A Kubernetes deployment works when nothing goes wrong, but does it really survive maintenance, a saturated node or a load spike? Making it reliable means configuring the cluster and the workloads so that availability is guaranteed by design, not by luck.
Four mechanisms cover most of application reliability: PodDisruptionBudget
objects for maintenance, anti-affinity for spreading replicas, probes to make
the Pod lifecycle visible, and requests and limits for stable
scheduling.
What you will learn
This page is an overview of the four settings that keep an application available. By the end, you will be able to:
- explain what a
PodDisruptionBudgetprotects during maintenance, and read one; - read an anti-affinity rule that spreads replicas across several nodes;
- tell the
readiness,livenessandstartupprobes apart; - explain why
requestsandlimitsmatter for evictions; - find the lesson that teaches each setting in depth.
The 4 pillars of application availability
1. PodDisruptionBudget (PDB)
A PodDisruptionBudget guarantees that Kubernetes does not remove more than a given number of pods during a voluntary operation (drain, update). It does not protect against a sudden failure of a node: a failure is not a voluntary operation, and no budget can prevent it.
apiVersion: policy/v1kind: PodDisruptionBudgetmetadata: name: mon-app-pdbspec: minAvailable: 2 # or maxUnavailable: 1 selector: matchLabels: app: mon-app2. Anti-affinity
Anti-affinity prevents several replicas of the same deployment from being placed on the same node, so a node failure does not take down the whole application.
affinity: podAntiAffinity: requiredDuringSchedulingIgnoredDuringExecution: - labelSelector: matchLabels: app: mon-app topologyKey: kubernetes.io/hostnameThe topologyKey says at what scale the replicas must be kept apart: with
kubernetes.io/hostname, the unit of separation is the node. The
requiredDuringScheduling... form is a strict rule: if no free node fits,
the Pod stays Pending rather than share a node.
3. Health probes
Three probes watch a container, and each answers a different question:
- Readiness: Kubernetes routes traffic to a pod only if it is ready to handle requests.
- Liveness: Kubernetes restarts a stuck container, for example on a deadlock.
- Startup: it avoids liveness false positives on a slow start. Until it succeeds, Kubernetes suspends the other two probes.
4. Requests and limits
requests define the resources guaranteed at scheduling time. limits cap
consumption. A Pod without requests can be evicted at any time as soon as
memory pressure appears on the node.
Where can you go deeper on each pillar?
This page gives the overview; each pillar has its own lesson in the course. The PodDisruptionBudget is practiced with the cordon, drain, uncordon cycle of preparing a maintenance (in French), where it shows its value. Anti-affinity and replica spreading belong to the lesson Affinities, taints and tolerations, probes to the lesson on probes, and requests and limits to their dedicated lesson.
The guide Application availability on Kubernetes (in French) then brings PodDisruptionBudget, anti-affinity and probes together in a single scenario, to check that service continuity holds when a node disappears.
Key takeaways
- Without a PDB, a
kubectl draincan evict all the replicas running on the node at once - Without anti-affinity, all your pods can land on the same node, and your deployment rests on a single point of failure
- Without a readiness probe, Kubernetes routes traffic to a pod that is not ready yet
- Without requests, the scheduler cannot guarantee the resources a pod needs, and evictions are unpredictable
- Combining PDB, anti-affinity, probes and requests covers most of resilience with no extra infrastructure
Knowledge check
Check that you have grasped the essentials of this guide. The questions cover only what has just been explained here.
Knowledge check
Check what you have learned with this interactive quiz
How it works
- The timer starts when you click Start
- Multiple choice, true/false and short answer questions
- You can move between questions
- Detailed results are shown at the end
Starts the quiz and the timer
Review
(0/0)Score by domain
A domain groups several guides and gets several questions, so its score is more reliable than a per-topic score, which often rests on a single question.
Summary
Strengths
No topic passed at 80% on at least two questions yet.
Weak areas
No topic below the pass mark.
Recommended next
Hints before you try again?
Continue the course: Kubernetes operations routine: daily, weekly and monthly
A new full quiz with random questions
Work again only on the questions you missed