Linux 7.0 Slashes PostgreSQL Throughput in Half
AWS says the current Linux 7.0 development kernel can cut PostgreSQL throughput to roughly 0.51x on Graviton4, with the slowdown tied to a preemption-mode change that increases time spent in a user-space spinlock. The proposed kernel-side fix looks unlikely, so PostgreSQL may need to adapt instead.
This smells like a kernel-vs-runtime boundary fight, not a simple bug fix. If the revert gets rejected, the pain shifts to PostgreSQL and any workload that leans on similar locking patterns.
- –The regression was bisected to Linux 7.0’s restriction of available preemption modes, specifically the move away from PREEMPT_NONE as a default.
- –Peter Zijlstra’s suggested path is to use rseq slice extension support, which pushes the workaround into user space rather than undoing the kernel change.
- –That makes this more than a one-off benchmark blip: database runtimes may need code changes to stay fast on newer kernels.
- –The impact appears workload-specific, but PostgreSQL is common enough that even a narrow regression becomes an ecosystem problem.
- –If Linux 7.0 stable ships before PostgreSQL adapts, cloud operators could see a real performance cliff on affected deployments.
DISCOVERED
52d ago
2026-04-05
PUBLISHED
52d ago
2026-04-05
RELEVANCE
AUTHOR
crcastle