Skip to content
Ameen Abozaid
SRB-005

High-availability design

  • 7 Lessons
  • 3.5 h practice (estimate)
On this page 9 topics

Overview

Topics

  • Redundancy
  • load shedding
  • failover
  • fault containment

Learn first

3.5 h of practice (estimate)

Learning outcomes

  1. Design redundancy at every layer and remove hidden single points of failure.
  2. Apply load shedding, admission control, and prioritization under overload.
  3. Contain faults with cells, shuffle sharding, and bulkheads at the infrastructure level.
  4. Compute composite availability and explain why dependencies cap it.

Lessons

  1. 01Availability math
  2. 02Redundancy and hidden SPOFs
  3. 03Load shedding and admission control
  4. 04Cell-based architecture
  5. 05Shuffle sharding
  6. 06Demo: shedding low-priority traffic during a flash sale
  7. 07Demo: computing checkout's real availability from its dependencies

Lab and exercises

Lab

SRB-005-LAB1 · Availability budget

Outcomes: 1, 2, 3, 4 · Time: ~3.5 h

  1. Map checkout dependencies including FulfillCo ones and mark hard or soft.
  2. Compute composite availability with measured figures.
  3. Redesign (fallbacks, cells, bulkheads) to reach 99.95%.
  4. Run a load-shedding experiment and record results.

Acceptance: math shown; load-shedding experiment run with before/after numbers.

Exercises

IDOutcomeTask
SRB-005-EX014Compute checkout availability: 99.95%, 99.99%, 99.9%, 99.9% in series.
SRB-005-EX024Recompute when the 99.9% stock check gets a cache fallback that serves 90% of failures.
SRB-005-EX031Audit a design for hidden SPOFs and list five.
SRB-005-EX042Write the shedding priority policy for the flash sale.
SRB-005-EX053Propose a cell design for FulfillCo shipping and the routing key.
SRB-005-EX063Explain shuffle sharding to a teammate with numbers.
Show solutions
  • EX01 Product about 99.74%; below 99.95% so redesign needed.
  • EX02 Effective stock failure drops to 0.01%, so about 99.99%; total about 99.83%; show the arithmetic.
  • EX03 Certificate expiry, shared config, single DNS provider, single CI pipeline, one on-call expert.
  • EX04 Protect payment confirmation and checkout of logged-in users; throttle browse; drop crawlers and recommendations first.
  • EX05 Cells by warehouse region with routing by warehouse id; rollouts cell by cell.
  • EX06 8 workers, 2 per customer: 28 pairs; chance two share both is 1/28.

Workplace evidence: an availability improvement you designed with before/after numbers.

Quiz

Outcomes: O1 redundancy and SPOFs · O2 shedding · O3 fault containment · O4 availability math.

A self-check for you: nothing is graded, saved or sent anywhere. Choose an answer, then check it. You can try again as often as you like.

  1. 1Two 99.9% serial services give about:O4
  2. 2Parallel redundancy math assumes:O4
  3. 3Dependencies cap availability because:O4
  4. 4A hidden SPOF could be:O1
  5. 5Redundancy that is never tested:O1
  6. 6Load shedding should drop first:O2
  7. 7Admission control means:O2
  8. 8A cell is:O3
  9. 9Cells mainly limit:O3
  10. 10Shuffle sharding reduces:O3

Practical part

Availability calculation problems (series, parallel, fallback) plus the lab's experiment evidence. Pass: 70% on the quiz and correct math in the lab.

Common mistakes

Multiplying nines without dependency analysis; shedding that drops paying customers first.

Assessment

Availability calculation problems.

Workplace evidence

An availability improvement you designed with before/after numbers.

Career relevance

Enterprise resilience (SRA-006).

Type to search levels, phases, modules and lessons.