iankelling.org Git - distro-setup/blob - system-status

   1 #!/bin/bash
   2 # Copyright (C) 2019 Ian Kelling
   3 # SPDX-License-Identifier: AGPL-3.0-or-later
   4
   5 # usage: runs 4 times every 15 seconds unless any args are passed, or we
   6 # are on battery power, then just runs once.
   7
   8 if [ -z "$BASH_VERSION" ]; then echo "error: shell is not bash" >&2; exit 1; fi
   9
  10 source /a/bin/errhandle/err
  11 status_file=/dev/shm/iank-status
  12
  13 shopt -s nullglob
  14 shopt -s dotglob
  15
  16 verbose=false
  17 if [[ $1 ]]; then
  18   verbose=true
  19 fi
  20 v() {
  21   if $verbose; then
  22     printf "%s\n" "$*"
  23   fi
  24 }
  25 lo() { /usr/local/bin/log-once "$@"; }
  26
  27 write-status() {
  28   chars=("${first_chars[@]}")
  29
  30   glob=(/nocow/btrfs-stale/*)
  31   if [[ -e ${glob[0]} ]]; then
  32     chars+=("STALE!")
  33   fi
  34   if [[ $(find /var/mail -type f \! -empty -print -quit) ]]; then
  35     var_mail_msg="message in /var/mail"
  36   fi
  37   lo -1 var_mail $var_mail_msg
  38   glob=(/m/md/bounces/new/*)
  39   if [[ -e ${glob[0]} ]]; then
  40     chars+=("BOUNCE!")
  41     bouncemsg="message in /m/md/bounces/new"
  42   fi
  43   lo -1 bounce $bouncemsg
  44   glob=(/m/md/alerts/new/* /m/md/alerts/cur/*)
  45   if [[ -e ${glob[0]} ]]; then
  46     chars+=("ALERT!")
  47   fi
  48   if [[ -e /nocow/user/mailtest-failure ]]; then
  49     chars+=("MAILPING!")
  50   fi
  51
  52   if ! qlen=$(/usr/sbin/exiqgrep -o 60 -c -b | awk '{print $1}'); then
  53     # early in install process, we dont have permission yet for exiqgrep
  54     qlen=$(sudo /usr/sbin/exiqgrep -o 60 -c -b | awk '{print $1}')
  55   fi
  56   if ((qlen)); then
  57     qmsg="queue length $qlen"
  58     chars+=("q $qlen")
  59   fi
  60   case $HOSTNAME in
  61     # No point in emailing about the mailq on a host where we don't
  62     # check email.
  63     $MAIL_HOST|l2)
  64       lo -1 qlen $qmsg
  65       ;;
  66   esac
  67
  68   begin=false
  69   if ! make -C /b/ds -q ~/.local/distro-begin || [[ $(<~/.local/distro-begin) != 0 ]]; then
  70     begin=true
  71   fi
  72
  73   end=false
  74   if ! make -C /b/ds -q ~/.local/distro-end || [[ $(<~/.local/distro-end) != 0 ]]; then
  75     end=true
  76   fi
  77
  78   # these conditions are so we dont have an overly verbose prompt
  79   if $begin && $end; then
  80     chars+=("DISTRO!")
  81   elif $begin; then
  82     chars+=("DISTRO-BEGIN!")
  83   elif $end; then
  84     chars+=("DISTRO-END!")
  85   else
  86     f=~/.local/conflink
  87     for _ in 1; do
  88       if [[ -e $f ]]; then
  89         now=$(date +%s)
  90         fsec=$(stat -c%Y $f)
  91         # the / 60 makes it 0-59 seconds less strict, +1 to help make sure we
  92         # dont have any false positives.
  93         fmin=$(( (fsec - now + 1 ) / 60 ))
  94         fminplus=$(( fmin + 60*24 ))
  95         # Filesystem files get copied, so find any newer than the last run.
  96         # The rest are hueristics:
  97         # Given the last time we added a file in git, is that newer than the last conflink run.
  98         # Given new files not added to git, were they modified more recently than the last conflink? but,
  99         # push their modification time back by a day so we can develop them before needing to add them to git.
 100
 101         all_dirs=({/a/bin/ds,/p/c}{/filesystem,/machine_specific/$HOSTNAME/filesystem})
 102         # This part is copied from conflink
 103         for x in /p/c/machine_specific/*.hosts /a/bin/ds/machine_specific/*.hosts; do
 104           if grep -qxF $HOSTNAME $x; then all_dirs+=( ${x%.hosts} ); fi
 105         done
 106
 107         # Just because i forget a lot, -mmin -NUM means files modified <= NUM minutes ago
 108         if (( fmin < 0 )) && [[ $(find ${all_dirs[@]} -mmin $fmin -type f -print -quit 2>/dev/null) ]]; then
 109           v conflink newer filesystem files
 110           chars+=("CONFLINK!")
 111           break
 112         fi
 113
 114         for d in /a/bin/distro-setup /p/c; do
 115           cd $d
 116           if [[ ! -e .git ]]; then
 117             # some hosts i dont push all of /p/c
 118             continue
 119           fi
 120           if (( $(date -d "$(git log --diff-filter=ACR --format=%aD -1)" +%s) > fsec )); then
 121             v conflink: newer files checked in to git
 122             chars+=("CONFLINK!")
 123             break
 124           fi
 125
 126           untracked=$(git ls-files -o --exclude-standard)
 127           if [[ $untracked && $(find $untracked -mmin $fminplus -type f -print -quit) ]]; then
 128             v conflink: untracked in $d
 129             chars+=("CONFLINK!")
 130             break
 131           fi
 132         done
 133         cd /
 134
 135       fi
 136       if [[ ! -e $f || $(<$f) != 0 ]]; then
 137         v conflink: last run not found or failed
 138         chars+=("CONFLINK!")
 139         break
 140       fi
 141     done
 142   fi
 143
 144
 145   ## Clean the paniclog, but only up to 4 times per day, or else we
 146   ## should investigate.
 147   loglog=/tmp/panicloglog-$(date --rfc-3339=date)
 148   if [[ -s $loglog ]]; then
 149     spamcount=$(stat -c%s $loglog)
 150   else
 151     spamcount=0
 152   fi
 153   if (( spamcount <= 4 )); then
 154     if grep -q 'spam acl condition' /var/log/exim4/paniclog &>/dev/null; then
 155       printf . >>$loglog
 156     fi
 157     /a/bin/distro-setup/epanic-clean
 158   fi
 159
 160   if [[ -s /var/log/exim4/paniclog ]]; then
 161     chars+=("PANIC!")
 162     tail -n 20 /var/log/exim4/paniclog | lo -1 paniclog
 163   else
 164     lo -1 paniclog
 165   fi
 166
 167   source /a/bin/bash_unpublished/source-state
 168   if [[ $MAIL_HOST == "$HOSTNAME" ]]; then
 169     bbkmsg=
 170     if [[ $(systemctl is-active btrbk.timer) != active ]]; then
 171       chars+=("BTRBK.TIMER!")
 172       bbkmsg="btrbk.timer not enabled"
 173     fi
 174     lo -60 btrbk.timer $bbkmsg
 175
 176     ## check if last snapshot was within an hour
 177     vol=o
 178     # this section generally copied from btrbk scripts, but
 179     # this part modified to speed things up by about half a second.
 180     # I'm not sure if its quite as reliable, but it looks pretty safe.
 181     # Profiled it using time and also adding to the top of the file:
 182     # set -x
 183     # PS4='+ $(date "+%2N") '
 184     snaps=($(ls -1avdr /mnt/root/btrbk/$vol.20*|head -n1 || [[ $? == 141 ]] ))
 185     now=$(date +%s)
 186     maxtime=0
 187     for s in ${snaps[@]}; do
 188       file=${s##*/}
 189       t=$(date -d $(sed -r  's/(.{4})(..)(.{5})(..)(.*)/\1-\2-\3:\4:\5/' <<<${file#$vol.}) +%s)
 190       if (( t > maxtime )); then
 191         maxtime=$t
 192       fi
 193     done
 194     if (( maxtime < now - 60*60 )); then
 195       chars+=("OLD-SNAP!")
 196       snapshotmsg="/o snapshot older than 1 hour"
 197     fi
 198     lo -1 old-snapshot $snapshotmsg
 199   fi
 200
 201   cat /a/bin/bash_unpublished/source-state >$status_file
 202
 203   if [[ ${chars[*]} ]]; then
 204     echo "ps_char=\"${chars[*]} \$ps_char\"" >>$status_file
 205   fi
 206
 207 }
 208 # use this if we want to do something just once per minute
 209 first_chars=()
 210
 211 power=true
 212 if [[ -e /sys/class/power_supply/AC/online && $(</sys/class/power_supply/AC/online) == 0 ]]; then
 213   power=false
 214 fi
 215
 216 write-status
 217 if [[ $1 ]]; then
 218   cat $status_file
 219   exit 0
 220 fi
 221
 222 if ! $power; then
 223   exit 0
 224 fi
 225 for ((i=1; i<=3; i++)); do
 226   sleep 15
 227   write-status
 228 done