where supported and governed. Never let an uncontrolled retry create an
infinite expensive workload.