fix alerts for down hosts

[distro-setup] / check-remote-mailqs

diff --git a/check-remote-mailqs b/check-remote-mailqs
index 6e9cf7c2cebbd1cfde4df4fe0d971dd9b1802055..2e98521d01e1359d2d6284953a9e2992029639fd 100755 (executable)
--- a/check-remote-mailqs
+++ b/check-remote-mailqs
@@ -9,8 +9,9 @@ source /a/bin/errhandle/err
 shopt -s nullglob
 shopt -s dotglob
 
+# temp disabled: x2wg kwwg
 # this list duplicated in brc2 ralerts
-for h in bk je li frodo kwwg x3wg x2wg kdwg sywg; do
+for h in bk je li frodo x3wg kdwg sywg; do
   statedir=/b/bash_unpublished/mailq-state
   statefile=$statedir/$h
   [[ -d $statedir ]] || continue
@@ -26,9 +27,16 @@ for h in bk je li frodo kwwg x3wg x2wg kdwg sywg; do
   else
     if [[ -s $statefile ]]; then
       logsec=$(date +%s -d "$(head -n1 $statefile | awk '{print $1,$2}')")
-      nowsec=$(date +%s)
-      if (( logsec < nowsec - 60*60*20 )); then
-        echo $0: host $h ssh /usr/local/bin/check-mailq fail for over 20 hours
+      case h in
+        frodo)
+          hours=200
+          ;;
+        *)
+          hours=20
+          ;;
+        esac
+      if (( logsec < EPOCHSECONDS - 60*60*hours )); then
+        echo $0: host $h ssh /usr/local/bin/check-mailq fail for over $hours hours
       fi
     fi
     printf "%s\n" "$c" | ts "%F %T" >> $statefile