Comparison
Recovery point objectivevsReplication lag
Recovery point objective
backups run nightly, so recovering means losing up to a day of writes, and nobody outside the platform team knew that.
The maximum acceptable data loss, expressed as time: how far back you may be forced to rewind. It is set by replication and backup frequency, so nightly snapshots mean a day and synchronous replication means approximately none. It is the number people conflate with recovery time, and the two are independent — you can be back in ten minutes having lost a day, or take a day to restore losing nothing.
Full entry →Replication lag
the replica is forty seconds behind the primary and the report you generated from it is missing this morning's orders.
How far behind a follower is from the leader, measured in time or log position. It grows under write bursts, long transactions, and single-threaded replication apply. It is the number to alert on before you send any read traffic to replicas, and the number that decides whether a failover loses data.
Full entry →