Field notes from production MySQL work.
September 21, 2026 · QueryKings
The index exists, EXPLAIN ignores it, and the query takes four seconds. Here is how InnoDB's cardinality estimates go stale, what histograms fix and what they do not, and the order to try ANALYZE TABLE, optimizer hints, and FORCE INDEX before you rewrite the query.
Read moreSeptember 21, 2026
Random UUID primary keys fight the clustered index: inserts land everywhere, pages split, secondary indexes carry a 36-byte key. Here is how to measure the damage, why UUIDv7 and UUID_TO_BIN(uuid, 1) fix most of it, and what migrating an existing table actually costs.
Read moreSeptember 21, 2026 · QueryKings
Most MySQL clusters fail over badly the first time it matters for real. Here is what asynchronous, semi-synchronous, and group replication each promise about data loss, how GTIDs and errant transactions decide whether a demoted primary can rejoin, and why the proxy layer — not the database — usually determines your actual outage length.
Read moreSeptember 18, 2026
Before you tune a query you have to pick the right one. How to rank a live workload with performance_schema statement digests, when the slow query log is still the better tool, and why total time beats worst-case latency as a target.
Read moreSeptember 17, 2026 · QueryKings
A 900 GB InnoDB table is not automatically a problem, and partitioning is not automatically the fix. How to tell which of the four real constraints you have hit — buffer pool, DDL time, delete cost, or a single-writer ceiling — and what each one actually calls for.
Read more