Advanced|23 hours|40 lessons

Running Databases and Stateful Systems on Kubernetes

Operating stateful workloads on Kubernetes: the operator pattern and its failure modes, CSI and the volume lifecycle, failover and leader election, consistent backups and point in time recovery, volume expansion and cross cluster migration, and the day to day of running a database somebody depends on. Lifetime updates included.

Text-based, no videos

9 modules, 40 lessons

Lifetime access

What you'll learn

Why a stateful pod cannot move, and what that removes from the scheduler

What a StatefulSet actually guarantees, and the longer list of things it does not

The operator pattern: what it adds, how to evaluate one, and why it is a failure domain you chose

CSI end to end, from provision through attach to mount, and where each step stalls

Failover and leader election, and what prevents two primaries

Why a volume snapshot is not a database backup, and what a consistent one requires

Point in time recovery, and the write log you have to have been keeping

Volume expansion, storage class migration, and moving a database between clusters

Tenant isolation for databases, and what the tenant distribution decides

The signals that predict a stateful incident rather than reporting one

Curriculum

9 modules · 40 lessons

What Makes Stateful Different

Pod identity bound to a volume in a zone, what a StatefulSet actually promises, and the failure modes that have no stateless equivalent.

4 lessons

The Pod That Cannot Move30 minFREE What a StatefulSet Actually Guarantees35 minFREE Why Databases Resist Orchestration30 minFREE The Failure Modes With No Stateless Equivalent30 minFREE

The Operator Pattern for Stateful Software

What an operator adds beyond a StatefulSet, how to evaluate one before adopting it, and the fact that it is a failure domain you chose.

6 lessons

What an Operator Does That a StatefulSet Cannot35 min Reading an Operator Before You Adopt It40 min The Operator as a Failure Domain35 min Upgrading an Operator Under a Live Database35 min When the Operator and Reality Disagree35 min Building Without an Operator30 min

CSI and the Volume Lifecycle

The plugin split, the lifecycle from provision to mount, topology binding, and what access modes actually mean.

5 lessons

Controller and Node Plugins35 min Provision, Attach, Mount, and Where It Hangs40 min Topology and Zone Binding35 min Access Modes and What They Actually Mean30 min StorageClass Design for a Platform30 min

Running a Database

Sizing, failover, connections, replicas, schema changes and version upgrades, on a workload that cannot simply be restarted.

6 lessons

Sizing a Database Pod35 min Failover and Leader Election40 min Connection Management and Pooling30 min Read Replicas and Scaling Reads35 min Schema Migrations Against a Live Cluster35 min Major Version Upgrades40 min

Backup, Restore, and Recovery

Why a volume snapshot is not a database backup, consistent backups, point in time recovery, and rehearsing the restore.

5 lessons

Why Volume Snapshots Are Not Database Backups35 min Consistent Backups for a Running Database35 min Point in Time Recovery40 min Restore Rehearsal and the Recovery Clock35 min Cross Cluster and Cross Region Recovery35 min

Data Lifecycle Operations

Expansion, migration between classes and clusters, snapshots as tools, and decommissioning with evidence.

5 lessons

Volume Expansion Without Downtime30 min Migrating Between Storage Classes35 min Snapshots and Clones as Operational Tools30 min Moving a Database Between Clusters40 min Decommissioning and Data Retention30 min

Multi-Tenancy and Isolation

Shared storage contention, per-tenant against shared databases, resource isolation, and credentials that rotate.

4 lessons

Noisy Neighbours on Shared Storage35 min Per-Tenant Database Against Shared With Isolation35 min Resource Isolation for Stateful Workloads30 min Credentials and Rotation30 min

Operating It Day to Day

The signals that matter, capacity you can see coming, the slow degradations, and a runbook someone can follow at three in the morning.

4 lessons

What to Monitor for a Stateful Workload35 min Capacity: Watching a Disk Fill Before It Does30 min The Slow Degradations35 min Runbooks and the On Call Story30 min

Capstone

A full stateful platform designed end to end, with the tradeoff stated for every decision.

1 lessons

Capstone: Design the Stateful Platform60 min

About the Author

Sharon Sahadevan

AI Infrastructure Engineer

Building production GPU clusters on Kubernetes. H100s, large-scale model serving, and end-to-end ML infrastructure across Azure and AWS.

10+ years designing cloud-native platforms with deep expertise in Kubernetes orchestration, GitOps (Argo CD), Terraform, and MLOps pipelines for LLM deployment.

Author of KubeNatives, a weekly newsletter read by 3,000+ DevOps and ML engineers for production insights on K8s internals, GPU scheduling, and model-serving patterns.

Ready to master this topic?

Start with the free preview lesson and see for yourself.