Skip to main contentSkip to user menuSkip to navigation
Interactive design lab

Reliability Engineering Workbench

What makes a service reliable?

A reliable service completes its critical user journey within an agreed SLO despite component failures, maintenance, and recovery. The journey is only as strong as its serial dependencies, shared failure domains, and tested restore path.

Step 1

Model

Shape the critical path, replica success rule, failure-domain placement, correlation assumption, MTBF, repair MTTR, detection and failover time, restore time, replication lag, RTO, RPO, SLO, and planned user impact.

Step 2

Observe

Compare independent and correlation-aware availability, translate the result into downtime and error-budget use, and see which architecture or recovery objective breaks first.

Step 3

Challenge

Start with the healthy plan, then inject common-mode failure, zone loss, slow recovery, maintenance overlap, and a cascading dependency. Use measured incident and restore-test data before treating the result as an SLO commitment.

Reliability workbench

Design the journey, then break its assumptions

Combine series dependencies, parallel replicas, failure domains, repair behavior, recovery objectives, and an SLO without pretending every failure is independent.

Scenario status

Objectives fit

Challenge the plan

Failure and operating scenarios

Scenarios change the evaluated month, not the assumptions you entered.

Evaluated dependency path

3 serial stages, 2 of 3 replicas required

Each stage uses parallel replicas. The complete journey succeeds only when every stage reaches its success rule.

Placement

3 domains · up to 1 replica/domain

User request

Journey begins

Critical stage 1

2-of-3

123

99.992% modeled stage

Critical stage 2

2-of-3

123

99.992% modeled stage

Critical stage 3

2-of-3

123

99.992% modeled stage

User outcome

99.975%

Available replicaScenario-unavailable replicaScroll is contained inside the path on narrow screens.
Availability model

Independent redundancy versus correlated reality

Parallel replicas improve a stage; serial stages multiply their risk. Correlation assigns a share of replica downtime to one shared cause, where adding copies cannot help.

One replica

99.9%

MTBF 2,000 hr / MTTR 2 hr

Independent parallel stage

>99.999%

Probability at least 2 of 3 are healthy

Correlation-aware stage

99.992%

8% shared-failure assumption

Complete journey

99.975%

3 stage probabilities multiplied in series

Model equations

A(replica) = MTBF / (MTBF + MTTR)

A(parallel) = P(at least k of n replicas healthy)

U(stage) ≈ c × U(replica) + (1 − c) × U(parallel)

A(journey) = A(stage) ^ serial stages

Assumption boundary

The correlation blend is a planning approximation, not a substitute for failure telemetry, load tests, or a dependency-specific Markov model. Rounded outputs communicate model scale rather than contractual precision.

Objectives and budgets

Turn availability into time the team can operate

99.95% monthly SLO

21.6 min

Total monthly error budget

Scenario month impact

12.7 min

59% of the error budget

Annual model downtime

2.6 hr

Steady-state estimate plus planned impact

Budget remaining

8.9 min

Reserve after this scenario

Error-budget consumption

12.7 min / 21.6 min

Includes 2 min planned impact and 0 sec from the selected challenge.

RTO: restore service

24 min / 30 min

2 min detect + 4 min fail over + 18 min restore

RPO: committed-data exposure

2 min / 5 min

Modeled from replication lag under the selected scenario.

Healthy plan

The configured path keeps its modeled error-budget reserve.

Use measured failure-domain correlation and incident data before treating the estimate as a commitment.

RTO

Met

RPO

Met

Quorum

Survives

Architecture consequences

What the current assumptions require

  • 13 of 3 replicas remain; the 2-of-3 success rule can still complete.
  • 2Shared failure accounts for 8% of replica unavailability; verify this assumption with incident and placement data.
  • 3The 24 min recovery path fits the 30 min RTO.
  • 4The 2 min replication exposure fits the 5 min RPO.

Planning rule: derive MTBF, MTTR, correlation, and recovery times from production evidence. Confirm the model with fault injection and restore tests before using it for an SLO commitment.

No quiz questions available
Could not load questions file
Spotted an issue or have a better explanation? This page is open source.Edit on GitHub·Suggest an improvement