Your p99 got worse after a kernel upgrade and nothing else changed? That is an EEVDF latency regression. Meta cut p99 28% with a two-pool sched_ext policy.
A zero downtime database migration is five feature-flagged phases, not a cutover. P95 said 4x slower, P90 said 17x, and the comparison code took a region down.
Run Kubernetes on bare metal and four integrations become yours: node identity, LoadBalancer IPs, provisioning, storage. Oxide shipped three; one is blocked.
Docker Swarm vs Kubernetes: a live SaaS runs for $166/year with zero crashes in 10 years, while the average Kubernetes cluster wastes 87% of its CPU.