Linux Signals

Signal delivery, common signals, signal masking, SIGTERM vs SIGKILL, and graceful shutdown patterns.

Overview

Signals are asynchronous notifications sent to a process by the kernel, another process, or the process itself. There are 64 standard signals. Signals interrupt normal execution and invoke a registered handler, the default action (usually terminate or ignore), or are blocked via a signal mask.

Key Signals for Production

SIGTERM (15) — request graceful termination; processes can catch and clean up
SIGKILL (9) — unconditional termination; cannot be caught, blocked, or ignored
SIGINT (2) — Ctrl+C from the terminal
SIGHUP (1) — trigger config reload in daemons
SIGCHLD (17) — sent to parent when a child changes state
SIGUSR1 / SIGUSR2 (10/12) — user-defined signals

Signal Delivery

If a process has a signal blocked in its signal mask, it is queued (pending). Signals are not queued past one — if the same signal fires three times while blocked, only one delivery occurs when unblocked. Real-time signals (SIGRTMIN to SIGRTMAX) are queued and ordered.

Graceful Shutdown Pattern

Receive SIGTERM → stop accepting new work → drain in-flight requests → close connections → exit 0. Kubernetes uses terminationGracePeriodSeconds for this.

Key concepts

SIGTERM (15): Request graceful shutdown. Can be caught and handled. Always try first — allows the process to flush buffers and exit cleanly.
SIGKILL (9): Unconditional kill. Cannot be caught, blocked, or ignored. Use as a last resort after SIGTERM has been given time to work.
SIGHUP (1): Modern convention: reload configuration without restarting. nginx, sshd, and most daemons handle this via a registered SIGHUP handler.
Signal mask: Per-thread bitmask of blocked signals. Blocked signals are queued (pending) and delivered when unblocked.
Real-time signals: SIGRTMIN through SIGRTMAX. Unlike standard signals, they are queued, ordered, and can carry a payload.

When to use it

Explaining why kill -9 should be the last resort — it prevents graceful cleanup
Designing container shutdown sequences in Kubernetes (preStop hooks + SIGTERM + grace period)
Debugging why a daemon does not pick up config changes — likely needs SIGHUP

Interview questions

What happens when you run kill -9 on a process vs kill -15? When should you use each?
How do you ensure your Docker container shuts down gracefully when Kubernetes drains a node?