Notes for the problems that show up after launch: bad plans, awkward migrations, index debt, vacuum pressure, replica lag, and the small decisions that make PostgreSQL easier to operate.
Two weeks after a routine OS upgrade, inserts started failing with duplicate-key errors on values that did not exist. The culprit was a glibc collation change that had quietly scrambled every btree on text.
Your ORM will connect to PostgreSQL on day one and still surprise you for weeks. These are the behavior changes I watch for: case sensitivity, booleans, upserts, RETURNING, and transaction aborts.
MariaDB has real sequence objects; MySQL never got them. They shine for shared id spaces and ids known before the insert, and they are a waste of a round trip for a single hot table.
Stored procedures are the part of a MySQL migration nothing automates. Field notes on porting routines and triggers to PL/pgSQL, and on deciding which logic belongs in the app instead.
The rule 'never use hash indexes' came from a real scar: before PG10 they skipped WAL, so replicas and crash recovery silently lost them. That was fixed eight major versions ago, and the reputation never recovered.
GTIDs make MySQL failover repointing almost trivial; replication slots make PostgreSQL WAL retention automatic but risky. A practitioner's comparison of the two position-tracking models.
Async replication, semi-sync, and Galera are not versions of the same feature — they are different trades in latency, consistency, and operations. A field guide to choosing between them on MariaDB.
Optimistic parallel replication bets that your transactions will not conflict and pays for the bet in rollbacks. Here is how the five slave_parallel_mode modes differ, when the bet wins, and how to measure real lag.
An errant transaction is a write that exists on a replica but not on the source, and it sits harmlessly in gtid_executed until the exact moment you fail over. Here is how they happen, how to find them, and how to fix them.
PgBouncer's transaction pooling mode is the right call for most fleets, but it silently breaks prepared statements, SET, advisory locks, and LISTEN/NOTIFY. Here is how to run it without the 3am surprises.
Btree indexes bloat quietly as updates and deletes churn. Here is how to measure bloat with pgstattuple, why it wrecks cache hit rates, and how REINDEX CONCURRENTLY fixes it without taking the table down.
Parallel query makes big scans faster until forty concurrent queries all want workers at once. The knobs, the cost model, Gather vs Gather Merge, and why Workers Launched says zero.