#!/usr/bin/env bash
#
# JbTecWiz Support Centre -- generated fix script
#
# Fault : "Too many open files" (EMFILE)
# Fix   : Track down a descriptor leak
# Source: https://jbtecwiz.com/support/lnx-too-many-open-files
#
# Run as     : Shell as root
# Expect     : 1 hour
# Risk       : low
# Reversible : yes
#
# WHEN THIS IS THE RIGHT FIX
#   The count climbs steadily and never falls. Raising the limit only
#   delays the failure.
#
# HOW TO UNDO IT
#   None -- diagnostic.
#
# 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 <n> <do> <why> <manual|cmd> [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 "  "Too many open files" (EMFILE)"
echo "  Track down a descriptor leak"
echo
echo "  Risk: low   Reversible   1 hour"
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 'Watch the count over time.' '' cmd 'pid=$(pgrep -f myapp | head -1)' 'watch -n5 "ls /proc/$pid/fd | wc -l"'; then
  pid=$(pgrep -f myapp | head -1)
  watch -n5 "ls /proc/$pid/fd | wc -l"
  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 'Group the open descriptors by type to see what is accumulating.' '' cmd 'sudo ls -l /proc/$pid/fd | awk '\''{print $11}'\'' | sed '\''s/[0-9]*$//'\'' | sort | uniq -c | sort -rn | head'; then
  sudo ls -l /proc/$pid/fd | awk '{print $11}' | sed 's/[0-9]*$//' | sort | uniq -c | sort -rn | head
  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 'Look for sockets stuck in CLOSE_WAIT -- that pattern means the application is not closing connections the peer has already finished with.' 'CLOSE_WAIT is entirely the local application'\''s responsibility. A large and growing count of them is a code bug, not a tuning problem.' cmd 'sudo ss -tanp | grep CLOSE-WAIT | head -20'; then
  sudo ss -tanp | grep CLOSE-WAIT | head -20
  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 'Check for repeatedly opened files that are never closed.' '' cmd 'sudo lsof -p $pid | awk '\''{print $9}'\'' | sort | uniq -c | sort -rn | head'; then
  sudo lsof -p $pid | awk '{print $9}' | sort | uniq -c | sort -rn | head
  if [ $? -ne 0 ]; then
    failed=$((failed+1))
    echo "  Step 4 failed. The rest of the fix may depend on it." >&2
  fi
fi

step 5 'Raise the limit as a stopgap so the service survives while the leak is fixed, and add a restart schedule if needed.' '' manual || true

rule "  Confirm it worked"
prose 'The descriptor count is stable under sustained load.'

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-too-many-open-files"
else
  echo "  Finished."
fi
echo
prose 'To undo: None -- diagnostic.'
rule