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 if [[ $EUID != 1000 ]]; then
  11   echo "$0: error, expected to be user 1000"
  12   exit 1
  13 fi
  14
  15 source /a/bin/errhandle/err
  16 status_file=/dev/shm/iank-status
  17
  18 shopt -s nullglob
  19 shopt -s dotglob
  20 shopt -s extglob
  21
  22 for p in ~/.gem/ruby/*/bin; do
  23   PATH="$PATH:$p"
  24 done
  25
  26
  27 verbose=false
  28 if [[ $1 ]]; then
  29   verbose=true
  30 fi
  31 v() {
  32   if $verbose; then
  33     printf "%s\n" "$*"
  34   fi
  35 }
  36 p() { printf "%s\n" "$*"; }
  37 # log-once COUNT NAME [MESSAGE]
  38 lo() {
  39   if type -p ifne &>/dev/null; then
  40     /usr/local/bin/log-once "$@" | ifne mail -s "$HOSTNAME: system-status $2" root@localhost
  41   fi
  42 }
  43
  44 loday() {
  45   if type -p ifne &>/dev/null; then
  46     /usr/local/bin/log-once "$@" | ifne mail -s "$HOSTNAME: system-status $2" daylert@iankelling.org
  47   fi
  48 }
  49
  50 # todo, consider migrating some of these alerts into prometheus
  51 write-status() {
  52   chars=("${first_chars[@]}")
  53
  54   services=( epanicclean )
  55   case $HOSTNAME in
  56     bk|je|li) : ;;
  57     *)
  58       services+=(
  59         systemstatus
  60         btrfsmaintstop
  61         dynamicipupdate
  62       )
  63       bads=()
  64       if systemctl show -p SubState --value ${services[@]} | egrep -v '^(running|)$' &>/dev/null; then
  65         for s in ${services[@]}; do
  66           if [[ $(systemctl show -p SubState --value $s 2>&1) != running ]]; then
  67             bads+=($s)
  68           fi
  69         done
  70         chars+=(MYSERS)
  71       fi
  72       p ${bads[*]} | lo -240 mysers
  73       ;;
  74   esac
  75
  76   case $HOSTNAME in
  77     kd)
  78       services=(
  79         prometheus-node-exporter
  80         prometheus-alertmanager
  81         prometheus
  82       )
  83       bads=()
  84       if systemctl show -p SubState --value ${services[@]} | egrep -v '^(running|)$' &>/dev/null; then
  85         for s in ${services[@]}; do
  86           if [[ $(systemctl show -p SubState --value $s 2>&1) != running ]]; then
  87             bads+=($s)
  88           fi
  89         done
  90         chars+=(PROM)
  91       fi
  92       p ${bads[*]} | lo -240 prom
  93       ;;
  94   esac
  95
  96
  97   if [[ -e /a/bin/bash_unpublished/source-state ]]; then
  98     # /a gets remounted due to btrbk, ignore error code for file doesnt exist
  99     source /a/bin/bash_unpublished/source-state || [[ $? == 1 ]]
 100   fi
 101   if [[ $MAIL_HOST == "$HOSTNAME" ]]; then
 102
 103     bouncemsg=
 104     glob=(/m/md/bounces/new/*)
 105     if [[ -e ${glob[0]} ]]; then
 106       chars+=(BOUNCE)
 107       bouncemsg="message in /m/md/bounces/new"
 108     fi
 109     p $bouncemsg | loday -1 bounce
 110     # emails without the S (seen) flag. this only checks the last flag,
 111     # but its good enough for me.
 112     glob=(/m/md/alerts/{new,cur}/!(*,S))
 113     if [[ -e ${glob[0]} ]]; then
 114       chars+=(A)
 115     fi
 116
 117     glob=(/m/md/daylert/{new,cur}/!(*,S))
 118     if [[ -e ${glob[0]} ]]; then
 119       chars+=(DAY)
 120     fi
 121
 122     bbkmsg=
 123     if [[ $(systemctl is-active btrbk.timer) != active ]]; then
 124       chars+=(BTRBK.TIMER)
 125       bbkmsg="not enabled"
 126     fi
 127     p "$bbkmsg" | lo -480 btrbk.timer
 128
 129     ## check if last snapshot was within an hour
 130     vol=o
 131     # this section generally copied from btrbk scripts, but
 132     # this part modified to speed things up by about half a second.
 133     # I'm not sure if its quite as reliable, but it looks pretty safe.
 134     # Profiled it using time and also adding to the top of the file:
 135     # set -x
 136     # PS4='+ $(date "+%2N") '
 137     # allow failure in case there are no snapshots yet.
 138     # shellcheck disable=SC2012
 139     shopt -u nullglob
 140     files=(/mnt/root/btrbk/$vol.20*)
 141     shopt -s nullglob
 142     snaps=()
 143     if (( ${#files[@]} )); then
 144       snaps=($(ls -1avdr "${files[@]}" 2>/dev/null |head -n1 || : ))
 145     fi
 146     now=$EPOCHSECONDS
 147     maxtime=0
 148     for s in ${snaps[@]}; do
 149       file=${s##*/}
 150       t=$(date -d $(sed -r  's/(.{4})(..)(.{5})(..)(.*)/\1-\2-\3:\4:\5/' <<<${file#$vol.}) +%s)
 151       if (( t > maxtime )); then
 152         maxtime=$t
 153       fi
 154     done
 155     snapshotmsg=
 156     if (( maxtime < now - 4*60*60 )); then
 157       chars+=(OLD-SNAP)
 158       snapshotmsg="/o snapshot older than 4 hours"
 159     fi
 160     p "$snapshotmsg" | lo -1 old-snapshot
 161
 162
 163     # commented out, only using timetrap retrospectively.
 164     # # clock us out in timetrap if are idle too long
 165     # if [[ -e /p/.timetrap.db ]]; then
 166     #   export DISPLAY=:0
 167     #   if type -p xprintidle &>/dev/null && xidle=$(xprintidle 2>/dev/null); then
 168     #     if [[ $xidle == [0-9]* ]]; then
 169     #       sheet=$(sqlite3 /p/.timetrap.db "select sheet from entries where end is NULL;")
 170     #       idle=300000
 171     #       if [[ $sheet == w ]]; then
 172     #         idle=900000
 173     #       fi
 174     #       if [[ $sheet && $xidle -gt $idle ]]; then
 175     #         timetrap out
 176     #       fi
 177     #     fi
 178     #   fi
 179     # fi
 180
 181   fi
 182
 183   if ip l show tunfsf &>/dev/null; then
 184     # this is for tracking dns over tls issue, which
 185     # fixvpndns() in brc2 fixes.
 186     stat=$(resolvectl dnsovertls tunfsf 2>/dev/null ||: )
 187     read _ _ _ istls <<<"$stat"
 188     case $istls in
 189       no) : ;;
 190       *)
 191         printf "%s\n" "$istls" | ts >> /tmp/istls.log
 192         chars+=("T:$istls")
 193         ;;
 194     esac
 195   fi
 196
 197   # We do this once every 5 minutes, since this is not a grave problem.
 198   # For formatted elisp, see /b/ds/unsaved-buffers.el
 199   elisp='(format "%s" (-reduce-from (lambda (acc buf) (let ((bpath (buffer-file-name buf))) (if (and bpath (buffer-modified-p buf)) (cons bpath acc ) acc))) nil (buffer-list)))'
 200   if [[ ! $last_emacs_check || $emacsfiles ]] || (( last_emacs_check < EPOCHSECONDS - 300 )); then
 201     if pgrep -G iank -u iank -f 'emacs --daemon' &>/dev/null; then
 202       # i dun care if this fails
 203       emacsfiles="$(timeout 1 emacsclient --eval "$elisp"| sed '/^"nil"$/d;s/^"(/E: /;s/)"$//' ||:)"
 204       if [[ $emacsfiles ]]; then
 205         chars+=("$emacsfiles")
 206       fi
 207     fi
 208     last_emacs_check=$EPOCHSECONDS
 209   fi
 210
 211
 212   glob=(/nocow/btrfs-stale/*)
 213   if [[ -e ${glob[0]} ]]; then
 214     chars+=(STALE)
 215   fi
 216   var_mail_msg=
 217   if [[ $(find /var/mail -type f \! -empty -print -quit) ]]; then
 218     var_mail_msg="message in /var/mail"
 219   fi
 220   p $var_mail_msg | loday -1 var_mail
 221
 222
 223   tmp=(/var/local/cron-errors/mailtest-check*)
 224   if (( ${#tmp[@]} )); then
 225     chars+=(MAILPING)
 226   fi
 227   tmp=(/var/local/cron-errors/mailtest-slow*)
 228   if (( ${#tmp[@]} )); then
 229     chars+=(SPAMD)
 230   fi
 231
 232   # early in install process, we dont have permission yet for exiqgrep.
 233   # 1100 helps allow for system restarts
 234   qlen=$(/usr/sbin/exiqgrep -o 1100 -c -b | awk '{print $1}') ||:
 235   qmsg=
 236   if ((qlen)); then
 237     qmsg="queue length $qlen"
 238     chars+=("q $qlen")
 239   fi
 240   case $HOSTNAME in
 241     # No point in emailing about the mailq on a host where we don't
 242     # check email.
 243     $MAIL_HOST)
 244       p $qmsg | loday -120 qlen
 245       ;;
 246   esac
 247
 248   begin=false
 249
 250   if ! make -C /b/ds -q ~/.local/distro-begin 2>/dev/null || [[ $(<~/.local/distro-begin) != 0 ]]; then
 251     begin=true
 252   fi
 253
 254   end=false
 255   if ! make -C /b/ds -q ~/.local/distro-end 2>/dev/null || [[ $(<~/.local/distro-end) != 0 ]]; then
 256     end=true
 257   fi
 258
 259   # these conditions are so we dont have an overly verbose prompt
 260   if $begin && $end; then
 261     chars+=(D)
 262   elif $begin; then
 263     chars+=(DB)
 264   elif $end; then
 265     chars+=(DE)
 266   else
 267     f=~/.local/conflink
 268     # shellcheck disable=SC2043
 269     for _ in 1; do
 270       if [[ -e $f ]]; then
 271         now=$EPOCHSECONDS
 272         fsec=$(stat -c%Y $f)
 273         # the / 60 makes it 0-59 seconds less strict, +1 to help make sure we
 274         # dont have any false positives.
 275         fmin=$(( (fsec - now + 1 ) / 60 ))
 276         fminplus=$(( fmin + 60*24 ))
 277         # Filesystem files get copied, so find any newer than the last run.
 278         # The rest are hueristics:
 279         # Given the last time we added a file in git, is that newer than the last conflink run.
 280         # Given new files not added to git, were they modified more recently than the last conflink? but,
 281         # push their modification time back by a day so we can develop them before needing to add them to git.
 282
 283         all_dirs=({/a/bin/ds,/p/c}{/filesystem,/machine_specific/$HOSTNAME/filesystem})
 284         # This part is copied from conflink
 285         for x in /p/c/machine_specific/*.hosts /a/bin/ds/machine_specific/*.hosts; do
 286           if grep -qxF $HOSTNAME $x; then all_dirs+=( ${x%.hosts} ); fi
 287         done
 288
 289         # Just because i forget a lot, -mmin -NUM means files modified <= NUM minutes ago
 290         if (( fmin < 0 )) && [[ $(find ${all_dirs[@]} -mmin $fmin -type f -print -quit 2>/dev/null) ]]; then
 291           v conflink newer filesystem files
 292           chars+=(CONFLINK)
 293           break
 294         fi
 295
 296         for d in /a/bin/distro-setup /p/c; do
 297           [[ -d $d ]] || continue
 298           cd $d
 299           if [[ ! -e .git ]]; then
 300             # some hosts i dont push all of /p/c
 301             continue
 302           fi
 303           if (( $(date -d "$(git log --diff-filter=ACR --format=%aD -1)" +%s) > fsec )); then
 304             v conflink: newer files checked in to git
 305             chars+=(CONFLINK)
 306             break
 307           fi
 308
 309           untracked=()
 310           while read -r l; do
 311             untracked+=("$l")
 312           done < <(git ls-files -o --exclude-standard)
 313           if [[ ${untracked[0]} && $(find "${untracked[@]}" -mmin $fminplus -type f -print -quit) ]]; then
 314             v conflink: untracked in $d
 315             chars+=(CONFLINK)
 316             break
 317           fi
 318         done
 319         cd /
 320
 321       fi
 322       if [[ ! -e $f || $(<$f) != 0 ]]; then
 323         v conflink: last run not found or failed
 324         chars+=(CONFLINK)
 325         break
 326       fi
 327     done
 328   fi
 329
 330   #  if [[ $(grep -v "exim user lost privilege for using -C option" /var/log/exim4/paniclog 2>/dev/null ||:) ]]; then
 331   if [[ -s /var/log/exim4/paniclog ]]; then
 332     chars+=("PANIC!")
 333     # leave it up to epanic-clean to send email notification
 334   fi
 335
 336   if [[ ! -e $status_file || -w $status_file ]]; then
 337     if [[ -e /a/bin/bash_unpublished/source-state ]]; then
 338       cat /a/bin/bash_unpublished/source-state >$status_file
 339     fi
 340
 341     if [[ ${chars[*]} ]]; then
 342       echo "ps_char=\"${chars[*]} \$ps_char\"" >>$status_file
 343     fi
 344   fi
 345 }
 346 # use this if we want to do something just once per minute
 347 first_chars=()
 348
 349
 350 write-status
 351 if [[ $1 ]]; then
 352   cat $status_file
 353   exit 0
 354 fi
 355
 356 main-loop() {
 357   while true; do
 358     power=true
 359     if [[ -e /sys/class/power_supply/AC/online && $(</sys/class/power_supply/AC/online) == 0 ]]; then
 360       power=false
 361     fi
 362     wait=15
 363     if ! $power; then
 364       wait=60
 365     fi
 366
 367     sleep $wait
 368     write-status
 369   done
 370 }
 371
 372 # ensure our long operations are one line so we are not prone errors
 373 # from this file being modified.
 374 main-loop