#!/usr/bin/env bash
#
# JbTecWiz Support Centre -- generated fix script
#
# Fault : A Kubernetes node goes NotReady
# Fix   : Relieve the pressure the node is reporting
# Source: https://jbtecwiz.com/support/lnx-ctr-node-notready
#
# Run as     : Root shell on the node
# Expect     : 40 minutes
# Risk       : medium
# Reversible : yes
#
# WHEN THIS IS THE RIGHT FIX
#   DiskPressure or MemoryPressure.
#
# HOW TO UNDO IT
#   Configuration changes to the kubelet can be reverted and the service
#   restarted.
#
# 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 "  A Kubernetes node goes NotReady"
echo "  Relieve the pressure the node is reporting"
echo
echo "  Risk: medium   Reversible   40 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 'Read the exact condition and its message.' '' cmd 'kubectl describe node node01 | sed -n '\''/Conditions:/,/Addresses:/p'\'''; then
  kubectl describe node node01 | sed -n '/Conditions:/,/Addresses:/p'
  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 'For DiskPressure, find what filled up -- it is usually images or logs.' '' cmd 'df -h /var/lib/containerd /var/log /' 'sudo du -sh /var/log/pods /var/lib/containerd 2>/dev/null'; then
  df -h /var/lib/containerd /var/log /
  sudo du -sh /var/log/pods /var/lib/containerd 2>/dev/null
  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 'Let the kubelet'\''s own garbage collection do the work, then prune what remains.' 'crictl rmi --prune removes images not used by any container, which is what the kubelet would eventually do itself. Deleting files under /var/lib/containerd by hand corrupts the runtime'\''s metadata.' cmd 'sudo crictl images' 'sudo crictl rmi --prune'; then
  sudo crictl images
  sudo crictl rmi --prune
  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 the eviction thresholds are sensible for the disk size -- the defaults are percentage-based and can be far too aggressive on a large disk.' '' cmd 'sudo grep -E '\''evictionHard|imageGC'\'' /var/lib/kubelet/config.yaml'; then
  sudo grep -E 'evictionHard|imageGC' /var/lib/kubelet/config.yaml
  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 'Cap the log sizes so this does not recur.' '' cmd 'sudo grep -E '\''containerLogMaxSize|containerLogMaxFiles'\'' /var/lib/kubelet/config.yaml'; then
  sudo grep -E 'containerLogMaxSize|containerLogMaxFiles' /var/lib/kubelet/config.yaml
  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 node returns to Ready and the condition clears.'
if [ "$DRYRUN" = "0" ]; then
  kubectl get nodes; kubectl describe node node01 | grep -A6 Conditions
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-ctr-node-notready"
else
  echo "  Finished."
fi
echo
prose 'To undo: Configuration changes to the kubelet can be reverted and the service restarted.'
rule