Reference · Cheat sheet

Replication

Async by default — fast, but a failover can lose writes. Print it, pin it.

From lesson 0006See also glossary

The model

Leader-follower: one primary, many replicas, exact copies. Replicas can cascade.
Replicas read-only by default (replica-read-only yes) — writes go to the primary.
Asynchronous by default: primary acks the client before replicas confirm.

Sync (PSYNC = replication ID + offset)

Resync	When	Cost
Partial	Reconnect, gap still in the backlog buffer	Ship only the missed stream slice
Full	First sync, or backlog too old	`BGSAVE` → RDB streamed → replay buffered writes

repl-diskless-sync: stream the RDB over the socket, skipping disk.

The async trade-off

Not a CP system

Ack comes before replicas confirm, so "acknowledged writes can still be lost during a failover." There is always a window for data loss.

Narrow the window with…	Effect
`WAIT n timeout`	Block until ≥ n replicas ack (or timeout). Best-effort, not CP.
`min-replicas-to-write N` + `min-replicas-max-lag M`	Refuse writes unless ≥ N replicas with lag < M s. Bounds loss to ~M s.

Replica expiry

Replicas don't expire keys — they wait for the primary's synthesized DEL, hiding logically-dead keys on reads. A promoted replica then expires independently.

Handy commands

REPLICAOF host port — make this node a replica. REPLICAOF NO ONE — promote to primary.
INFO replication — role, connected_slaves, master_link_status, offsets.
WAIT n timeout — how many replicas have my writes.

← lesson 0006 · Sources: Redis Docs — Replication · WAIT