Why ultra-fast COMMIT is dangerous for DBMS and how to verify data persistence

This article from Postgres Professional explores the critical issue of data persistence in DBMS. The author notes that enabling fsync in the configuration does not always guarantee physical data writes to disk. System calls can often be ignored by drivers, virtualization layers, or the OS itself, while the database incorrectly confirms a successful COMMIT to the client. This creates a risk of data loss during failures. The material details verification methods, ranging from system call analysis to stress testing with hard reboots of the test environment. The goal is to help DBAs ensure that WAL files actually reach non-volatile storage rather than remaining in the OS cache. This guide serves as an essential tool for ensuring data integrity in high-load systems.
This is a summary. Read the full article at the original source:
HabrRelated stories
Databricks has officially acquired Row Zero, a cloud-based spreadsheet startup designed to handle massive datasets with high performance. This acquisi…
Reconciling Wildberries financial reports: Why totals match but item-level data fails
The author analyzes the issue of financial data discrepancies when working with the Wildberries marketplace. Although total account balances match the…
The author shares their experience of discovering critical errors in a published open dataset regarding parking occupancy. After an anomalous occupanc…

