What are consistency models (strong vs eventual)?

Question

IT Interview Practice · Accepted Answer

**Consistency models** define guarantees about when and how data updates become visible across a distributed system — ranging from **strong consistency** (everyone sees the latest data immediately) to **eventual consistency** (updates propagate over time). The choice involves trade-offs with availability and performance.

## Strong vs eventual consistency

```text
STRONG CONSISTENCY → every read returns the MOST RECENT write (all nodes agree immediately):
  ✓ simple to reason about; always correct/current data
  ✗ requires coordination → higher latency, lower availability (especially during partitions)
  → for: data that must be correct/current (financial balances, inventory, bookings)
EVENTUAL CONSISTENCY → updates propagate over time; reads MAY return stale data briefly,
  but all nodes CONVERGE eventually:
  ✓ high availability, low latency, scalable
  ✗ reads can be stale temporarily (must tolerate this)
  → for: data where brief staleness is OK (social feeds, likes, view counts, caches)
```

## The trade-off

```text
Consistency vs availability/performance (relates to CAP):
  → STRONGER consistency → more coordination → higher latency, lower availability
  → WEAKER (eventual) consistency → faster, more available, scalable, but stale reads
→ choose based on whether the data NEEDS to be immediately consistent for correctness.
```

## Other models and nuance

```text
→ a SPECTRUM exists: read-your-writes, monotonic reads, causal consistency, etc.
→ Many systems offer TUNABLE consistency (choose per operation/query)
→ Use STRONG where correctness demands it; EVENTUAL where availability/scale matters and
  staleness is acceptable → often MIX within one system
```

## Why it matters

Understanding consistency models is valuable because **the consistency vs availability/performance trade-off is fundamental to distributed systems**, so it's important system-design knowledge.

Consistency models define when data updates become visible across a distributed system, and the key distinction — **strong consistency** (every read returns the latest write, all nodes agreeing immediately) versus **eventual consistency** (updates propagate over time, reads may be briefly stale but converge) — represents a fundamental trade-off.

Understanding this is important because **strong consistency** provides correctness and simplicity (always current data) but requires coordination that increases latency and reduces availability (suiting data that must be correct, like financial balances, inventory, and bookings), while **eventual consistency** offers high availability, low latency, and scalability at the cost of temporary staleness (suiting data where brief staleness is acceptable, like social feeds, likes, and view counts).

Understanding **the trade-off** (stronger consistency means more coordination, higher latency, and lower availability; weaker consistency means faster, more available, and scalable but with stale reads — relating to the CAP theorem) is the key insight that guides choosing the right model for each use case.

Understanding the **nuance** — that a spectrum of models exists (read-your-writes, causal consistency, etc.), that many systems offer **tunable consistency** (choosing per operation), and that systems often **mix models** (strong where correctness demands it, eventual where availability and scale matter) — reflects sophisticated understanding.

Choosing consistency appropriately is a key design decision affecting correctness, availability, and performance.

Since the consistency vs availability/performance trade-off is fundamental to distributed systems and choosing the right consistency model (strong vs eventual, often mixed) significantly affects correctness, availability, and performance, and since understanding the models and their trade-offs is important for designing distributed systems and choosing databases, understanding consistency models is valuable, practically-relevant system-design knowledge — a fundamental distributed-systems trade-off (consistency vs availability/performance), important for choosing appropriate data handling based on correctness needs, and central to designing distributed systems and selecting databases with the right consistency guarantees.

Strong vs eventual consistency

The trade-off

Other models and nuance

Why it matters

Related questions