Skip to content

~/avetyan.com/cheatsheets

Cheat sheets

Linux processes cheat sheet

Find, inspect, signal and prioritise processes on Linux: ps, top, kill and its signals, jobs, nohup, nice, ionice and /proc.

3 min read

Find processes

bash
ps aux                                  # everything, BSD style
ps -ef --forest                         # with the parent tree
ps -eo pid,ppid,user,%cpu,%mem,etime,cmd --sort=-%cpu | head
ps -o pid,stat,cmd -p 1234
pgrep -af nginx                         # PIDs and command lines
pidof nginx
pstree -p 1234

ps STAT codes

CodeMeaning
RRunning or runnable
SSleeping, waiting for something
DUninterruptible sleep, usually disk or NFS I/O
ZZombie: finished, parent hasn’t collected it
TStopped (Ctrl-Z or SIGSTOP)
< / NHigh / low priority
s / +Session leader / foreground group

Many processes in D push the load average up without using CPU.

top keys

KeyWhat it does
P / M / TSort by CPU / memory / time
1One line per CPU
cFull command lines
HShow threads
uOnly one user’s processes
kSend a signal
rRenice

htop does the same with a friendlier screen.

Signals

SignalNumber (x86, ARM)Default effect
SIGHUP1Hang up; many daemons reload their config
SIGINT2Interrupt (Ctrl-C)
SIGQUIT3Quit and dump core (Ctrl-\)
SIGKILL9Kill now; can’t be caught or ignored
SIGTERM15Ask politely to stop (the default)
SIGSTOP19Pause; can’t be caught
SIGCONT18Resume
SIGUSR110Program-defined (log reopen in nginx, for example)
bash
kill 1234                               # SIGTERM
kill -HUP 1234
kill -9 1234                            # last resort: no cleanup happens
pkill -f 'python worker.py'
killall -TERM nginx
kill -l                                 # list all signals

Jobs

bash
long-task &                             # run in the background
jobs -l
fg %1
bg %1                                   # resume a Ctrl-Z'd job in the background
disown %1                               # keep it running after logout
nohup long-task > task.log 2>&1 &
setsid long-task                        # in a new session, detached from the terminal

For anything that must survive a reboot, write a systemd unit instead.

Priority

bash
nice -n 10 tar -czf backup.tgz /data    # lower CPU priority (19 is lowest)
renice -n 5 -p 1234
ionice -c3 rsync -a /data /backup       # idle I/O class
ionice -c2 -n7 -p 1234                  # best effort, lowest
taskset -c 0,1 ./app                    # pin to CPUs 0 and 1

Look inside /proc

bash
cat /proc/1234/status                   # state, memory, threads
tr '\0' ' ' < /proc/1234/cmdline; echo  # full command line
tr '\0' '\n' < /proc/1234/environ       # environment (root or owner)
ls -l /proc/1234/fd                     # open files
ls -l /proc/1234/cwd                    # working directory
cat /proc/1234/limits                   # ulimits
cat /proc/1234/io                       # bytes read and written

Other tools

bash
lsof -p 1234                            # files and sockets it holds
strace -p 1234 -f -e trace=network      # what syscalls it makes
pmap -x 1234                            # memory map
timeout 30s ./slow-command              # kill it after 30 seconds
watch -n 2 'ps -eo pid,%cpu,cmd --sort=-%cpu | head'

Keep in mind

  • Zombies can’t be killed; they’re already dead. Fix or restart the parent that isn’t reaping them.
  • Send SIGTERM, wait a few seconds, then SIGKILL. A killed database or queue may need recovery on restart.

Comments

Sign in with Google, GitHub or Facebook to comment.

Loading comments…