A payment balance needs the latest truth; a feed like-count can lag a moment. Toggle between strong and eventual consistency and watch freshness trade against speed and availability.
Same write, two guarantees. Toggle the mode, optionally knock a replica offline, then write a new value and watch when each node reflects it — and at what cost.
🗄️
Primary
v1
fresh
📄
Replica 2
v1
fresh
📄
Replica 3
v1
fresh
Strong
Eventual
Read freshness
Always latest
May be stale briefly
Write latency
Higher (waits for replicas)
Low (acks immediately)
Availability on node loss
Write refused
Stays available
Use for
Payments, balances, inventory
Feeds, likes, view counts, search
What just happened
▹Strong consistency means every read sees the latest write. To guarantee it, a write must reach all replicas before it's acknowledged — so writes are slower, and if a replica is unreachable the write is refused.
▹Eventual consistency acknowledges the write immediately and lets replicas catch up in the background. Writes are fast and stay available even when a node is down — but reads can be briefly stale.
▹It's a per-data-type decision. A payment balance demands strong; a feed like-count, view count, or recommendation is perfectly happy being a second behind.