#!/usr/bin/env bash # # JbTecWiz Support Centre -- generated fix script # # Fault : "server reached pm.max_children setting" -- the site stalls under load # Fix : Size the pool for the memory available # Source: https://jbtecwiz.com/support/lnx-web-fpm-children # # Run as : Root shell # Expect : 30 minutes # Risk : medium # Reversible : yes # # WHEN THIS IS THE RIGHT FIX # Requests are reasonable and there are genuinely more of them. # # HOW TO UNDO IT # Restore the previous values in www.conf and reload. # # Walks the fix one step at a time and asks before each. Steps with no # command are yours to do -- it prints those and waits. DRYRUN=1 prints # without executing; UNATTENDED=1 does not ask. # # -------------------------------------------------------------------- # NO WARRANTY - USE AT YOUR OWN RISK # # This script is provided by JbTecWiz as-is and with no warranty of any # kind, express or implied. You run it entirely at your own risk. # # JbTecWiz accepts no liability for any loss or damage arising from its # use, including but not limited to data loss, downtime, or configuration # changes that turn out to be wrong for your system. # # You are responsible for reading this script before running it, for # satisfying yourself that it suits the machine in front of you, and for # having a working backup first. Some steps cannot be undone. # -------------------------------------------------------------------- set -uo pipefail DRYRUN="${DRYRUN:-0}" UNATTENDED="${UNATTENDED:-0}" failed=0 if [ "$(id -u)" -ne 0 ]; then echo " This fix is documented as needing root. Re-run with sudo." >&2 exit 3 fi rule() { printf "\n%s\n" "$(printf '-%.0s' $(seq 1 70))"; if [ $# -gt 0 ]; then echo "$1"; fi; } prose() { echo "$1" | fold -s -w 74 | sed "s/^/ /"; } # Returns 0 when the caller should run the command, 1 when it should not. # A manual step always returns 1 -- there is nothing for the caller to run. step() { # step [command lines...] local n="$1" dotext="$2" why="$3" mode="$4"; shift 4 rule " Step $n of 5" prose "$dotext" if [ -n "$why" ]; then echo; prose "$why"; fi if [ "$mode" = "manual" ]; then echo; echo " -> Do this yourself, then press Enter to carry on." if [ "$UNATTENDED" = "0" ] && [ "$DRYRUN" = "0" ]; then read -r _; fi return 1 fi echo; printf " %s\n" "$@"; echo if [ "$DRYRUN" = "1" ]; then echo " (dry run -- not executed)"; return 1; fi if [ "$UNATTENDED" = "0" ]; then read -r -p " Run this step? [Y]es / [S]kip / [Q]uit " a case "$a" in [Qq]*) echo " Stopped at your request."; exit 0 ;; [Ss]*) echo " Skipped."; return 1 ;; esac fi return 0 } rule echo " "server reached pm.max_children setting" -- the site stalls under load" echo " Size the pool for the memory available" echo echo " Risk: medium Reversible 30 minutes" echo prose 'No warranty. Use at your own risk - JbTecWiz accepts no liability. Read it before you run it, and have a backup.' rule echo if [ "$UNATTENDED" = "0" ] && [ "$DRYRUN" = "0" ]; then read -r -p " Ready? [y/N] " go case "$go" in [Yy]*) ;; *) echo " Nothing was changed."; exit 0;; esac fi if step 1 'Measure what one worker actually uses rather than guessing.' 'The right value is available memory divided by the real per-worker figure, leaving room for the database and the operating system. A number copied from a blog post is how servers end up swapping under load, which is far worse than queueing.' cmd 'ps -o rss=,comm= -C php-fpm8.2 | awk '\''{s+=$1; n++} END {printf "workers=%d avg=%dMB\n", n, s/n/1024}'\'''; then ps -o rss=,comm= -C php-fpm8.2 | awk '{s+=$1; n++} END {printf "workers=%d avg=%dMB\n", n, s/n/1024}' if [ $? -ne 0 ]; then failed=$((failed+1)) echo " Step 1 failed. The rest of the fix may depend on it." >&2 fi fi if step 2 'Check total memory and what else is using it.' '' cmd 'free -m' 'ps -eo rss,comm --sort=-rss | head -10'; then free -m ps -eo rss,comm --sort=-rss | head -10 if [ $? -ne 0 ]; then failed=$((failed+1)) echo " Step 2 failed. The rest of the fix may depend on it." >&2 fi fi if step 3 'Set the values consistently -- max_children is the ceiling, and the start/min/max spare values must be below it.' '' cmd 'sudo sed -i '\''s/^pm.max_children.*/pm.max_children = 30/; s/^pm.start_servers.*/pm.start_servers = 8/; s/^pm.min_spare_servers.*/pm.min_spare_servers = 5/; s/^pm.max_spare_servers.*/pm.max_spare_servers = 12/'\'' /etc/php/8.2/fpm/pool.d/www.conf'; then sudo sed -i 's/^pm.max_children.*/pm.max_children = 30/; s/^pm.start_servers.*/pm.start_servers = 8/; s/^pm.min_spare_servers.*/pm.min_spare_servers = 5/; s/^pm.max_spare_servers.*/pm.max_spare_servers = 12/' /etc/php/8.2/fpm/pool.d/www.conf if [ $? -ne 0 ]; then failed=$((failed+1)) echo " Step 3 failed. The rest of the fix may depend on it." >&2 fi fi if step 4 'Add a request limit so a leaking worker is recycled rather than growing forever.' '' cmd 'sudo sed -i '\''s/^;*pm.max_requests.*/pm.max_requests = 500/'\'' /etc/php/8.2/fpm/pool.d/www.conf' 'sudo systemctl reload php8.2-fpm'; then sudo sed -i 's/^;*pm.max_requests.*/pm.max_requests = 500/' /etc/php/8.2/fpm/pool.d/www.conf sudo systemctl reload php8.2-fpm if [ $? -ne 0 ]; then failed=$((failed+1)) echo " Step 4 failed. The rest of the fix may depend on it." >&2 fi fi if step 5 'Confirm the machine does not swap under peak load -- if it does, the number is too high regardless of what the calculation said.' '' cmd 'vmstat 2 10'; then vmstat 2 10 if [ $? -ne 0 ]; then failed=$((failed+1)) echo " Step 5 failed. The rest of the fix may depend on it." >&2 fi fi rule " Confirm it worked" prose 'The warning stops appearing and free memory stays positive at peak.' if [ "$DRYRUN" = "0" ]; then sudo grep -c max_children /var/log/php8.2-fpm.log fi rule if [ "$failed" -gt 0 ]; then echo " Finished with $failed failed step(s)." echo " Read the full write-up at https://jbtecwiz.com/support/lnx-web-fpm-children" else echo " Finished." fi echo prose 'To undo: Restore the previous values in www.conf and reload.' rule