xDev Asia

Deployment strategies: RollingUpdate vs Recreate. Kubectl rollout commands, maxUnavailable/maxSurge. Revision history và rollback kỹ thuật cho CKAD.

Rolling Update và Rollback — maxUnavailable, maxSurge, ReplicaSet history

1. Deployment Strategies

Strategy	Cách hoạt động	Downtime?	Khi dùng
RollingUpdate	Replace pods dần dần, maintain availability	Không	Default, production
Recreate	Kill tất cả pods cũ, tạo mới	Có	Dev/test, breaking changes

spec:
  strategy:
    type: RollingUpdate
    rollingUpdate:
      maxUnavailable: 1   # OR "25%"  — Max pods unavailable during update
      maxSurge: 1         # OR "25%"  — Max extra pods above desired count

      ┌─────────────────────────────────────────────────┐
      │ Desired: 4 pods                                  │
      │                                                  │
      │ maxUnavailable: 1 → min 3 pods must be running  │
      │ maxSurge: 1       → max 5 pods total at once    │
      │                                                  │
      │ Step 1: Create 1 new pod (5 total = desired+surge)│
      │ Step 2: Terminate 1 old pod (4 total)           │
      │ Step 3: Repeat until all replaced               │
      └─────────────────────────────────────────────────┘

Exam tip: maxUnavailable và maxSurge KHÔNG thể cùng lúc là 0. Nếu cần zero-downtime update: set maxUnavailable: 0 và maxSurge: 1 (hoặc cao hơn).

2. kubectl rollout Commands

# Xem trạng thái rollout
kubectl rollout status deployment/myapp

# Xem revision history
kubectl rollout history deployment/myapp
kubectl rollout history deployment/myapp --revision=2

# Rollback về version trước
kubectl rollout undo deployment/myapp

# Rollback về revision cụ thể
kubectl rollout undo deployment/myapp --to-revision=2

# Tạm dừng rollout
kubectl rollout pause deployment/myapp

# Resume rollout
kubectl rollout resume deployment/myapp

Command	Tác dụng
`rollout status`	Wait/show current rollout progress
`rollout history`	List revision history
`rollout undo`	Rollback to previous (or specific) revision
`rollout pause/resume`	Pause để canary test, rồi resume
`rollout restart`	Force restart tất cả pods (rolling)

Exam tip: Để lưu CHANGE-CAUSE trong revision history, thêm annotation: kubectl annotate deployment/myapp kubernetes.io/change-cause="Updated image to v2" TRƯỚC khi update. Hoặc dùng --record flag (deprecated nhưng vẫn hoạt động trong exam).

3. Trigger & Monitor Update

# Update image (trigger rolling update)
kubectl set image deployment/myapp container-name=nginx:1.25

# Xem ReplicaSet history (mỗi update tạo mới 1 RS)
kubectl get rs
# NAME              DESIRED   CURRENT   READY
# myapp-7d9b8c      4         4         4     ← current
# myapp-6f5a2b      0         0         0     ← old (kept for rollback)

# Scale deployment
kubectl scale deployment/myapp --replicas=6

# Edit deployment trực tiếp
kubectl edit deployment/myapp

4. Revision History Limit

spec:
  revisionHistoryLimit: 10  # Default: 10 old RS kept for rollback
                             # Set to 0 to disable rollback capability

5. Cheat Sheet

Tình huống exam	Command
Update image	`kubectl set image deploy/app c=image:v2`
Check rollout	`kubectl rollout status deploy/app`
Rollback nhanh	`kubectl rollout undo deploy/app`
Rollback về rev 3	`kubectl rollout undo deploy/app --to-revision=3`
Zero-downtime config	`maxUnavailable: 0, maxSurge: 1`

6. Practice Questions

Q1: A Deployment with 10 replicas is configured with maxUnavailable: 2 and maxSurge: 3. During a rolling update, what is the maximum number of pods that can exist at any given time?

A) 10
B) 12
C) 13 ✓
D) 15

Explanation: maxSurge=3 means up to 3 extra pods above the desired count (10) can exist simultaneously. So maximum = 10 + 3 = 13 pods. Meanwhile, maxUnavailable=2 means at least 8 pods must be available.

Q2: You updated a Deployment and then realized the new version has a bug. Which command quickly reverts to the previous working version?

A) kubectl delete deployment myapp && kubectl apply -f old.yaml
B) kubectl rollout undo deployment/myapp ✓
C) kubectl rollout history deployment/myapp
D) kubectl set image deployment/myapp container=old-image

Explanation: rollout undo is the fastest way to revert to the previous revision. It creates a new rolling update back to the previous ReplicaSet. Option D also works but requires knowing the exact old image name.

Q3: A Deployment uses Recreate strategy. What is the expected behavior during an update?

A) Pods are replaced one at a time with no downtime
B) All existing pods are terminated before new pods are created, causing downtime ✓
C) Half the pods are updated at once while the other half serve traffic
D) New pods are created first, then old pods are terminated