An Amazon/AWS engineer discovered that Linux 7.0 development kernel cuts PostgreSQL performance in half compared to previous versions. Salvatore Dipietro found throughput dropping to just 0.51x on Graviton4 servers due to increased time spent in user-space spinlocks. The regression was traced to Linux 7.0's restriction of preemption modes, limiting available scheduling options for modern CPUs.

This change, previously covered by Phoronix, was part of broader scheduler updates focusing on full and lazy preemption models. A patch was proposed to restore PREEMPT_NONE as the default preemption model, but the original author suggests PostgreSQL should instead adopt Restartable Sequences (RSEQ) time slice extensions. The debate centers on whether the database should adapt or the kernel should revert its optimization.

If PostgreSQL must implement RSEQ support, Linux 7.0 stable could ship with significant performance regressions for database workloads until the popular server is updated. With Linux 7.0 stable arriving in about two weeks and powering Ubuntu 26.04 LTS in April, database administrators face a potential performance cliff unless they can apply workarounds or delay kernel upgrades.