Redis

Replication Backlog Too Small for Network Instability

warning
ReplicationUpdated Mar 2, 2026

When redis.replication.repl_backlog_size is too small relative to write rate and network latency, replicas require full resync after brief disconnections, causing replication storms and increased master load.

Technologies:
How to detect:

Monitor redis.replication.master_link_down_since_seconds events combined with redis.clients.connected_slaves dropping and recovering. If replicas frequently require full resync rather than partial, backlog is too small. Calculate required backlog: write_rate * expected_disconnection_time * safety_factor.

Recommended action:

Calculate appropriate repl-backlog-size: (write_bytes_per_sec * max_expected_disconnection_seconds * 2). Increase repl-backlog-size in redis.conf if replicas frequently require full resync. Monitor redis.replication.offset divergence to detect when backlog is insufficient. For high-write workloads, consider 64MB-256MB backlog size instead of default 1MB.