after each chapter

Compare

You started with a normal single caster, then bolted on AZ HA, then region HA. The software never became a cluster — only the recovery plane changed.

Teardown multi-region Back: Multi-region

Progression

Chapter Example Failure you absorb Typical gap
Baseline relay_iport None — host down = stream down until you repair
Dual-AZ relay_iport_ha AZ / single host ~20–40s (~34s measured)
Multi-region relay_iport_ha_mr Region 1–3+ min (DNS)
Two HA patterns for NTRIP Path A: clients via Route 53 alias to one shared NLB with blue and green EC2 casters. Path B: clients via Route 53 failover to PRIMARY Auckland or SECONDARY Sydney NLB and caster. Path A / Dual-AZ L4 shared NLB / ~20-40s NTRIP :2101 Route 53 alias VIP NLB TCP health Blue / AZ-a Green / AZ-b relay pull Path B / Multi-region DNS AKL + SYD / DNS TTL 1-3+ min same VIP Route 53 failover PRIMARY AKL SECONDARY SYD both pull

Dual-AZ vs multi-region

  Dual-AZ L4 Multi-region DNS
Mechanism Shared NLB target health Route53 failover aliases
VIP on failover Unchanged Same FQDN, new answer
Dominant delay NLB unhealthy + reconnect Resolver / TTL
Keep NLB? Yes (the failover plane) Yes (stable alias + health)

Dual-AZ wins on recovery time. Multi-region wins on regional blast radius. Neither migrates an established TCP stream — NtripCaster never did that.

Decision

Choose When
Baseline only Lab / single site; accept host repair time
Dual-AZ AZ / host resilience; shortest reconnect gap
Multi-region Region diversity; accept minute-scale DNS
Neither HA alone Continuous fixed / safety-critical without a hotter path

Normal setup → prove the stream → dual-AZ if you need AZ HA → multi-region if you need regional diversity.

Teardown index Links into each chapter’s Teardown Baseline Re-run normal setup