Redis Replication

A Redis replica maintains a copy of a primary's dataset and serves reads. Replication is asynchronous: the primary acknowledges a write to the client before the replica has it.

# On the replica.
replicaof 10.20.1.10 6379
masterauth <password>
replica-read-only yes

Full and partial resynchronisation

On first connection the replica performs a full sync: the primary forks, produces an RDB snapshot, sends it, and streams the writes that happened meanwhile.

On a brief disconnect the replica attempts a partial resync using the replication backlog — a ring buffer of recent writes on the primary.

repl-backlog-size 64mb
repl-backlog-ttl 3600

Diskless replication

repl-diskless-sync yes
repl-diskless-sync-delay 5
repl-diskless-load swapdb

Diskless sync streams the RDB directly to the replica socket instead of writing it to disk first. This helps when disks are slow or space is tight; it means the primary cannot serve the same snapshot to several replicas that connect at different moments, which is what repl-diskless-sync-delay batches for.

Consistency

WAIT numreplicas timeout blocks until the given number of replicas have acknowledged all writes issued by this connection:

SET order:4711:status paid
WAIT 1 100

It narrows the window and does not close it: WAIT confirms receipt, not durability on disk, and it is not a consensus protocol. Treat it as a risk reduction, not a guarantee.

Preventing stale reads

Replicas serve stale data by definition. Two controls:

# On the replica: stop serving reads if it has been disconnected too long.
replica-serve-stale-data no
 
# On the primary: refuse writes unless N replicas are connected and lagging < M seconds.
min-replicas-to-write 1
min-replicas-max-lag 10

min-replicas-to-write makes the primary fail writes rather than accept data it cannot replicate. That converts a silent durability loss into a visible error — usually the right trade for data you care about, and the wrong one for a pure cache.

Monitoring

INFO replication

Key fields: role, connected_slaves, master_link_status (must be up on a replica), master_repl_offset versus each replica's offset for lag in bytes, and sync_full / sync_partial_err in INFO stats to spot repeated full resynchronisations.