Уровень 0 · материалов: 3
В кластер включаются документы, посвященные выявлению и устранению факторов, вызывающих системное падение производительности баз данных.
Общие признаки: снижение производительности БД, поиск первопричин (root cause), влияние индексов и файловых систем, мониторинг и диагностика
Группа выше: Диагностика деградации производительности
Смысл: The main idea is that systemic performance degradation in a database can threaten the survival of a software project, but a methodical approach—upgrading tools, implementing custom monitoring, and analyzing execution plans—can identify and fix root causes like inefficient indexing.
The author saved an online game from closure by upgrading MySQL, building a custom lock-logging tool, and fixing inefficient composite key indexes that caused full table scans.
Смысл: The main idea is that severe database performance degradation can be caused by low-level file system bugs (specifically ZFS lock contention on NUMA architectures) rather than the database engine itself, highlighting the necessity of full-stack diagnostic tools to find the root cause.
PostgreSQL performance plummeted on high-core NUMA servers due to a ZFS lock contention bug in the aggsum_add function, which was resolved by updating ZFS to version 2.1+.
Смысл: The main idea is that updating database indexes in high-load environments requires a precise sequence of operations because index utilization depends on the specific structure of the application's SQL queries. A failure to align code deployment with index changes can lead to catastrophic performance degradation.
A developer accidentally crashed a high-load production environment by deleting database indexes before deploying the code required to utilize their partial replacements.