various fixes, a few new features

[distro-setup] / check-stale-alerts

diff --git a/check-stale-alerts b/check-stale-alerts
index a6d8c822ee0f9f3db3859f7c3e4e40607b701cc1..88fa0962823fab41b9dd16757c8cfb6dc7be3cd6 100755 (executable)
--- a/check-stale-alerts
+++ b/check-stale-alerts
@@ -1,25 +1,19 @@
 #!/bin/bash
 
-time_arg="-ctime +4"
-case $1 in
-  now)
-    time_arg=
-    ;;
-esac
 
 if [[ ! -e /dev/shm/iank-status ]]; then
   exit 0
 fi
 eval $(< /dev/shm/iank-status)
 
-if [[ $HOSTNAME != "$MAIL_HOST" ]]; then
-  exit 0
-fi
-out=$(find /var/local/cron-errors /home/iank/cron-errors /sysd-mail-once-state -type f -ctime +4)
+out=$(find /var/local/cron-errors /home/iank/cron-errors /sysd-mail-once-state -type f)
 if [[ $out ]]; then
   echo HOSTNAME: $HOSTNAME
   printf "%s\n" "$out"
 fi
+if [[ $HOSTNAME != "$MAIL_HOST" ]]; then
+  exit 0
+fi
 for h in {li,bk,je}.b8.nz; do
   out=$(ssh $h find /m/md/bounces/new /var/local/cron-errors /home/iank/cron-errors /sysd-mail-once-state -type f)
   if [[ $out ]]; then
@@ -27,3 +21,8 @@ for h in {li,bk,je}.b8.nz; do
     printf "%s\n" "$out"
   fi
 done
+
+# for .prom.$PID files, they will naturally exist when
+# scripts like system-status get killed on reboot.
+# Just clean them up here.
+find /var/lib/prometheus/node-exporter -name '*[0-9]' -type f -mtime 9 -delete