🖥️ Operating Systems

Why: thread pools, GC pauses, container limits, file I/O, and “why is this slow” all come back to the OS.

Core → Advanced

• Processes vs threads, context switching, and its cost
• CPU scheduling (CFS basics), user vs kernel mode, system calls
• Virtual memory, paging, TLB, page faults, the stack vs the heap
• File systems, page cache, fsync, and why sequential I/O is fast
• I/O models: blocking, non-blocking, epoll, io_uring. This is how Netty, Go’s netpoller, and Node work
• Synchronization primitives: mutex, semaphore, condition variable, futex
• Deadlock conditions (Coffman) and prevention
• Linux: signals, file descriptors, ulimit, cgroups + namespaces (this is what containers are)
• Memory: OOM killer, swap, container memory limits vs JVM heap (-XX:MaxRAMPercentage)

🧪 Labs (🟢 warm-up → 🟡 core → 🔴 hard → ⚫ boss)

• 🟢 strace -c a Java and a Go HTTP server under load; compare syscall mixes
• 🟡 Tiny shell in Go: fork/exec via os/exec, pipes, signals (Ctrl-C forwarding)
• 🟡 10k platform threads vs 10k virtual threads: vmstat 1 context switches + RSS
• 🔴 Write 1 GB with vs without fsync per 4 KB; explain it with the page cache
• ⚫ Build a “container” with unshare + cgroups v2 by hand (then in Go in W16)

🧠 Cognitive tasks

• Predict → verify: RSS of 10k platform threads (reserved vs committed stack)
• Symptom → hypotheses: “load average 40 on 8 cores, CPU 30%”
• Feynman: explain epoll to a junior in 3 minutes

🛰️ Orbit integration

• Size worker concurrency from measured context-switch and CPU numbers
• Container memory limits vs JVM MaxRAMPercentage for orbit-api

Go deeper

⚙️ Linux & OS Internals · Container Internals

Resources

• Operating Systems: Three Easy Pieces (free, ostep.org)
• The Linux Programming Interface (Kerrisk): reference
• Systems Performance 2e (Brendan Gregg): ch. 1–7

Interview questions interview-q

• Process vs thread? Why are goroutines/virtual threads cheaper?
• What happens on a page fault?
• How does epoll let one thread handle 10k connections?