Field notes from production MySQL work.
September 16, 2026 · QueryKings
One idle transaction left open can slow every read on the server, grow the undo tablespace, and make a routine query plan degrade overnight. Here is how MVCC purge lag builds, how to measure it, and how to keep it from happening again.
Read moreSeptember 15, 2026 · QueryKings
A backup you have never restored is a hypothesis. How to pick between logical dumps, physical snapshots and clones, keep a binlog chain that actually supports point-in-time recovery, and run a timed drill that produces a real RTO number instead of a policy document.
Read moreSeptember 14, 2026
A join predicate with two different collations silently stops using the index. Here is how utf8mb3, utf8mb4_general_ci and utf8mb4_0900_ai_ci end up mixed in one schema, how to find every mismatch before it costs you a table scan, and how to migrate large tables without a maintenance window.
Read moreSeptember 14, 2026
A platform move is a replication problem, not a dump-and-load problem. How to seed the target, run it as a replica of production, decide between RDS MySQL and Aurora, use Blue/Green where it fits, and cut over with a write pause measured in seconds and a rollback you have tested.
Read moreSeptember 14, 2026
ERROR 1040 is rarely a max_connections problem. How to find the limit that is really binding — memory per connection, thread scheduling, or a slow query holding the pool — and where a pooler like ProxySQL or RDS Proxy helps and where it hides the fault.
Read more