#!/usr/bin/env bash
#
# JbTecWiz Support Centre -- generated fix script
#
# Fault : "bash: fork: retry: Resource temporarily unavailable"
# Fix   : Recover a machine that cannot fork at all
# Source: https://jbtecwiz.com/support/lnx-fork-resource
#
# Run as     : Existing shell or console
# Expect     : 20 minutes
# Risk       : high
# Reversible : yes
#
# WHEN THIS IS THE RIGHT FIX
#   Nothing can start. Use shell builtins, which need no fork, wherever
#   possible.
#
# HOW TO UNDO IT
#   Delete /etc/sysctl.d/99-pidmax.conf and run sysctl --system.
#
# 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

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 <n> <do> <why> <manual|cmd> [command lines...]
  local n="$1" dotext="$2" why="$3" mode="$4"; shift 4
  rule "  Step $n of 6"
  prose "$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 "  "bash: fork: retry: Resource temporarily unavailable""
echo "  Recover a machine that cannot fork at all"
echo
echo "  Risk: high   Reversible   20 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 'Use builtins to look around -- echo and read are builtins and will work when /bin/ls will not.' 'Globbing and echo are handled inside bash with no fork. Anything that calls an external binary needs a process slot you do not have.' cmd 'echo /proc/[0-9]*/ | wc -w'; then
  echo /proc/[0-9]*/ | wc -w
  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 'Read the system-wide pid ceiling.' '' cmd 'cat /proc/sys/kernel/pid_max'; then
  cat /proc/sys/kernel/pid_max
  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 'Raise it temporarily to buy room to work.' '' cmd 'echo 131072 > /proc/sys/kernel/pid_max'; then
  echo 131072 > /proc/sys/kernel/pid_max
  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 'Now that you can fork, find the offender.' '' cmd 'ps -eo user,pid,ppid,cmd --sort=-ppid | head -40'; then
  ps -eo user,pid,ppid,cmd --sort=-ppid | head -40
  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 'Kill the whole process group rather than individual processes, which respawn faster than you can kill them.' '' cmd 'sudo systemctl stop runaway.service' 'sudo pkill -9 -u appuser'; then
  sudo systemctl stop runaway.service
  sudo pkill -9 -u appuser
  if [ $? -ne 0 ]; then
    failed=$((failed+1))
    echo "  Step 5 failed. The rest of the fix may depend on it." >&2
  fi
fi

if step 6 'Make the ceiling permanent if it was genuinely too low for the workload.' '' cmd 'echo '\''kernel.pid_max = 131072'\'' | sudo tee /etc/sysctl.d/99-pidmax.conf' 'sudo sysctl --system'; then
  echo 'kernel.pid_max = 131072' | sudo tee /etc/sysctl.d/99-pidmax.conf
  sudo sysctl --system
  if [ $? -ne 0 ]; then
    failed=$((failed+1))
    echo "  Step 6 failed. The rest of the fix may depend on it." >&2
  fi
fi

rule "  Confirm it worked"
prose 'Commands run normally and the process count is stable.'
if [ "$DRYRUN" = "0" ]; then
  ps -eo user= | wc -l; cat /proc/sys/kernel/pid_max
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-fork-resource"
else
  echo "  Finished."
fi
echo
prose 'To undo: Delete /etc/sysctl.d/99-pidmax.conf and run sysctl --system.'
rule