4 · High Availability & Fault Tolerance — Revise¶

Resilient · 26% · Compressed recall. Read the Learn guide first.

Key facts¶

Fact	Value
Default resilient web tier	ALB → ASG across ≥ 2 AZs
Multi-AZ	Protects vs AZ failure (HA default)
Multi-Region	Protects vs Region failure (DR — Topic 6)
ASG self-healing	Replaces instances failing health checks
ASG default health check	EC2 only → set to ELB behind a load balancer
Simplest scaling policy	Target tracking (e.g. CPU 50%)
Cross-zone LB	ALB: on + free; NLB: off by default (may add inter-AZ cost)

ELB types¶

LB	Layer	Pick when
ALB	7	HTTP/HTTPS, path/host routing, Lambda/container targets, WAF
NLB	4	Static/Elastic IP, millions rps, ultra-low latency, TCP/UDP
GLB	3	Inline 3rd-party appliances (firewall/IDS/IPS)
CLB	legacy	Avoid

Scaling policies¶

Policy	Use
Target tracking	Keep a metric at a value (default choice)
Step / Simple	Adjust by alarm magnitude
Scheduled	Known time-based patterns
Predictive	ML forecasts load, scales ahead

Decision cues¶

Survive AZ failure + auto-scale?     → ASG across ≥2 AZs behind ALB
HTTP path/host routing / Lambda?     → ALB
Static IP / raw TCP-UDP / extreme?   → NLB
Inline firewall appliance?           → GLB
App crashes not being replaced?      → ASG health check type = ELB
Predictable daily spike?             → Scheduled scaling (+ target tracking)
Users randomly logged out behind LB? → Sticky sessions (better: externalize session)

Flashcards¶

What's the standard resilient web-tier architecture?

An Application Load Balancer in front of an Auto Scaling group spanning at least two Availability Zones.

ASG behind an ALB doesn't replace app-crashed instances. Why?

Default ASG health check is EC2 status only (OS running). Set health check type to ELB so it acts on the ALB's health checks.

Need a static IP and Layer-4 performance from a load balancer?

Network Load Balancer (static/Elastic IP per AZ, millions rps, ultra-low latency).

Cross-zone load balancing defaults?

ALB: enabled and free. NLB: disabled by default; enabling can add inter-AZ data transfer cost.

Which scaling policy for a predictable weekday 8am spike?

Scheduled scaling to pre-warm capacity, plus target tracking for the rest.

Does an Auto Scaling group provide HA on its own?

Only if it spans multiple AZs. In one AZ it's still a single point of failure.

Which LB routes by URL path or hostname, and can target Lambda?

Application Load Balancer (Layer 7).

Common traps¶

ALB + ASG across ≥ 2 AZs is the resilient default; one AZ = SPOF.
ASG health check defaults to EC2 → use ELB behind a load balancer.
NLB = static IP / L4 / extreme perf; ALB = HTTP routing / Lambda targets.
Cross-zone: ALB on+free, NLB off by default.
Target tracking first; scheduled/predictive only for time-based/forecastable load.
LB and ASG are Regional (multi-AZ), not multi-Region.