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 and have verbose output. On battery power, run
   7 # once per minute.
   8
   9 if [ -z "$BASH_VERSION" ]; then echo "error: shell is not bash" >&2; exit 1; fi
  10
  11 if [[ $EUID != 1000 ]]; then
  12   echo "$0: error, expected to be user 1000"
  13   exit 1
  14 fi
  15
  16 source /a/bin/errhandle/err
  17 status_file=/dev/shm/iank-status
  18
  19 shopt -s nullglob
  20 shopt -s dotglob
  21 shopt -s extglob
  22
  23 for p in ~/.gem/ruby/*/bin; do
  24   PATH="$PATH:$p"
  25 done
  26
  27
  28 verbose=false
  29 if [[ $1 ]]; then
  30   verbose=true
  31 fi
  32 v() {
  33   if $verbose; then
  34     printf "%s\n" "$*"
  35   fi
  36 }
  37 p() { printf "%s\n" "$*"; }
  38 # log-once COUNT NAME [MESSAGE]
  39 lo() {
  40   if type -p ifne &>/dev/null; then
  41     /usr/local/bin/log-once "$@" | ifne mail -s "$HOSTNAME: system-status $2" root@localhost
  42   fi
  43 }
  44
  45 loday() {
  46   if type -p ifne &>/dev/null; then
  47     /usr/local/bin/log-once "$@" | ifne mail -s "$HOSTNAME: system-status $2" daylert@iankelling.org
  48   fi
  49 }
  50 # rm glob
  51 rmg() {
  52   if (( $# )); then
  53     rm -f "$@"
  54   fi
  55 }
  56
  57 # todo, consider migrating some of these alerts into prometheus
  58 write-status() {
  59   chars=("${first_chars[@]}")
  60
  61   services=( epanicclean )
  62   case $HOSTNAME in
  63     bk|je|li) : ;;
  64     *)
  65       services+=(
  66         systemstatus
  67         btrfsmaintstop
  68         dynamicipupdate
  69       )
  70       bads=()
  71       if systemctl show -p SubState --value ${services[@]} | grep -E -v '^(running|)$' &>/dev/null; then
  72         for s in ${services[@]}; do
  73           if [[ $(systemctl show -p SubState --value $s 2>&1) != running ]]; then
  74             bads+=($s)
  75           fi
  76         done
  77         chars+=(MYSERS)
  78       fi
  79       p ${bads[*]} | lo -240 mysers
  80       ;;
  81   esac
  82
  83   case $HOSTNAME in
  84     kd)
  85       services=(
  86         prometheus-node-exporter
  87         prometheus-alertmanager
  88         prometheus
  89       )
  90       bads=()
  91       if systemctl show -p SubState --value ${services[@]} | grep -E -v '^(running|)$' &>/dev/null; then
  92         for s in ${services[@]}; do
  93           if [[ $(systemctl show -p SubState --value $s 2>&1) != running ]]; then
  94             bads+=($s)
  95           fi
  96         done
  97         chars+=(PROM)
  98       fi
  99       p ${bads[*]} | lo -240 prom
 100       ;;
 101   esac
 102
 103
 104   # this section copied from servicepid()
 105   unit=exim4
 106   pid=$(systemctl show --property MainPID --value $unit ||:)
 107   case $pid in
 108     [1-9]*) : ;;
 109     *)
 110       dir=/sys/fs/cgroup/system.slice
 111       if [[ ! -d $dir ]]; then
 112         dir=/sys/fs/cgroup/systemd/system.slice
 113       fi;
 114       pid=$(head -n1 $dir/${unit%.service}.service/cgroup.procs ||:)
 115       ;;
 116   esac
 117   if [[ ! $pid ]]; then
 118     chars+=(EXIM)
 119   fi
 120
 121
 122   if [[ -e /a/bin/bash_unpublished/source-state ]]; then
 123     # /a gets remounted due to btrbk, ignore error code for file doesnt exist
 124     source /a/bin/bash_unpublished/source-state || [[ $? == 1 ]]
 125   fi
 126
 127
 128   ## check if last snapshot was recent
 129   old_snap_limit=$(( 3 * 60 * 60 ))
 130   for vol in a o q; do
 131     case $vol in
 132       o) btrbk_root=/mnt/o/btrbk ;;
 133       *) btrbk_root=/mnt/root/btrbk ;;
 134     esac
 135     # this section generally copied from btrbk scripts, but
 136     # this part modified to speed things up by about half a second.
 137     # I'm not sure if its quite as reliable, but it looks pretty safe.
 138     # Profiled it using time and also adding to the top of the file:
 139     # set -x
 140     # PS4='+ $(date "+%2N") '
 141     # allow failure in case there are no snapshots yet.
 142     # shellcheck disable=SC2012
 143     shopt -s nullglob
 144     files=($btrbk_root/$vol.20*)
 145     shopt -u nullglob
 146     if (( ! ${#files[@]} )); then
 147       continue
 148     fi
 149     snaps=($(ls -1avdr "${files[@]}" 2>/dev/null |head -n1 || : ))
 150     now=$EPOCHSECONDS
 151     maxtime=0
 152     for s in ${snaps[@]}; do
 153       file=${s##*/}
 154       t=$(date -d $(sed -r  's/(.{4})(..)(.{5})(..)(.*)/\1-\2-\3:\4:\5/' <<<${file#$vol.}) +%s)
 155       if (( t > maxtime )); then
 156         maxtime=$t
 157       fi
 158     done
 159     snapshotmsg=
 160     last_snap_age=$(( now - maxtime ))
 161     last_snap_hours=$(( last_snap_age / 60 / 60 ))
 162     if (( last_snap_age > old_snap_limit )); then
 163       chars+=(OLD-SNAP-${last_snap_hours}h)
 164       snapshotmsg="/$vol snapshot older than 4 hours"
 165       if [[ $MAIL_HOST == "$HOSTNAME" ]]; then
 166         p "$snapshotmsg" | lo -1 old-snapshot
 167       fi
 168       # not bothering to get info on all volumes if we find an old one.
 169       break
 170     fi
 171   done
 172
 173
 174   if [[ $MAIL_HOST == "$HOSTNAME" ]]; then
 175
 176     bouncemsg=
 177     glob=(/m/md/bounces/new/*)
 178     if [[ -e ${glob[0]} ]]; then
 179       chars+=(BOUNCE)
 180       bouncemsg="message in /m/md/bounces/new"
 181     fi
 182     p $bouncemsg | loday -1 bounce
 183     # emails without the S (seen) flag. this only checks the last flag,
 184     # but its good enough for me.
 185     glob=(/m/md/alerts/{new,cur}/!(*,S))
 186     if [[ -e ${glob[0]} ]]; then
 187       chars+=(A)
 188     fi
 189
 190     glob=(/m/md/daylert/{new,cur}/!(*,S))
 191     if [[ -e ${glob[0]} ]]; then
 192       chars+=(DAY)
 193     fi
 194
 195     bbkmsg=
 196     if [[ $(systemctl is-active btrbk.timer) != active ]]; then
 197       chars+=(BTRBK.TIMER)
 198       bbkmsg="not enabled"
 199     fi
 200     p "$bbkmsg" | lo -480 btrbk.timer
 201
 202
 203
 204     # commented out, only using timetrap retrospectively.
 205     # # clock us out in timetrap if are idle too long
 206     # if [[ -e /p/.timetrap.db ]]; then
 207     #   export DISPLAY=:0
 208     #   if type -p xprintidle &>/dev/null && xidle=$(xprintidle 2>/dev/null); then
 209     #     if [[ $xidle == [0-9]* ]]; then
 210     #       sheet=$(sqlite3 /p/.timetrap.db "select sheet from entries where end is NULL;")
 211     #       idle=300000
 212     #       if [[ $sheet == w ]]; then
 213     #         idle=900000
 214     #       fi
 215     #       if [[ $sheet && $xidle -gt $idle ]]; then
 216     #         timetrap out
 217     #       fi
 218     #     fi
 219     #   fi
 220     # fi
 221   else # end if $MAIL_HOST
 222     rmg /home/iank/cron-errors/bounce* \
 223         /home/iank/cron-errors/btrbk.timer* \
 224         /home/iank/cron-errors/old-snapshot*
 225   fi
 226
 227   if ip l show tunfsf &>/dev/null; then
 228     # this is for tracking dns over tls issue, which
 229     # fixvpndns() in brc2 fixes.
 230     stat=$(resolvectl dnsovertls tunfsf 2>/dev/null ||: )
 231     read -r _ _ _ istls <<<"$stat"
 232     case $istls in
 233       no) : ;;
 234       *)
 235         printf "%s\n" "$istls" | ts >> /tmp/istls.log
 236         chars+=("T:$istls")
 237         ;;
 238     esac
 239   fi
 240
 241   # We do this once every 5 minutes, since this is not a grave problem.
 242   # For formatted elisp, see /b/ds/unsaved-buffers.el
 243   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)))'
 244   if [[ ! $last_emacs_check || $emacsfiles ]] || (( last_emacs_check < EPOCHSECONDS - 300 )); then
 245     if pgrep -G iank -u iank -f 'emacs --daemon' &>/dev/null; then
 246       # i dun care if this fails
 247       emacsfiles="$(timeout 1 emacsclient -a /usr/bin/true --eval "$elisp" 2>/dev/null | sed '/^"nil"$/d;s/^"(/E: /;s/)"$//' ||:)"
 248       if [[ $emacsfiles ]]; then
 249         chars+=("$emacsfiles")
 250       fi
 251     fi
 252     last_emacs_check=$EPOCHSECONDS
 253   fi
 254
 255
 256   glob=(/nocow/btrfs-stale/*)
 257   if [[ -e ${glob[0]} ]]; then
 258     chars+=(STALE)
 259   fi
 260   var_mail_msg=
 261   if [[ $(find /var/mail -type f \! -empty -print -quit) ]]; then
 262     var_mail_msg="message in /var/mail"
 263   fi
 264   p $var_mail_msg | loday -1 var_mail
 265
 266   # Note, early in install process, we dont have permission yet for exiqgrep.
 267   #
 268   # todo: don't do this every 15 seconds, more like once every 2 minutes to
 269   # save cpu cycles.
 270   #
 271   # 2400 = 40 mins. This should allow for system restarts, and
 272   # 30 minute message delay plus 10 minute queu runs.
 273   qlen=$(/usr/sbin/exiqgrep -o 2400 -c -b | awk '{print $1}') ||:
 274   qmsg=
 275   if ((qlen)); then
 276     # Do sending of long delayed messages, and dont count them in our queue warnings.
 277     for mid in $(exiqgrep -o 2400 -zi); do
 278       if exim -Mvh $mid | awk 'tolower($2) == "fdate:"' | grep -q .; then
 279         qlen=$(( qlen - 1 ))
 280         if (( $(date -d "$(exim -Mset $mid -be <<<'$h_date:' | sed -n 's/^> *//;/./p')" +%s) < EPOCHSECONDS )); then
 281           if ip a show veth0-mail &>/dev/null; then
 282             pid=$(pgrep -f "/usr/sbin/exim4 -bd -q30m -C /etc/exim4/my.conf"|head -n1);
 283             nsenter -t $pid -n -m /usr/sbin/exim4 -C /etc/exim4/my.conf -M $mid
 284           else
 285             /usr/sbin/exim4 -M $mid
 286           fi
 287         fi
 288       fi
 289     done
 290
 291     if ((qlen)); then
 292       qmsg="queue length $qlen"
 293       chars+=("q $qlen")
 294     fi
 295   fi
 296   case $HOSTNAME in
 297     # No point in emailing about the mailq on a host where we don't
 298     # check email.
 299     $MAIL_HOST)
 300       p $qmsg | loday -120 qlen
 301       ;;
 302     *)
 303       rmg /home/iank/cron-errors/qlen*
 304       ;;
 305   esac
 306
 307   begin=false
 308
 309   # todo: make this robust to the case of /a not being mounted
 310   if ! make -C /b/ds -q ~/.local/distro-begin 2>/dev/null || [[ $(<~/.local/distro-begin) != 0 ]]; then
 311     begin=true
 312   fi
 313
 314   end=false
 315   if ! make -C /b/ds -q ~/.local/distro-end 2>/dev/null || [[ $(<~/.local/distro-end) != 0 ]]; then
 316     end=true
 317   fi
 318
 319   # these conditions are so we dont have an overly verbose prompt
 320   if $begin && $end; then
 321     chars+=(D)
 322   elif $begin; then
 323     chars+=(DB)
 324   elif $end; then
 325     chars+=(DE)
 326   else
 327     source /a/bin/ds/script-files
 328     f=~/.local/conflink
 329     # shellcheck disable=SC2043
 330     for _ in 1; do
 331       if [[ -e $f ]]; then
 332         now=$EPOCHSECONDS
 333         fsec=$(stat -c%Y $f)
 334         # the / 60 makes it 0-59 seconds less strict, +1 to help make sure we
 335         # dont have any false positives.
 336         fmin=$(( (fsec - now + 1 ) / 60 ))
 337         fminplus=$(( fmin + 60*24 ))
 338         # Filesystem files get copied, so find any newer than the last run.
 339         # The rest are hueristics:
 340         # Given the last time we added a file in git, is that newer than the last conflink run.
 341         # Given new files not added to git, were they modified more recently than the last conflink? but,
 342         # push their modification time back by a day so we can develop them before needing to add them to git.
 343
 344         all_dirs=({/a/bin/ds,/p/c}{/filesystem,/machine_specific/$HOSTNAME/filesystem})
 345         # This part is copied from conflink
 346         for x in /p/c/machine_specific/*.hosts /a/bin/ds/machine_specific/*.hosts; do
 347           if grep -qxF $HOSTNAME $x; then all_dirs+=( ${x%.hosts} ); fi
 348         done
 349
 350         # Just because i forget a lot, -mmin -NUM means files modified <= NUM minutes ago
 351         if (( fmin < 0 )) && [[ $(find ${all_my_scripts[@]} ${all_dirs[@]} -mmin $fmin -type f -print -quit 2>/dev/null) ]]; then
 352           v conflink newer filesystem files
 353           chars+=(CONFLINK)
 354           break
 355         fi
 356
 357         for d in /a/bin/distro-setup /p/c; do
 358           [[ -d $d ]] || continue
 359           cd $d
 360           if [[ ! -e .git ]]; then
 361             # some hosts i dont push all of /p/c
 362             continue
 363           fi
 364           if (( $(date -d "$(git log --diff-filter=ACR --format=%aD -1)" +%s) > fsec )); then
 365             v conflink: newer files checked in to git
 366             chars+=(CONFLINK)
 367             break
 368           fi
 369
 370           untracked=()
 371           while read -r l; do
 372             untracked+=("$l")
 373           done < <(git ls-files -o --exclude-standard)
 374           if [[ ${untracked[0]} && $(find "${untracked[@]}" -mmin $fminplus -type f -print -quit) ]]; then
 375             v conflink: untracked in $d
 376             chars+=(CONFLINK)
 377             break
 378           fi
 379         done
 380         cd /
 381
 382       fi
 383       if [[ ! -e $f || $(<$f) != 0 ]]; then
 384         v conflink: last run not found or failed
 385         chars+=(CONFLINK)
 386         break
 387       fi
 388     done
 389   fi
 390
 391   #  if [[ $(grep -v "exim user lost privilege for using -C option" /var/log/exim4/paniclog 2>/dev/null ||:) ]]; then
 392   if [[ -s /var/log/exim4/paniclog ]]; then
 393     chars+=("PANIC!")
 394     # leave it up to epanic-clean to send email notification
 395   fi
 396
 397   if [[ ! -e $status_file || -w $status_file ]]; then
 398     if [[ -e /a/bin/bash_unpublished/source-state ]]; then
 399       cat /a/bin/bash_unpublished/source-state >$status_file
 400     fi
 401
 402     if [[ ${chars[*]} ]]; then
 403       echo "ps_char=\"${chars[*]} \$ps_char\"" >>$status_file
 404     fi
 405   fi
 406 }
 407
 408 # This prevents me having to mute notifications when I'm going to bed.
 409 mute() {
 410   local locked
 411   export DISPLAY=:0
 412   locked=false
 413   if lock_info=$(xscreensaver-command -time); then
 414     if [[ $lock_info != *non-blanked* ]]; then
 415       locked=true
 416     fi
 417   else
 418     locked=true
 419   fi
 420   midnight=$(date -d 00:00 +%s)
 421   mdiff=$(( EPOCHSECONDS - midnight ))
 422   if $locked && (( mdiff < 6 || mdiff > 21 )); then
 423     case $(pactl get-sink-mute @DEFAULT_SINK@ | awk '{print $2}') in
 424       no)
 425         # for log purposes
 426         echo unmuted
 427         pactl set-sink-mute @DEFAULT_SINK@ true
 428         ;;
 429     esac
 430   fi
 431   if ! $locked && ((  mdiff > 6 || mdiff < 12  )) && [[ ! -e /tmp/ianknap ]]; then
 432     case $(pactl get-sink-mute @DEFAULT_SINK@ | awk '{print $2}') in
 433       yes)
 434         # for log purposes
 435         echo muted
 436         pactl set-sink-mute @DEFAULT_SINK@ false
 437         ;;
 438     esac
 439   fi
 440 }
 441
 442 # use this if we want to do something just once per minute
 443 first_chars=()
 444
 445 write-status
 446 if [[ $1 ]]; then
 447   cat $status_file
 448   exit 0
 449 fi
 450
 451 main-loop() {
 452   while true; do
 453     power=true
 454     if [[ -e /sys/class/power_supply/AC/online && $(</sys/class/power_supply/AC/online) == 0 ]]; then
 455       power=false
 456     fi
 457     wait=15
 458     if ! $power; then
 459       if systemctl -q is-active bitcoind; then
 460         bitcoinoff
 461       fi
 462       wait=60
 463     fi
 464
 465     sleep $wait
 466     write-status
 467     mute
 468   done
 469 }
 470
 471 # ensure our long operations are one line so we are not prone errors
 472 # from this file being modified.
 473 main-loop