Episode 3. WAL and replication. Day 3 of your 12 day sprint.

Scenario: "Your replica is ten seconds behind at peak. Walk me through it." Start with the LSN chain, then measure.

Say: "Every change gets an LSN. The primary streams it, the replica receives, flushes, then applies. Lag lives in exactly one of those gaps."

Say the sync trade: "Synchronous replication buys zero data loss and spends commit latency. Asynchronous buys speed and risks lag."

Lab: read pg_stat_replication on a live or paper replica, name the sent, flush, and replay LSNs, and write a switchover versus failover paragraph.

Say the fencing line: "Planned switchover keeps one writer. Unplanned failover must fence the old primary first, or you get two writers."

Bridge from Oracle: Say: "This is Data Guard redo transport with new names. Sync versus async, lag triage, observer fencing, same discipline."

Drill: write the three lag gaps on paper: sent, flush, replay.
Say: receive lag is network, flush lag is disk, replay lag is apply load.
Say the RPO line: sync commits on two nodes, async promises nothing yet.
Done-check: state your cluster's RPO for sync and for async without notes.
Depth tag tonight: Debug on lag triage, Recognize on WAL-insert internals.

Teaser: tomorrow, Episode 4, indexing. The B-tree descent and the write bill every index charges.