SIGSEGV in worker process after prolonged HTTP/1.1 keepalive traffic

Goal: verify that nginx ≥ 1.19.8 resolves the upstream keepalive memory leak that manifested as SIGSEGV in worker processes under prolonged HTTP/1.1 keepalive traffic, and determine whether the worker_processes auto setting still produces uneven load distribution on heterogeneous CPU architectures when worker_process_auto_affinity is enabled.

Constraints: while the leak fix is well‑documented, the interaction between worker_processes auto and worker_process_auto_affinity on big.LITTLE or NUMA systems remains unspecified; it is unclear whether nginx should automatically adjust the worker count to match available affinity or if administrators must set worker processes and affinity explicitly for predictable performance.

Questions: Does enabling worker_process_auto_affinity guarantee balanced worker placement across core types when worker_processes is set to auto? Is explicit worker count and CPU affinity configuration required to avoid suboptimal throughput on big.LITTLE systems?