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

Consistency Model Explorer

What is a consistency model?

A consistency model is a precise promise about which values a client may observe when writes, reads, replication, and failures overlap.

Step 1

Model

Choose the promised model, write acknowledgement, read route, session protection, operation timing, and a failure or recovery history.

Step 2

Observe

Inspect replica values, trace every operation, and see whether the observed history satisfies or disproves the promise.

Step 3

Challenge

Force a stale read, then add read-your-writes protection. Compare that result with a network partition, concurrent writes, and three recovery policies.

Replicated history workbench

Make the guarantee prove itself

Promise a consistency model, configure the read and write protocol, then inject a history that can confirm or disprove the claim.

3. Failure injection

Challenge the healthy path

User-visible outcome

The UI appears to undo the save

Client A saved V1, refreshed, and saw V0. A user may retry and create duplicate work.

Observed read

V0

Observed system

Replica state at read time

Primary region

Replica A

V1

Write coordinator

Stored V1 at 100 ms

Primary region

Replica B

V0

Quorum peer

V1 visible at 980 ms

Remote region

Replica C

V0

Client A local read

Still on V0; V1 arrives at 1,700 ms

Write reply

130 ms

Read begins

500 ms

Replica C catches up

1,700 ms

Operation timeline

Trace the observed history

  1. 0 ms

    Client A

    Previous read

    Client A observed V0.

  2. 100 ms

    Client A

    Write V1

    Client A sends V1 to Replica A.

  3. 120 ms

    Replica A

    Store V1

    Replica A starts replication.

  4. 500 ms

    Client A

    Read returns V0

    The nearest replica responds before V1 arrives.

  5. 980 ms

    Replica B

    Apply V1

    Replica B advances from V0 to V1.

  6. 1,700 ms

    Replica C

    Apply V1

    Replica C catches up to V1.

Contract verdicts

Test each guarantee against this history

Promised contract

Read-your-writes

History disproves it

Client A completed V1 and then observed V0 in the same session.

Forced stale read

What changed and why

Client observation

The nearest replica responds before V1 arrives.

Design consequence

A consistency label is only valid if every allowed history satisfies its promise. Change the route, acknowledgement, timing, or session token and compare the same fault again.

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