php-fpm strace all (or some) instances

php-fpm_strace.sh

#!/bin/bash

# Default number of seconds to strace for
DEFAULT_STRACE_TIMEOUT=$((60*30))

# Job check frequency
DEFAULT_CHECK_INTERVAL=30

# Use timeout?
USE_TIMEOUT_CMD=n

# Set a maximum number of strace jobs to run
# If zero, will do all of them
MAX_WORKERS=0

msg_stamp ()
{
    date --iso-8601=seconds
}

msg_out ()
{
    echo -e "[$(msg_stamp)] $@"
}

msg_err ()
{
    msg_out $@ 1>&2
}

msg_fatal ()
{
    local exit_val=$1
    shift
    msg_err $@
    exit $exit_val
}

get_master_pid ()
{
    ps aux \
        | grep 'php-fpm: master process' \
        | grep -v grep \
        | awk '{ print $2 }'
}

get_worker_pids ()
{
    local master_pid=$(get_master_pid)
    msg_err "[DEBUG] master_pid=${master_pid}"
    pidof php5-fpm \
        | sed -e "s/\(^\|\s\)${master_pid}\(\s\|$\)/\2/g"
}

start_worker_strace ()
{
    local worker_pid=$1
    local strace_timeout=${2:-${DEFAULT_STRACE_TIMEOUT}}

    if [ "$USE_TIMEOUT_CMD" == "y" ]
    then
        ( ( \
            timeout ${strace_timeout} \
            strace -c -p ${worker_pid} \
          ) \
          2>&1 \
          | tee strace_php.${worker_pid} \
        ) &
    else
        ( ( strace -c -p ${worker_pid} ) \
          >> strace_php.${worker_pid} 2>&1 \
        ) &
    fi
    msg_out "(${worker_pid}) monitored by ($!)"
}

wait_on_workers ()
{
    local check_interval=${1:-${DEFAULT_CHECK_INTERVAL}}
    local running_pids=( )

    while :
    do
        # Fetch running jobs
        running_pids=( $(jobs -p) )

        # If no jobs running, break the loop
        [[ ${#running_pids[@]} -eq 0 ]] && break

        # Show waiting
        msg_out "(${#running_pids[@]}) remaining..."
        sleep $check_interval
    done

}

# Check that we're root
if [[ $UID != 0 ]]
then
    echo "ERROR: This script must be run as root or with sudo" >&2
    exit 1
fi

pids=( $(get_worker_pids) )
msg_out "Found ${#pids[@]} PID(s) to monitor..."
for pid in ${pids[@]}
do
    start_worker_strace \
        $pid \
        30

    # Should we limit how many strace workerss we're running?
    if [[ ${MAX_WORKERS} -gt 0 ]]
    then
        # Yes, limit them. If at the limit, break this loop
        [[ $(jobs -p | wc -l) -ge ${MAX_WORKERS} ]] \
            && break
    fi
done

# Wait on all the strace invocations to complete
wait_on_workers