Mastering Pkill And Pgrep For Process Management

pgrep and pkill are essential tools for finding and signaling processes by name and attributes. They’re far more precise than grepping ps output and should be your go-to for process management scripts and one-liners.

pgrep: Find Process IDs

pgrep searches running processes and outputs matching PIDs. All selection criteria must match unless you use alternatives (comma-separated values act as OR).

pgrep -u root sshd        # sshd processes owned by root
pgrep -u root,daemon      # processes owned by root OR daemon
pgrep -f "python main.py" # match full command line
pgrep -x nginx            # exact process name match only

pkill: Signal Processes

pkill sends a signal (default SIGTERM) to matching processes instead of listing them:

pkill -HUP syslogd           # send SIGHUP to syslogd
pkill -9 firefox             # SIGKILL unresponsive processes
pkill -TERM -u username      # terminate all processes by user
pkill -f "node.*worker"      # signal by command line pattern

Common Options

Selection criteria:

-u euid – match by effective user ID (name or number)
-U uid – match by real user ID
-g pgrp – match by process group ID
-G gid – match by group ID (name or number)
-P ppid – match by parent process ID
-t term – match by controlling terminal (e.g., pts/0)
-s sid – match by session ID

Matching:

-f – match against full command line, not just process name. Crucial for scripts with generic names
-x – exact match only (no substring matching)
-v – invert match (exclude matching processes)

Process selection:

-n – select only the newest matching process
-o – select only the oldest matching process
-l – list process names alongside PIDs (pgrep only)
-d delimiter – change output delimiter (pgrep only)

Signals (pkill):

Use numeric (-9) or symbolic (-KILL) form
Common: TERM (15, graceful), KILL (9, force), HUP (1, reload config)

Practical Examples

Reload a service without finding the PID manually:

pkill -HUP nginx          # nginx rereads config
pkill -TERM -u appuser    # stop all processes for a user

Combine with other tools:

# Get detailed info on matching processes
ps -fp $(pgrep -d, -x python3)

# Count matching processes
pgrep -c redis

# Kill all zombies from a parent
pkill -P 1234 -9

Match by command line:

# Kill specific Python script (not all Python processes)
pkill -f "celery worker"

# Only exact full command match
pkill -fx "/usr/bin/python3 /opt/app/main.py"

Age-based selection:

pgrep -n -u postgres    # newest postgres process
pgrep -o java           # oldest Java process

Important Behavior

pgrep and pkill never match themselves
Process names are limited to 15 characters in /proc/[pid]/stat — use -f to match against the full /proc/[pid]/cmdline
Extended Regular Expressions are supported in patterns
Exit status: 0 if matches found, 1 if no matches, 2 for syntax error, 3 for fatal error

Avoid Common Mistakes

Don’t use bare process names with common tools:

# BAD: might match unexpected processes
pkill java

# GOOD: be specific with full command matching
pkill -f "java.*myapp.jar"

Verify matches before signaling:

# Preview what you're about to kill
pgrep -f "nginx master"

# Then signal it
pkill -f "nginx master"

Combine criteria carefully:

# This finds processes owned by root AND in session 1234
pgrep -u root -s 1234

# Comma-separated values create OR logic
pgrep -u root,daemon   # owned by root OR daemon

Related Commands

ps(1) – detailed process listing
kill(1) – signal a process by PID
killall(1) – similar to pkill but doesn’t support regex
systemctl(1) – preferred for managing services on systemd systems

pgrep and pkill outperform grepping ps output because they’re atomic operations that avoid race conditions and handle edge cases properly. Use them consistently in scripts rather than shell pipelines for robustness.

Mastering pkill and pgrep for Process Management