Why a Crontab Script Miscounts Processes: PATH and pidof Gotchas

Problem Origin

A program needs to restart automatically after a crash, so a Linux crontab entry runs a script every minute to check whether the video_checkup process is running. When the script is executed manually in a shell, it works, but when run by cron it repeatedly starts the process, creating many instances.

Problem Description

The monitoring script /home/watch.sh contains:

#!/bin/sh
shell_log_file=/home/start.log
pid_count=`pidof video_checkup | wc -w`
path=$(cd "$(dirname "$0")"; pwd)
run_command="${path}/video_checkup"
config_path="${path}/config.json"
if [ $pid_count -eq 0 ]; then
    echo `date +%Y-%m-%d_%H:%M:%S` " run $run_command $config_path" >> $shell_log_file
    $run_command $config_path
else
    echo `date +%Y-%m-%d_%H:%M:%S` " video_checkup already running" >> $shell_log_file
fi

The crontab line added is: * * * * * /home/watch.sh > /dev/null 2>&1 Despite the logic, the video_checkup process count keeps increasing because pid_count is always evaluated as zero.

Analysis

Debugging showed that the conditional if [ $pid_count -eq 0 ] is always true when the script runs under cron. Logging the environment with export revealed that PATH in the cron job is limited to /usr/bin:/bin, while an interactive shell has a much richer PATH (e.g.,

/usr/local/sbin:/usr/local/bin:/sbin:/bin:/usr/sbin:/usr/bin:/root/bin

).

The pidof command resides in /sbin. Because /sbin is missing from the cron PATH, the command is not found. The failure is silent when piped to wc -w; wc receives no input and returns 0, so pid_count becomes zero every time.

Solution

Explicitly set a proper PATH at the beginning of watch.sh so that pidof can be located:

PATH="/usr/local/sbin:/usr/local/bin:/sbin:/bin:/usr/sbin:/usr/bin:/root/bin"

After adding this line, the script correctly detects the running video_checkup process and stops launching duplicate instances.

This case illustrates how differences between the cron environment and an interactive shell can cause subtle bugs in automation scripts.