Notesdistributed🌱 seedling

Causal Consistency

Also: Causally Consistent · Causal+ Consistency · Real-Time Causal Consistency

If one operation could have influenced another, every replica shows them in that order; concurrent operations may appear in different orders. It is about the strongest model that stays available during a partition.

Why it exists

The anomalies users notice are causal: a reply before its post, a photo visible to someone just removed from the audience. Fixing them doesn't need a global order, only cause before effect, and that can be done without cross-region round trips.

How it works

  • Causality is Happens-Before: same client in sequence, or a read of a write followed by a later operation.
  • Each write carries its dependencies (a Version Vector, or the versions the client has observed).
  • A replica delays making a write visible until everything it depends on is visible.
  • Implies the session guarantees: Read-Your-Writes Consistency, Monotonic Reads, monotonic writes, writes-follow-reads.
  • Real-time causal, a slightly stronger variant, is proved the strongest model in an always-available, convergent system.

Trade-offs

Gains Costs
no causal anomalies dependency metadata on every write
local reads and writes; sticky available concurrent writes still conflict and need merging
no global coordination only sees causality that flows through the store

In practice

  • MongoDB: causally consistent sessions pass cluster and operation times between operations so a session reads its own writes in causal order.
  • Typical use: social feeds, comments, collaborative apps across regions.

Pitfalls

  • Hidden channels: a phone call or a message queue carries causality the database never saw.
  • Clients that hop replicas lose the "sticky" availability; the session must carry its dependency state.

Recall

These cards are also in your review deck, scheduled with spaced repetition.