mainly changes to keep systems up to date

[distro-setup] / system-status

diff --git a/system-status b/system-status
index a36240ca24c631296d5d944da3a6d2251ab02c87..25a637ee9606788c97e22949b470a811428abf16 100755 (executable)
--- a/system-status
+++ b/system-status
@@ -6,41 +6,97 @@
 
 if [ -z "$BASH_VERSION" ]; then echo "error: shell is not bash" >&2; exit 1; fi
 
-set -eE -o pipefail
-trap 'echo "$0:$LINENO:error: \"$BASH_COMMAND\" returned $?" >&2' ERR
-
+source /a/bin/errhandle/err
 status_file=/dev/shm/iank-status
 
 shopt -s nullglob
 shopt -s dotglob
 
+lo() { /usr/local/bin/log-once "$@"; }
+
 write-status() {
   chars=()
 
   glob=(/nocow/btrfs-stale/*)
-  if [[ -e $glob ]] ; then
+  if [[ -e ${glob[0]} ]]; then
     chars+=("STALE!")
   fi
   glob=(/m/md/bounces/new/*)
-  if [[ -e $glob ]]; then
+  if [[ -e ${glob[0]} ]]; then
     chars+=("BOUNCE!")
+    bouncemsg="message in /m/md/bounces/new"
   fi
-  glob=(/m/md/alerts/new/*)
-  if [[ -e $glob ]]; then
+  lo -1 bounce $bouncemsg
+  glob=(/m/md/alerts/new/* /m/md/alerts/cur/*)
+  if [[ -e ${glob[0]} ]]; then
     chars+=("ALERT!")
   fi
   if [[ -e /nocow/user/mailtest-failure ]]; then
     chars+=("MAILPING!")
   fi
+
+
+  if ! make -q ~/.local/distro-begin || [[ $(<~/.local/distro-begin) != 0 ]]; then
+    chars+=("DISTRO-BEGIN!")
+  fi
+
+  if ! make -q ~/.local/distro-end || [[ $(<~/.local/distro-end) != 0 ]]; then
+    chars+=("DISTRO-END!")
+  fi
+
+  f=~/.local/conflink
+  if [[ -e $f ]]; then
+    cd /b/ds
+    now=$(date +%s)
+    fsec=$(stat -c%Y $f)
+    fmin=$(( (fsec - now ) / 60 + 1 ))
+    fminplus=$(( fmin + 60*24 ))
+    # Filesystem files get copied, so find any newer than the last run.
+    # The rest are hueristics:
+    # Given the last time we added a file in git, is that newer than the last conflink run.
+    # Given new files not added to git, were they modified more recently than the last conflink? but,
+    # push their modification time back by a day so we can develop them before needing to add them to git.
+    if (( $(date -d "$(git log --diff-filter=ACR --format=%aD -1)" +%s) > fsec )) || \
+         [[ $(find {/a/bin/ds,/p/c}{/filesystem,/machine_specific/$HOSTNAME/filesystem} -mmin $fmin -type f -print -quit 2>/dev/null) ]] \
+         || [[ $(find $(git ls-files -o --exclude-standard) -mmin $fminplus -type f -print -quit) ]]; then
+      chars+=("CONFLINK!")
+    fi
+  fi
+
+  if [[ ! -e $f || $(<$f) != 0 ]]; then
+    chars+=("CONFLINK!")
+  fi
+
+
+  ## Clean the paniclog, but only up to 4 times per day, or else we
+  ## should investigate.
+  loglog=/tmp/panicloglog-$(date --rfc-3339=date)
+  if [[ -s $loglog ]]; then
+    spamcount=$(stat -c%s $loglog)
+  else
+    spamcount=0
+  fi
+  if (( spamcount <= 4 )); then
+    if grep -q 'spam acl condition' /var/log/exim4/paniclog; then
+      printf . >>$loglog
+    fi
+    /a/bin/distro-setup/epanic-clean
+  fi
+
   if [[ -s /var/log/exim4/paniclog ]]; then
     chars+=("PANIC!")
+    tail -n 20 /var/log/exim4/paniclog | lo -1 paniclog
+  else
+    lo -1 paniclog
   fi
 
   source /a/bin/bash_unpublished/source-state
-  if [[ $MAIL_HOST == $HOSTNAME ]]; then
+  if [[ $MAIL_HOST == "$HOSTNAME" ]]; then
     if [[ $(systemctl is-active btrbk.timer) != active ]]; then
-      chars+=("BTRBK-TIMER!")
+      chars+=("BTRBK.TIMER!")
+      bbkmsg="btrbk.timer not enabled"
     fi
+    lo -60 btrbk.timer $bbkmsg
 
     ## check if last snapshot was within an hour
     vol=o
@@ -56,18 +112,20 @@ write-status() {
     done
     if (( maxtime < now - 60*60 )); then
       chars+=("OLD-SNAPSHOT!")
+      snapshotmsg="/o snapshot older than 1 hour"
     fi
+    lo -1 old-snapshot $snapshotmsg
   fi
 
   cat /a/bin/bash_unpublished/source-state >$status_file
 
-  if [[ $chars ]]; then
+  if [[ ${chars[*]} ]]; then
     echo "ps_char=\"${chars[*]} \$ps_char\"" >>$status_file
   fi
 
 }
 write-status
-if [[ $@ ]]; then
+if [[ $1 ]]; then
   cat $status_file
   exit 0
 fi