fix
[distro-setup] / mailtest-check
index c7d40c17502e967b808e2e251a6b50b63cb1464c..599d4b6a4157ab0ddc31e74b58b0766e8cc61bc4 100755 (executable)
@@ -28,11 +28,6 @@ getspamdpid() {
     done
   fi
 }
-pr() {
-  if $doprom && [[ -e /var/lib/prometheus/node-exporter ]]; then
-    cat >>/var/lib/prometheus/node-exporter/mailtest-check.prom.$$
-  fi
-}
 
 
 #### begin arg processing ####
@@ -57,10 +52,11 @@ if [[ $1 == nonint ]]; then
 fi
 #### end arg processing ####
 
-
-if ! $int; then
-  sleep 60
-fi
+# we put this in to avoid dns errors that happen on reboot,
+# but I want to debug them.
+# if ! $int; then
+#   sleep 60
+# fi
 
 
 # TODO, get je to deliver the local mailbox: /m/md/INBOX
@@ -80,6 +76,7 @@ esac
 
 main() {
 
+  local -a unexpected_spamd_results missing_dnswl last_usec
   case $HOSTNAME in
     bk)
       folders=(/m/md/{expertpathologyreview.com,amnimal.ninja}/testignore)
@@ -103,6 +100,7 @@ main() {
           try_left=$(( try_limit - ( EPOCHSECONDS - try_start_time) ))
           timeout=120 # somewhat arbitrary value
           if (( try_left < 0 )); then
+            echo "mailtest-check: failed to rsync fencepost > $try_limit seconds"
             break
           fi
           if (( try_left < timeout )); then
@@ -144,9 +142,11 @@ EOF
   fi
   tmpfile=$(mktemp)
   declare -i unexpected=0
-  declare -i missing_dnswl=0
   for folder in ${folders[@]}; do
     for from in ${froms[@]}; do
+      declare -i missing_dnswl=0
+      declare -i dnsfail=0
+      declare -i unexpected=0
       latest=
       last_sec=0
 
@@ -215,8 +215,8 @@ EOF
             # eggs has RCVD_IN_DNSWL_MED
             keys+=(RCVD_IN_DNSWL_MED)
           elif [[ $from == *@gnu.org ]]; then
-            # eggs has these
-            keys+=(RCVD_IN_DNSWL_MED DKIMWL_WL_HIGH)
+            # eggs has this. it used to have DKIMWL_WL_HIGH sometime in 2022
+            keys+=(RCVD_IN_DNSWL_MED)
           fi
 
           for t in  ${keys[@]}; do
@@ -264,9 +264,10 @@ EOF
           rm -f $resultfile
           for r in ${results[@]}; do
             case $r in
-              DKIM_INVALID|T_SPF_TEMPERROR|T_SPF_HELO_TEMPERROR)
-                missing_dnswl+=1
-                ;;
+              # iank: for when we want to handle dns errors differently
+              # DKIM_INVALID|T_SPF_TEMPERROR|T_SPF_HELO_TEMPERROR)
+              #   dnsfail+=1
+              #   ;;
               *)
                 unexpected=$(( unexpected + 1 ))
                 ;;
@@ -276,7 +277,9 @@ EOF
             # We expect dns failures from time to time, so
             # we count them separately and alert differently.
             case $miss in
-              DKIM_VALID|DKIM_VALID_AU|DKIM_VALID_EF|SPF_HELO_PASS|SPF_PASS|RCVD_IN_DNSWL_MED|DKIMWL_WL_HIGH)
+              # iank: dns fail
+              # DKIM_VALID|DKIM_VALID_AU|DKIM_VALID_EF|SPF_HELO_PASS|SPF_PASS|
+              RCVD_IN_DNSWL_MED|DKIMWL_WL_HIGH)
                 missing_dnswl+=1
                 ;;
               *)
@@ -284,6 +287,12 @@ EOF
                 ;;
             esac
           done
+          mapfile -O ${#missing_dnswl[@]} -t missing_dnswl <<EOF
+mailtest_check_missing_dnswl{folder="$folder",from="$from"} $missing_dnswl
+EOF
+          mapfile -O ${#unexpected_spamd_results[@]} -t unexpected_spamd_results <<EOF
+mailtest_check_unexpected_spamd_results{folder="$folder",from="$from"} $unexpected
+EOF
         fi # if spamdpid
       fi # if $slow
 
@@ -292,21 +301,25 @@ EOF
       e $((age_sec / 60)):$(( age_sec % 60 )) ago. to:$to from:$from $latest
 
       # usec = unix seconds
-      pr <<EOF
+      mapfile -O ${#last_usec[@]} -t last_usec <<EOF
 mailtest_check_last_usec{folder="$folder",from="$from"} $last_sec
 EOF
     done # end for from in ${froms[@]}
   done # end for folder in ${folders[@]}
-  if $slow; then
-    pr <<EOF
-mailtest_check_missing_dnswl $missing_dnswl
-mailtest_check_unexpected_spamd_results $unexpected
-EOF
-  fi
 
   dir=/var/lib/prometheus/node-exporter
-  if [[ -e $dir  ]]; then
-    mv $dir/mailtest-check.prom.$$ $dir/mailtest-check.prom
+  path=$dir/mailtest-check.prom.$$
+  if $doprom && [[ -e $dir  ]]; then
+    for l in "${unexpected_spamd_results missing_dnswl[@]}"; do
+      printf "%s\n" "$l" >>$path
+      done
+    for l in "${missing_dnswl[@]}"; do
+      printf "%s\n" "$l" >>$path
+    done
+    for l in "${last_usec[@]}"; do
+      printf "%s\n" "$l" >>$path
+    done
+    mv $path $dir/mailtest-check.prom
     # note: node_textfile_mtime_seconds will tell us when this last happened. useful for debugging.
   fi
 }