E-Ink News Daily

Back to list

PREEMPT_NONE Is Dead; Your Postgres Probably Doesn’t Care

An AWS benchmark revealed a significant PostgreSQL throughput regression on Linux 7.0, primarily caused by the removal of the PREEMPT_NONE kernel scheduler mode. The author argues this is largely a configuration issue related to large shared memory setups rather than a fundamental flaw in the Linux kernel, advising most users that they do not need to delay upgrades.

Background

Linux 7.0 removed the PREEMPT_NONE scheduling option, shifting server kernels towards PREEMPT_LAZY or PREEMPT_FULL by default. This change has sparked debate regarding its impact on database workloads like PostgreSQL that rely on specific preemption behaviors for optimal throughput.

Source
Lobsters
Published
Jul 6, 2026 at 08:31 PM
Score
7.0 / 10