Replication

FrogDB supports primary-replica replication using an orchestrated model (not gossip). Standard Redis replication commands (REPLICAOF, ROLE, INFO replication) work as expected — see Redis replication docs.

How FrogDB Differs from Redis Replication

• Orchestrated failover — an external orchestrator manages topology, not Sentinel or gossip
• RocksDB checkpoints for full resync instead of RDB transfer
• SHA256 integrity verification during checkpoint streaming
• Replicas do not evict independently — all eviction decisions come from the primary via replicated DEL commands
• Memory-aware sync — replicas monitor memory during full sync and can pause/abort if limits are exceeded

Configuration

[replication]
role = "replica"                          # standalone, primary, replica
primary-host = "primary.example.com"
primary-port = 6379
min-replicas-to-write = 1                 # Write quorum (0 = disabled)
min-replicas-timeout-ms = 5000

WAL Retention

WAL retention determines how long replicas can be disconnected before requiring full resync:

[rocksdb]
min-wal-retention-secs = 3600   # Keep WAL for 1 hour
min-wal-files-to-keep = 10

Increase if replicas may be offline for extended periods.

Replication ID

Each dataset history has a unique replication ID. On failover:

1. Promoted replica generates a new replication ID
2. Promoted replica remembers the old primary’s ID (secondary ID)
3. Other replicas can continue with partial sync using either ID

Failover

Managed by the external orchestrator:

1. Orchestrator detects primary failure via health checks
2. Orchestrator selects a replica for promotion
3. Promoted replica generates new replication ID, begins accepting writes
4. Other replicas reconnect to new primary (PSYNC using secondary ID)
5. Clients receive -MOVED redirections

Memory Considerations

• Provision replicas with 10-20% more memory than the primary (replication buffers)
• When a promoted replica becomes primary, maxmemory becomes active and eviction begins immediately if over limit
• Full sync memory is configurable:

[replication]
fullsync-max-memory-mb = 512    # Max memory for full sync buffering

Monitoring

INFO replication

Compare master_repl_offset (primary) with slave_repl_offset (replica) to measure lag.

Symptom	Likely Cause	Fix
Frequent full resyncs	WAL retention too short	Increase min-wal-retention-secs
Replica OOM during sync	Primary dataset exceeds replica memory	Increase replica memory