Process & Resource Management

Debug running Rust services with ps, htop, signals, and perf. Validate graceful shutdown, thread counts, and memory growth under load.

Recipe

pgrep -a my-service
kill -TERM "$(pgrep my-service)"
sleep 2
pgrep my-service || echo "exited cleanly"
 
htop -p "$(pgrep -d',' my-service)"

When to reach for this:

• Service won't stop on deploy
• CPU pegged after traffic spike
• Investigating memory leak suspicions

Working Example

# Watch open files for connection leak
PID=$(pgrep my-service)
ls -l /proc/$PID/fd | wc -l
 
# Sample stack with perf (Linux)
sudo perf record -g -p $PID -- sleep 30
sudo perf report
 
# Memory map summary
pmap -x $PID | tail -1

// Ensure SIGTERM handled in service
use tokio::signal;
 
async fn shutdown_signal() {
    signal::ctrl_c().await.ok();
    signal::unix::signal(signal::unix::SignalKind::terminate())
        .unwrap()
        .recv()
        .await
        .ok();
}

What this demonstrates:

• TERM before KILL matches Kubernetes pod termination
• /proc/$PID/fd counts open sockets/files
• perf attributes CPU to symbols (with debug symbols)

Deep Dive

Signals

Tokio Runtime Threads

Default multi-thread runtime uses ~N worker threads; blocking pool adds more under spawn_blocking load.

Gotchas

• SIGKILL in tests - hides shutdown bugs. Fix: integration test SIGTERM path.
• perf without permissions - perf_event_paranoid. Fix: sysctl adjust or sudo.
• Debug symbols stripped - useless perf stacks. Fix: retain symbols in staging builds.
• Reading RSS only - allocator caches memory. Fix: track over time and load patterns.
• Wrong PID after reload - systemd restarts. Fix: systemctl status for current PID.

Alternatives

FAQs

Related

• Graceful shutdown
• Production debugging
• Performance Skill
• Sysadmin essentials

Stack versions: This page was written for Rust 1.97.0 (edition 2024), Tokio 1.x, Axum 0.8, serde 1.0, sqlx 0.8, clap 4, and Polars 0.46+.