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   glob=(/m/md/bounces/new/*)
  35   if [[ -e ${glob[0]} ]]; then
  36     chars+=("BOUNCE!")
  37     bouncemsg="message in /m/md/bounces/new"
  38   fi
  39   lo -1 bounce $bouncemsg
  40   glob=(/m/md/alerts/new/* /m/md/alerts/cur/*)
  41   if [[ -e ${glob[0]} ]]; then
  42     chars+=("ALERT!")
  43   fi
  44   if [[ -e /nocow/user/mailtest-failure ]]; then
  45     chars+=("MAILPING!")
  46   fi
  47
  48   if ! qlen=$(/usr/sbin/exiqgrep -o 60 -c -b | awk '{print $1}'); then
  49     # early in install process, we dont have permission yet for exiqgrep
  50     qlen=$(sudo /usr/sbin/exiqgrep -o 60 -c -b | awk '{print $1}')
  51   fi
  52   if ((qlen)); then
  53     chars+=("q $qlen")
  54   fi
  55
  56   cd /b/ds
  57   if ! make -q ~/.local/distro-begin || [[ $(<~/.local/distro-begin) != 0 ]]; then
  58     chars+=("DISTRO-BEGIN!")
  59   fi
  60
  61   if ! make -q ~/.local/distro-end || [[ $(<~/.local/distro-end) != 0 ]]; then
  62     chars+=("DISTRO-END!")
  63   fi
  64
  65   f=~/.local/conflink
  66   for _ in 1; do
  67     if [[ -e $f ]]; then
  68       now=$(date +%s)
  69       fsec=$(stat -c%Y $f)
  70       # the / 60 makes it 0-59 seconds less strict, +1 makes sure we
  71       # dont have any false positives.
  72       fmin=$(( (fsec - now ) / 60 + 1 ))
  73       fminplus=$(( fmin + 60*24 ))
  74       # Filesystem files get copied, so find any newer than the last run.
  75       # The rest are hueristics:
  76       # Given the last time we added a file in git, is that newer than the last conflink run.
  77       # Given new files not added to git, were they modified more recently than the last conflink? but,
  78       # push their modification time back by a day so we can develop them before needing to add them to git.
  79
  80       all_dirs=({/a/bin/ds,/p/c}{/filesystem,/machine_specific/$HOSTNAME/filesystem})
  81       # This part is copied from conflink
  82       for x in /p/c/machine_specific/*.hosts /a/bin/ds/machine_specific/*.hosts; do
  83         if grep -qxF $HOSTNAME $x; then all_dirs+=( ${x%.hosts} ); fi
  84       done
  85
  86       # just because i forget a lot, -mmin -NUM means files modified <= NUM minutes ago
  87       if (( $(date -d "$(git log --diff-filter=ACR --format=%aD -1)" +%s) > fsec )) || \
  88            [[ $(find ${all_dirs[@]} -mmin $fmin -type f -print -quit 2>/dev/null) ]]; then
  89         v conflink newer git or newer filesystem files
  90         chars+=("CONFLINK!")
  91         break
  92       fi
  93
  94       for d in /a/bin/distro-setup /p/c; do
  95         cd $d
  96         untracked=$(git ls-files -o --exclude-standard)
  97         if [[ $untracked && $(find $untracked -mmin $fminplus -type f -print -quit) ]]; then
  98           v conflink: untracked in $d
  99           chars+=("CONFLINK!")
 100           break
 101         fi
 102       done
 103
 104     fi
 105     if [[ ! -e $f || $(<$f) != 0 ]]; then
 106       v conflink: last run not found or failed
 107       chars+=("CONFLINK!")
 108       break
 109     fi
 110   done
 111
 112
 113   ## Clean the paniclog, but only up to 4 times per day, or else we
 114   ## should investigate.
 115   loglog=/tmp/panicloglog-$(date --rfc-3339=date)
 116   if [[ -s $loglog ]]; then
 117     spamcount=$(stat -c%s $loglog)
 118   else
 119     spamcount=0
 120   fi
 121   if (( spamcount <= 4 )); then
 122     if grep -q 'spam acl condition' /var/log/exim4/paniclog &>/dev/null; then
 123       printf . >>$loglog
 124     fi
 125     /a/bin/distro-setup/epanic-clean
 126   fi
 127
 128   if [[ -s /var/log/exim4/paniclog ]]; then
 129     chars+=("PANIC!")
 130     tail -n 20 /var/log/exim4/paniclog | lo -1 paniclog
 131   else
 132     lo -1 paniclog
 133   fi
 134
 135   source /a/bin/bash_unpublished/source-state
 136   if [[ $MAIL_HOST == "$HOSTNAME" ]]; then
 137     bbkmsg=
 138     if [[ $(systemctl is-active btrbk.timer) != active ]]; then
 139       chars+=("BTRBK.TIMER!")
 140       bbkmsg="btrbk.timer not enabled"
 141     fi
 142     lo -60 btrbk.timer $bbkmsg
 143
 144     ## check if last snapshot was within an hour
 145     vol=o
 146     # this section generally copied from btrbk scripts, but
 147     # this part modified to speed things up by about half a second.
 148     # I'm not sure if its quite as reliable, but it looks pretty safe.
 149     # Profiled it using time and also adding to the top of the file:
 150     # set -x
 151     # PS4='+ $(date "+%2N") '
 152     snaps=($(ls -1avdr /mnt/root/btrbk/$vol.20*|head -n1))
 153     now=$(date +%s)
 154     maxtime=0
 155     for s in ${snaps[@]}; do
 156       file=${s##*/}
 157       t=$(date -d $(sed -r  's/(.{4})(..)(.{5})(..)(.*)/\1-\2-\3:\4:\5/' <<<${file#$vol.}) +%s)
 158       if (( t > maxtime )); then
 159         maxtime=$t
 160       fi
 161     done
 162     if (( maxtime < now - 60*60 )); then
 163       chars+=("OLD-SNAPSHOT!")
 164       snapshotmsg="/o snapshot older than 1 hour"
 165     fi
 166     lo -1 old-snapshot $snapshotmsg
 167   fi
 168
 169   cat /a/bin/bash_unpublished/source-state >$status_file
 170
 171   if [[ ${chars[*]} ]]; then
 172     echo "ps_char=\"${chars[*]} \$ps_char\"" >>$status_file
 173   fi
 174
 175 }
 176 # use this if we want to do something just once per minute
 177 first_chars=()
 178
 179 power=true
 180 if [[ -e /sys/class/power_supply/AC/online && $(</sys/class/power_supply/AC/online) == 0 ]]; then
 181   power=false
 182 fi
 183
 184 write-status
 185 if [[ $1 ]]; then
 186   cat $status_file
 187   exit 0
 188 fi
 189
 190 if ! $power; then
 191   exit 0
 192 fi
 193 for ((i=1; i<=3; i++)); do
 194   sleep 15
 195   write-status
 196 done