Tag
The prime directive (code and schema never jump together), batched backfills as jobs, each engine's locking personality — CONCURRENTLY, NOT VALID, gh-ost — and the destructive-change quarantine.
The lost update, the phantom check, the inconsistent report, the cross-row invariant — each real bug matched to its cheapest cure, from atomic updates to Serializable retry loops, plus deadlock etiquette.
Why the bug is syntactically invisible, making the framework tattle with preventLazyLoading, the fix taxonomy beyond with(), and the two siblings — over-eager loading and unbounded gets — hiding behind it.
Where Postgres genuinely pulls ahead, where MySQL holds real ground, the differences everyone overweights, and a decision procedure whose top factor is organizational, not technical.
The B-tree mental model everything follows from, composite column order as the whole game, covering and partial indexes, reading EXPLAIN without a priesthood, and the quarterly maintenance contract.
The honest tour by operational weight: Postgres tsvector with trigrams, the search-box specialists with superb typo tolerance, and the real signals that justify the heavyweight cluster.
The three machines under the API: inverted indexes (why search is fast and updates are weird), analyzers with a Turkish-aware example, permanent mappings, and BM25 relevance in one honest paragraph.
Organized by the incident it prevents: maxmemory and eviction policies, AOF vs RDB persistence blends, Sentinel vs Cluster reality, the fork-memory trap and the five-number dashboard.
The 95% of Redis teams never open: atomic counters, rate limiters, sorted-set schedulers, pub/sub vs streams (one of them forgets!), queue pragmatics and the eviction-policy line that forces two instances.
Cache-aside with the two mandatory upgrades — TTL jitter and stampede protection — plus version-stamped invalidation, the after-commit trap, and the three metrics that keep a cache honest.
