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 ${services[@]} | egrep -v '^(running|)$' &>/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 ${services[@]} | egrep -v '^(running|)$' &>/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   if [[ $MAIL_HOST == "$HOSTNAME" ]]; then
 127
 128     bouncemsg=
 129     glob=(/m/md/bounces/new/*)
 130     if [[ -e ${glob[0]} ]]; then
 131       chars+=(BOUNCE)
 132       bouncemsg="message in /m/md/bounces/new"
 133     fi
 134     p $bouncemsg | loday -1 bounce
 135     # emails without the S (seen) flag. this only checks the last flag,
 136     # but its good enough for me.
 137     glob=(/m/md/alerts/{new,cur}/!(*,S))
 138     if [[ -e ${glob[0]} ]]; then
 139       chars+=(A)
 140     fi
 141
 142     glob=(/m/md/daylert/{new,cur}/!(*,S))
 143     if [[ -e ${glob[0]} ]]; then
 144       chars+=(DAY)
 145     fi
 146
 147     bbkmsg=
 148     if [[ $(systemctl is-active btrbk.timer) != active ]]; then
 149       chars+=(BTRBK.TIMER)
 150       bbkmsg="not enabled"
 151     fi
 152     p "$bbkmsg" | lo -480 btrbk.timer
 153
 154     ## check if last snapshot was within an hour
 155     vol=o
 156     # this section generally copied from btrbk scripts, but
 157     # this part modified to speed things up by about half a second.
 158     # I'm not sure if its quite as reliable, but it looks pretty safe.
 159     # Profiled it using time and also adding to the top of the file:
 160     # set -x
 161     # PS4='+ $(date "+%2N") '
 162     # allow failure in case there are no snapshots yet.
 163     # shellcheck disable=SC2012
 164     shopt -u nullglob
 165     files=(/mnt/o/btrbk/$vol.20*)
 166     shopt -s nullglob
 167     snaps=()
 168     if (( ${#files[@]} )); then
 169       snaps=($(ls -1avdr "${files[@]}" 2>/dev/null |head -n1 || : ))
 170     fi
 171     now=$EPOCHSECONDS
 172     maxtime=0
 173     for s in ${snaps[@]}; do
 174       file=${s##*/}
 175       t=$(date -d $(sed -r  's/(.{4})(..)(.{5})(..)(.*)/\1-\2-\3:\4:\5/' <<<${file#$vol.}) +%s)
 176       if (( t > maxtime )); then
 177         maxtime=$t
 178       fi
 179     done
 180     snapshotmsg=
 181     if (( maxtime < now - 4*60*60 )); then
 182       chars+=(OLD-SNAP)
 183       snapshotmsg="/o snapshot older than 4 hours"
 184     fi
 185     p "$snapshotmsg" | lo -1 old-snapshot
 186
 187
 188     # commented out, only using timetrap retrospectively.
 189     # # clock us out in timetrap if are idle too long
 190     # if [[ -e /p/.timetrap.db ]]; then
 191     #   export DISPLAY=:0
 192     #   if type -p xprintidle &>/dev/null && xidle=$(xprintidle 2>/dev/null); then
 193     #     if [[ $xidle == [0-9]* ]]; then
 194     #       sheet=$(sqlite3 /p/.timetrap.db "select sheet from entries where end is NULL;")
 195     #       idle=300000
 196     #       if [[ $sheet == w ]]; then
 197     #         idle=900000
 198     #       fi
 199     #       if [[ $sheet && $xidle -gt $idle ]]; then
 200     #         timetrap out
 201     #       fi
 202     #     fi
 203     #   fi
 204     # fi
 205   else # end if $MAIL_HOST
 206     rmg /home/iank/cron-errors/bounce* \
 207         /home/iank/cron-errors/btrbk.timer* \
 208         /home/iank/cron-errors/old-snapshot*
 209   fi
 210
 211   if ip l show tunfsf &>/dev/null; then
 212     # this is for tracking dns over tls issue, which
 213     # fixvpndns() in brc2 fixes.
 214     stat=$(resolvectl dnsovertls tunfsf 2>/dev/null ||: )
 215     read _ _ _ istls <<<"$stat"
 216     case $istls in
 217       no) : ;;
 218       *)
 219         printf "%s\n" "$istls" | ts >> /tmp/istls.log
 220         chars+=("T:$istls")
 221         ;;
 222     esac
 223   fi
 224
 225   # We do this once every 5 minutes, since this is not a grave problem.
 226   # For formatted elisp, see /b/ds/unsaved-buffers.el
 227   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)))'
 228   if [[ ! $last_emacs_check || $emacsfiles ]] || (( last_emacs_check < EPOCHSECONDS - 300 )); then
 229     if pgrep -G iank -u iank -f 'emacs --daemon' &>/dev/null; then
 230       # i dun care if this fails
 231       emacsfiles="$(timeout 1 emacsclient -a /usr/bin/true --eval "$elisp" 2>/dev/null | sed '/^"nil"$/d;s/^"(/E: /;s/)"$//' ||:)"
 232       if [[ $emacsfiles ]]; then
 233         chars+=("$emacsfiles")
 234       fi
 235     fi
 236     last_emacs_check=$EPOCHSECONDS
 237   fi
 238
 239
 240   glob=(/nocow/btrfs-stale/*)
 241   if [[ -e ${glob[0]} ]]; then
 242     chars+=(STALE)
 243   fi
 244   var_mail_msg=
 245   if [[ $(find /var/mail -type f \! -empty -print -quit) ]]; then
 246     var_mail_msg="message in /var/mail"
 247   fi
 248   p $var_mail_msg | loday -1 var_mail
 249
 250   # early in install process, we dont have permission yet for exiqgrep.
 251   # 1100 helps allow for system restarts
 252   qlen=$(/usr/sbin/exiqgrep -o 1100 -c -b | awk '{print $1}') ||:
 253   qmsg=
 254   if ((qlen)); then
 255     qmsg="queue length $qlen"
 256     chars+=("q $qlen")
 257   fi
 258   case $HOSTNAME in
 259     # No point in emailing about the mailq on a host where we don't
 260     # check email.
 261     $MAIL_HOST)
 262       p $qmsg | loday -120 qlen
 263       ;;
 264     *)
 265       rmg /home/iank/cron-errors/qlen*
 266       ;;
 267   esac
 268
 269   begin=false
 270
 271   # todo: make this robust to the case of /a not being mounted
 272   if ! make -C /b/ds -q ~/.local/distro-begin 2>/dev/null || [[ $(<~/.local/distro-begin) != 0 ]]; then
 273     begin=true
 274   fi
 275
 276   end=false
 277   if ! make -C /b/ds -q ~/.local/distro-end 2>/dev/null || [[ $(<~/.local/distro-end) != 0 ]]; then
 278     end=true
 279   fi
 280
 281   # these conditions are so we dont have an overly verbose prompt
 282   if $begin && $end; then
 283     chars+=(D)
 284   elif $begin; then
 285     chars+=(DB)
 286   elif $end; then
 287     chars+=(DE)
 288   else
 289     source /a/bin/ds/script-files
 290     f=~/.local/conflink
 291     # shellcheck disable=SC2043
 292     for _ in 1; do
 293       if [[ -e $f ]]; then
 294         now=$EPOCHSECONDS
 295         fsec=$(stat -c%Y $f)
 296         # the / 60 makes it 0-59 seconds less strict, +1 to help make sure we
 297         # dont have any false positives.
 298         fmin=$(( (fsec - now + 1 ) / 60 ))
 299         fminplus=$(( fmin + 60*24 ))
 300         # Filesystem files get copied, so find any newer than the last run.
 301         # The rest are hueristics:
 302         # Given the last time we added a file in git, is that newer than the last conflink run.
 303         # Given new files not added to git, were they modified more recently than the last conflink? but,
 304         # push their modification time back by a day so we can develop them before needing to add them to git.
 305
 306         all_dirs=({/a/bin/ds,/p/c}{/filesystem,/machine_specific/$HOSTNAME/filesystem})
 307         # This part is copied from conflink
 308         for x in /p/c/machine_specific/*.hosts /a/bin/ds/machine_specific/*.hosts; do
 309           if grep -qxF $HOSTNAME $x; then all_dirs+=( ${x%.hosts} ); fi
 310         done
 311
 312         # Just because i forget a lot, -mmin -NUM means files modified <= NUM minutes ago
 313         if (( fmin < 0 )) && [[ $(find ${all_my_scripts[@]} ${all_dirs[@]} -mmin $fmin -type f -print -quit 2>/dev/null) ]]; then
 314           v conflink newer filesystem files
 315           chars+=(CONFLINK)
 316           break
 317         fi
 318
 319         for d in /a/bin/distro-setup /p/c; do
 320           [[ -d $d ]] || continue
 321           cd $d
 322           if [[ ! -e .git ]]; then
 323             # some hosts i dont push all of /p/c
 324             continue
 325           fi
 326           if (( $(date -d "$(git log --diff-filter=ACR --format=%aD -1)" +%s) > fsec )); then
 327             v conflink: newer files checked in to git
 328             chars+=(CONFLINK)
 329             break
 330           fi
 331
 332           untracked=()
 333           while read -r l; do
 334             untracked+=("$l")
 335           done < <(git ls-files -o --exclude-standard)
 336           if [[ ${untracked[0]} && $(find "${untracked[@]}" -mmin $fminplus -type f -print -quit) ]]; then
 337             v conflink: untracked in $d
 338             chars+=(CONFLINK)
 339             break
 340           fi
 341         done
 342         cd /
 343
 344       fi
 345       if [[ ! -e $f || $(<$f) != 0 ]]; then
 346         v conflink: last run not found or failed
 347         chars+=(CONFLINK)
 348         break
 349       fi
 350     done
 351   fi
 352
 353   #  if [[ $(grep -v "exim user lost privilege for using -C option" /var/log/exim4/paniclog 2>/dev/null ||:) ]]; then
 354   if [[ -s /var/log/exim4/paniclog ]]; then
 355     chars+=("PANIC!")
 356     # leave it up to epanic-clean to send email notification
 357   fi
 358
 359   if [[ ! -e $status_file || -w $status_file ]]; then
 360     if [[ -e /a/bin/bash_unpublished/source-state ]]; then
 361       cat /a/bin/bash_unpublished/source-state >$status_file
 362     fi
 363
 364     if [[ ${chars[*]} ]]; then
 365       echo "ps_char=\"${chars[*]} \$ps_char\"" >>$status_file
 366     fi
 367   fi
 368 }
 369 # use this if we want to do something just once per minute
 370 first_chars=()
 371
 372
 373 write-status
 374 if [[ $1 ]]; then
 375   cat $status_file
 376   exit 0
 377 fi
 378
 379 main-loop() {
 380   while true; do
 381     power=true
 382     if [[ -e /sys/class/power_supply/AC/online && $(</sys/class/power_supply/AC/online) == 0 ]]; then
 383       power=false
 384     fi
 385     wait=15
 386     if ! $power; then
 387       if systemctl -q is-active bitcoind; then
 388         bitcoinoff
 389       fi
 390       wait=60
 391     fi
 392
 393     sleep $wait
 394     write-status
 395   done
 396 }
 397
 398 # ensure our long operations are one line so we are not prone errors
 399 # from this file being modified.
 400 main-loop