[distro-setup] / system-status

#!/bin/bash
# Copyright (C) 2019 Ian Kelling
# SPDX-License-Identifier: AGPL-3.0-or-later

# usage: runs 4 times every 15 seconds unless any args are passed, or we
# are on battery power, then just runs once.

if [ -z "$BASH_VERSION" ]; then echo "error: shell is not bash" >&2; exit 1; fi

source /a/bin/errhandle/err
status_file=/dev/shm/iank-status

shopt -s nullglob
shopt -s dotglob

verbose=false
if [[ $1 ]]; then
  verbose=true
fi
v() {
  if $verbose; then
    printf "%s\n" "$*"
  fi
}
lo() { /usr/local/bin/log-once "$@"; }

write-status() {
  chars=("${first_chars[@]}")

  glob=(/nocow/btrfs-stale/*)
  if [[ -e ${glob[0]} ]]; then
    chars+=("STALE!")
  fi
  glob=(/m/md/bounces/new/*)
  if [[ -e ${glob[0]} ]]; then
    chars+=("BOUNCE!")
    bouncemsg="message in /m/md/bounces/new"
  fi
  lo -1 bounce $bouncemsg
  glob=(/m/md/alerts/new/* /m/md/alerts/cur/*)
  if [[ -e ${glob[0]} ]]; then
    chars+=("ALERT!")
  fi
  if [[ -e /nocow/user/mailtest-failure ]]; then
    chars+=("MAILPING!")
  fi

  if ! qlen=$(/usr/sbin/exiqgrep -o 60 -c -b | awk '{print $1}'); then
    # early in install process, we dont have permission yet for exiqgrep
    qlen=$(sudo /usr/sbin/exiqgrep -o 60 -c -b | awk '{print $1}')
  fi
  if ((qlen)); then
    chars+=("q $qlen")
  fi

  begin=false
  cd /b/ds
  if ! make -q ~/.local/distro-begin || [[ $(<~/.local/distro-begin) != 0 ]]; then
    begin=true
  fi

  end=false
  if ! make -q ~/.local/distro-end || [[ $(<~/.local/distro-end) != 0 ]]; then
    end=true
  fi

  # these conditions are so we dont have an overly verbose prompt
  if $begin && $end; then
    chars+=("DISTRO!")
  elif $begin; then
    chars+=("DISTRO-BEGIN!")
  elif $end; then
    chars+=("DISTRO-END!")
  else
    f=~/.local/conflink
    for _ in 1; do
      if [[ -e $f ]]; then
        now=$(date +%s)
        fsec=$(stat -c%Y $f)
        # the / 60 makes it 0-59 seconds less strict, +1 makes sure we
        # dont have any false positives.
        fmin=$(( (fsec - now ) / 60 + 1 ))
        fminplus=$(( fmin + 60*24 ))
        # Filesystem files get copied, so find any newer than the last run.
        # The rest are hueristics:
        # Given the last time we added a file in git, is that newer than the last conflink run.
        # Given new files not added to git, were they modified more recently than the last conflink? but,
        # push their modification time back by a day so we can develop them before needing to add them to git.

        all_dirs=({/a/bin/ds,/p/c}{/filesystem,/machine_specific/$HOSTNAME/filesystem})
        # This part is copied from conflink
        for x in /p/c/machine_specific/*.hosts /a/bin/ds/machine_specific/*.hosts; do
          if grep -qxF $HOSTNAME $x; then all_dirs+=( ${x%.hosts} ); fi
        done

        # just because i forget a lot, -mmin -NUM means files modified <= NUM minutes ago
        if (( $(date -d "$(git log --diff-filter=ACR --format=%aD -1)" +%s) > fsec )) || \
             [[ $(find ${all_dirs[@]} -mmin $fmin -type f -print -quit 2>/dev/null) ]]; then
          v conflink newer git or newer filesystem files
          chars+=("CONFLINK!")
          break
        fi

        for d in /a/bin/distro-setup /p/c; do
          cd $d
          if [[ ! -e .git ]]; then
            # some hosts i dont push all of /p/c
            continue
          fi
          untracked=$(git ls-files -o --exclude-standard)
          if [[ $untracked && $(find $untracked -mmin $fminplus -type f -print -quit) ]]; then
            v conflink: untracked in $d
            chars+=("CONFLINK!")
            break
          fi
        done

      fi
      if [[ ! -e $f || $(<$f) != 0 ]]; then
        v conflink: last run not found or failed
        chars+=("CONFLINK!")
        break
      fi
    done
  fi


  ## Clean the paniclog, but only up to 4 times per day, or else we
  ## should investigate.
  loglog=/tmp/panicloglog-$(date --rfc-3339=date)
  if [[ -s $loglog ]]; then
    spamcount=$(stat -c%s $loglog)
  else
    spamcount=0
  fi
  if (( spamcount <= 4 )); then
    if grep -q 'spam acl condition' /var/log/exim4/paniclog &>/dev/null; then
      printf . >>$loglog
    fi
    /a/bin/distro-setup/epanic-clean
  fi

  if [[ -s /var/log/exim4/paniclog ]]; then
    chars+=("PANIC!")
    tail -n 20 /var/log/exim4/paniclog | lo -1 paniclog
  else
    lo -1 paniclog
  fi

  source /a/bin/bash_unpublished/source-state
  if [[ $MAIL_HOST == "$HOSTNAME" ]]; then
    bbkmsg=
    if [[ $(systemctl is-active btrbk.timer) != active ]]; then
      chars+=("BTRBK.TIMER!")
      bbkmsg="btrbk.timer not enabled"
    fi
    lo -60 btrbk.timer $bbkmsg

    ## check if last snapshot was within an hour
    vol=o
    # this section generally copied from btrbk scripts, but
    # this part modified to speed things up by about half a second.
    # I'm not sure if its quite as reliable, but it looks pretty safe.
    # Profiled it using time and also adding to the top of the file:
    # set -x
    # PS4='+ $(date "+%2N") '
    snaps=($(ls -1avdr /mnt/root/btrbk/$vol.20*|head -n1))
    now=$(date +%s)
    maxtime=0
    for s in ${snaps[@]}; do
      file=${s##*/}
      t=$(date -d $(sed -r  's/(.{4})(..)(.{5})(..)(.*)/\1-\2-\3:\4:\5/' <<<${file#$vol.}) +%s)
      if (( t > maxtime )); then
        maxtime=$t
      fi
    done
    if (( maxtime < now - 60*60 )); then
      chars+=("OLD-SNAP!")
      snapshotmsg="/o snapshot older than 1 hour"
    fi
    lo -1 old-snapshot $snapshotmsg
  fi

  cat /a/bin/bash_unpublished/source-state >$status_file

  if [[ ${chars[*]} ]]; then
    echo "ps_char=\"${chars[*]} \$ps_char\"" >>$status_file
  fi

}
# use this if we want to do something just once per minute
first_chars=()

power=true
if [[ -e /sys/class/power_supply/AC/online && $(</sys/class/power_supply/AC/online) == 0 ]]; then
  power=false
fi

write-status
if [[ $1 ]]; then
  cat $status_file
  exit 0
fi

if ! $power; then
  exit 0
fi
for ((i=1; i<=3; i++)); do
  sleep 15
  write-status
done