File systems and I/O drill

You are building a small embedded job queue that persists jobs to a local file so they survive crashes and power loss. Specify exactly how you append jobs, when you acknowledge them to producers, and how you recover, and estimate throughput if each fsync takes 0.5 ms.

A logging agent on each host writes application logs to /var/log and rotates them nightly. Last week several hosts filled their disks: df showed 100 percent, but du over /var/log summed to only 20 GB of a 100 GB disk. A different host failed with 'No space left on device' while df showed 35 percent free. Diagnose both incidents and prevent them.

A nightly backup job on the same NVMe-backed database host makes the database's p99 write latency jump from 2 ms to 40 ms for two hours. The backup reads the data directory with a single-threaded file copy. Explain what competes, how you would measure it, and three fixes.

A static-file and video-segment server on Linux serves 20 Gbit/s and is CPU-bound, with profiles showing time in copying and in read and write system calls. Explain where the copies and system calls come from and redesign the serving path.