Process Management on Linux
Everything that runs on a Linux system is a process. The kernel schedules them, the init system supervises them, and you control them through a handful of tools that haven't changed much in thirty years — for good reason.
ps: snapshot the process table
The Unix ps command is the oldest way to list processes. It supports two competing flag styles (BSD and SysV) — most people memorize one incantation:
ps aux # all processes, user-friendly
ps -ef # all processes, full format
ps -fp 12345 # specific PID
ps --forest -e # tree view, every processThe columns you care about: PID, %CPU, %MEM, VSZ (virtual size), RSS (resident set), STAT (state codes — R running, S sleeping, D uninterruptible, Z zombie).
top and htop: live view
top is the BSD-era live monitor and ships everywhere. htop is the modern replacement: colorized, mouse-aware, tree mode with F5, kill via F9. Install it and never look back.
htop # interactive
top -b -n 1 | head -20 # batch mode for scripts
top -p $(pgrep -d, nginx) # only show nginx procsSignals and kill
kill doesn't necessarily kill — it sends a signal. The kernel delivers signals to processes; what happens next depends on what the process does with them.
kill -TERM 12345 # SIGTERM — "please exit cleanly" (default)
kill -HUP 12345 # SIGHUP — most daemons reload config
kill -KILL 12345 # SIGKILL — uncatchable, the kernel reaps the process
kill -9 12345 # same thing, numeric
pkill -f my-script.py # kill by command-line match
killall nginx # kill by exact nameSIGKILL is the option of last resort. A process stuck in D state (uninterruptible sleep — usually waiting on disk or NFS) won't even respond to SIGKILL until the syscall returns.
Priority: nice and ionice
The scheduler decides who runs next. nice lets you suggest priority for CPU; values range from -20 (highest) to +19 (lowest). Only root can request negative values.
nice -n 10 ./long-build.sh # run at lower priority
renice 15 -p 12345 # change priority of running process
ionice -c 3 ./backup.sh # idle I/O class — yields to anything else/proc: the kernel's filesystem
Every running process has a directory under /proc/<pid>/. Reading from it is how ps works.
cat /proc/12345/status # human-readable state
cat /proc/12345/cmdline | tr '\0' ' ' # full command line
ls -l /proc/12345/fd # all open file descriptors
readlink /proc/12345/exe # path to the binary
cat /proc/12345/maps # memory mapWhen memory runs out: the OOM killer
When the kernel can't allocate memory, it picks a victim and sends SIGKILL. Selection is based on the oom_score in /proc/<pid>/oom_score. You can bias it:
echo -1000 > /proc/$$/oom_score_adj # never kill this shell
dmesg | grep -i 'killed process' # see who got reapedcommand & runs in background. jobs lists them, fg %1 brings job 1 to foreground, bg %1 resumes a suspended job. disown detaches a job from the shell so it survives logout.