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 once every 15 seconds unless any args are passed, or we
   6 # then just runs once. On battery power, run once per minute.
   7
   8 if [ -z "$BASH_VERSION" ]; then echo "error: shell is not bash" >&2; exit 1; fi
   9
  10 source /a/bin/errhandle/err
  11 status_file=/dev/shm/iank-status
  12
  13 shopt -s nullglob
  14 shopt -s dotglob
  15 shopt -s extglob
  16
  17 for p in ~/.gem/ruby/*/bin; do
  18   PATH="$PATH:$p"
  19 done
  20
  21
  22 verbose=false
  23 if [[ $1 ]]; then
  24   verbose=true
  25 fi
  26 v() {
  27   if $verbose; then
  28     printf "%s\n" "$*"
  29   fi
  30 }
  31 # log-once COUNT NAME [MESSAGE]
  32 lo() {
  33   /usr/local/bin/log-once "$@" | ifne mail -s "$HOSTNAME: system-status $2" root@localhost
  34 }
  35
  36 loday() {
  37   /usr/local/bin/log-once "$@" | ifne mail -s "$HOSTNAME: system-status $2" daylerts@iankelling.org
  38 }
  39
  40 # todo, consider migrating some of these alerts into prometheus
  41 write-status() {
  42   chars=("${first_chars[@]}")
  43
  44   services=(
  45     epanicclean
  46     systemstatus
  47     btrfsmaintstop
  48     dynamicipupdate
  49   )
  50   bads=()
  51   if systemctl show -p SubState --value ${services[@]} | egrep -v '^(running|)$'; then
  52     for s in ${services[@]}; do
  53       if [[ $(systemctl show -p SubState --value $s) != running ]]; then
  54         bads+=($s)
  55       fi
  56     done
  57     chars+=(MYSERS)
  58
  59   fi
  60   lo -240 mysers ${bads[*]}
  61
  62   services=(
  63     prometheus-node-exporter
  64     prometheus-alertmanager
  65     prometheus
  66   )
  67   case $HOSTNAME in
  68     kd)
  69       bads=()
  70       if systemctl show -p SubState --value ${services[@]} | egrep -v '^(running|)$'; then
  71         for s in ${services[@]}; do
  72           if [[ $(systemctl show -p SubState --value $s) != running ]]; then
  73             bads+=($s)
  74           fi
  75         done
  76         chars+=(PROM)
  77       fi
  78       lo -240 prom ${bads[*]}
  79       ;;
  80   esac
  81
  82   # clock us out in timetrap if are idle too long
  83   if [[ -e /p/.timetrap.db ]]; then
  84     export DISPLAY=:0
  85     if type -p xprintidle &>/dev/null && xidle=$(xprintidle 2>/dev/null); then
  86       if [[ $xidle == [0-9]* ]]; then
  87         sheet=$(sqlite3 /p/.timetrap.db "select sheet from entries where end is NULL;")
  88         idle=300000
  89         if [[ $sheet == w ]]; then
  90           idle=900000
  91         fi
  92         if [[ $sheet && $xidle -gt $idle ]]; then
  93           timetrap out
  94         fi
  95       fi
  96     fi
  97   fi
  98
  99
 100   if ip l show tunfsf &>/dev/null; then
 101     # this is for tracking dns over tls issue, which
 102     # fixvpndns() in brc2 fixes.
 103     stat=$(resolvectl dnsovertls tunfsf 2>/dev/null ||: )
 104     read _ _ _ istls <<<"$stat"
 105     case $istls in
 106       no) : ;;
 107       *)
 108         printf "%s\n" "$istls" | ts >> /tmp/istls.log
 109         chars+=("T:$istls")
 110         ;;
 111     esac
 112   fi
 113
 114
 115   if pgrep -G iank -u iank -f 'emacs --daemon' &>/dev/null; then
 116     emacsfiles="$(emacsclient --eval "$(cat /usr/local/bin/unsaved-buffers.el)"| sed '/^"nil"$/d;s/^"(/E: /;s/)"$//')"
 117     if [[ $emacsfiles ]]; then
 118       chars+=("$emacsfiles")
 119     fi
 120   fi
 121
 122   glob=(/nocow/btrfs-stale/*)
 123   if [[ -e ${glob[0]} ]]; then
 124     chars+=(STALE)
 125   fi
 126   var_mail_msg=
 127   if [[ $(find /var/mail -type f \! -empty -print -quit) ]]; then
 128     var_mail_msg="message in /var/mail"
 129   fi
 130   loday -1 var_mail $var_mail_msg
 131
 132   bouncemsg=
 133   glob=(/m/md/bounces/new/*)
 134   if [[ -e ${glob[0]} ]]; then
 135     chars+=(BOUNCE)
 136     bouncemsg="message in /m/md/bounces/new"
 137   fi
 138   loday -1 bounce $bouncemsg
 139   # emails without the S (seen) flag. this only checks the last flag,
 140   # but its good enough for me.
 141   glob=(/m/md/alerts/{new,cur}/!(*,S))
 142   if [[ -e ${glob[0]} ]]; then
 143     chars+=(A)
 144   fi
 145
 146   glob=(/m/md/daylerts/{new,cur}/!(*,S))
 147   if [[ -e ${glob[0]} ]]; then
 148     chars+=(DAY)
 149   fi
 150
 151
 152   tmp=(/var/local/cron-errors/mailtest-check*)
 153   if (( ${#tmp[@]} )); then
 154     chars+=(MAILPING)
 155   fi
 156   tmp=(/var/local/cron-errors/mailtest-slow*)
 157   if (( ${#tmp[@]} )); then
 158     chars+=(SPAMD)
 159   fi
 160
 161   # early in install process, we dont have permission yet for exiqgrep.
 162   # 1100 helps allow for system restarts
 163   qlen=$(/usr/sbin/exiqgrep -o 1100 -c -b | awk '{print $1}') ||:
 164   qmsg=
 165   if ((qlen)); then
 166     qmsg="queue length $qlen"
 167     chars+=("q $qlen")
 168   fi
 169   case $HOSTNAME in
 170     # No point in emailing about the mailq on a host where we don't
 171     # check email.
 172     $MAIL_HOST|bk)
 173       loday -120 qlen $qmsg
 174       ;;
 175   esac
 176
 177   begin=false
 178   if ! make -C /b/ds -q ~/.local/distro-begin || [[ $(<~/.local/distro-begin) != 0 ]]; then
 179     begin=true
 180   fi
 181
 182   end=false
 183   if ! make -C /b/ds -q ~/.local/distro-end || [[ $(<~/.local/distro-end) != 0 ]]; then
 184     end=true
 185   fi
 186
 187   # these conditions are so we dont have an overly verbose prompt
 188   if $begin && $end; then
 189     chars+=(D)
 190   elif $begin; then
 191     chars+=(DB)
 192   elif $end; then
 193     chars+=(DE)
 194   else
 195     f=~/.local/conflink
 196     # shellcheck disable=SC2043
 197     for _ in 1; do
 198       if [[ -e $f ]]; then
 199         now=$(date +%s)
 200         fsec=$(stat -c%Y $f)
 201         # the / 60 makes it 0-59 seconds less strict, +1 to help make sure we
 202         # dont have any false positives.
 203         fmin=$(( (fsec - now + 1 ) / 60 ))
 204         fminplus=$(( fmin + 60*24 ))
 205         # Filesystem files get copied, so find any newer than the last run.
 206         # The rest are hueristics:
 207         # Given the last time we added a file in git, is that newer than the last conflink run.
 208         # Given new files not added to git, were they modified more recently than the last conflink? but,
 209         # push their modification time back by a day so we can develop them before needing to add them to git.
 210
 211         all_dirs=({/a/bin/ds,/p/c}{/filesystem,/machine_specific/$HOSTNAME/filesystem})
 212         # This part is copied from conflink
 213         for x in /p/c/machine_specific/*.hosts /a/bin/ds/machine_specific/*.hosts; do
 214           if grep -qxF $HOSTNAME $x; then all_dirs+=( ${x%.hosts} ); fi
 215         done
 216
 217         # Just because i forget a lot, -mmin -NUM means files modified <= NUM minutes ago
 218         if (( fmin < 0 )) && [[ $(find ${all_dirs[@]} -mmin $fmin -type f -print -quit 2>/dev/null) ]]; then
 219           v conflink newer filesystem files
 220           chars+=(CONFLINK)
 221           break
 222         fi
 223
 224         for d in /a/bin/distro-setup /p/c; do
 225           [[ -d $d ]] || continue
 226           cd $d
 227           if [[ ! -e .git ]]; then
 228             # some hosts i dont push all of /p/c
 229             continue
 230           fi
 231           if (( $(date -d "$(git log --diff-filter=ACR --format=%aD -1)" +%s) > fsec )); then
 232             v conflink: newer files checked in to git
 233             chars+=(CONFLINK)
 234             break
 235           fi
 236
 237           untracked=()
 238           while read -r l; do
 239             untracked+=("$l")
 240           done < <(git ls-files -o --exclude-standard)
 241           if [[ ${untracked[0]} && $(find "${untracked[@]}" -mmin $fminplus -type f -print -quit) ]]; then
 242             v conflink: untracked in $d
 243             chars+=(CONFLINK)
 244             break
 245           fi
 246         done
 247         cd /
 248
 249       fi
 250       if [[ ! -e $f || $(<$f) != 0 ]]; then
 251         v conflink: last run not found or failed
 252         chars+=(CONFLINK)
 253         break
 254       fi
 255     done
 256   fi
 257
 258   #  if [[ $(grep -v "exim user lost privilege for using -C option" /var/log/exim4/paniclog 2>/dev/null ||:) ]]; then
 259   if [[ -s /var/log/exim4/paniclog ]]; then
 260     chars+=("PANIC!")
 261     # leave it up to epanic-clean to send email notification
 262   fi
 263
 264   source /a/bin/bash_unpublished/source-state
 265   if [[ $MAIL_HOST == "$HOSTNAME" ]]; then
 266     bbkmsg=
 267     if [[ $(systemctl is-active btrbk.timer) != active ]]; then
 268       chars+=(BTRBK.TIMER)
 269       bbkmsg="not enabled"
 270     fi
 271     lo -480 btrbk.timer $bbkmsg
 272
 273     ## check if last snapshot was within an hour
 274     vol=o
 275     # this section generally copied from btrbk scripts, but
 276     # this part modified to speed things up by about half a second.
 277     # I'm not sure if its quite as reliable, but it looks pretty safe.
 278     # Profiled it using time and also adding to the top of the file:
 279     # set -x
 280     # PS4='+ $(date "+%2N") '
 281     # allow failure in case there are no snapshots yet.
 282     # shellcheck disable=SC2012
 283     shopt -u nullglob
 284     files=(/mnt/root/btrbk/$vol.20*)
 285     shopt -s nullglob
 286     snaps=()
 287     if (( ${#files[@]} )); then
 288       snaps=($(ls -1avdr "${files[@]}" 2>/dev/null |head -n1 || : ))
 289     fi
 290     now=$(date +%s)
 291     maxtime=0
 292     for s in ${snaps[@]}; do
 293       file=${s##*/}
 294       t=$(date -d $(sed -r  's/(.{4})(..)(.{5})(..)(.*)/\1-\2-\3:\4:\5/' <<<${file#$vol.}) +%s)
 295       if (( t > maxtime )); then
 296         maxtime=$t
 297       fi
 298     done
 299     snapshotmsg=
 300     if (( maxtime < now - 4*60*60 )); then
 301       chars+=(OLD-SNAP)
 302       snapshotmsg="/o snapshot older than 4 hours"
 303     fi
 304     lo -1 old-snapshot $snapshotmsg
 305   fi
 306
 307   cat /a/bin/bash_unpublished/source-state >$status_file
 308
 309   if [[ ${chars[*]} ]]; then
 310     echo "ps_char=\"${chars[*]} \$ps_char\"" >>$status_file
 311   fi
 312
 313 }
 314 # use this if we want to do something just once per minute
 315 first_chars=()
 316
 317
 318 write-status
 319 if [[ $1 ]]; then
 320   cat $status_file
 321   exit 0
 322 fi
 323
 324 main-loop() {
 325   while true; do
 326     power=true
 327     if [[ -e /sys/class/power_supply/AC/online && $(</sys/class/power_supply/AC/online) == 0 ]]; then
 328       power=false
 329     fi
 330     wait=15
 331     if ! $power; then
 332       wait=60
 333     fi
 334
 335     sleep $wait
 336     write-status
 337   done
 338 }
 339
 340 # ensure our long operations are one line so we are not prone errors
 341 # from this file being modified.
 342 main-loop