From ab4ff14abc8627ea48212dc4298d04f899285346 Mon Sep 17 00:00:00 2001
From: "Raymond P. Burkholder" <github@oneunified.net>
Date: Fri, 2 Sep 2016 17:18:20 -0300
Subject: [PATCH] provide pagerduty alert

---
 path-loss.sh | 76 +++++++++++++++++++++++++++++++++++++++++-----------
 1 file changed, 61 insertions(+), 15 deletions(-)

diff --git a/path-loss.sh b/path-loss.sh
index 0ef5ca6..75a42a8 100644
--- a/path-loss.sh
+++ b/path-loss.sh
@@ -1,42 +1,78 @@
 #!/bin/bash
 
 # rburkholder@quovadis.bm
+
 # raymond@burkholder.net
+# http://blog.raymond.burkholder.net
+
+# requires use of (with some modifications, will post):
+#   https://github.com/enigma-io/pd-trigger
 
 # user defined settings
-txtEmail="user@example.com"
+txtEmail="name@example.com"
 txtSubject="Path Loss Report"
 txtSeparator="==============================\n"
 nTrigger=4
 nAttempts=5
+txtPdServiceKey="000000000000000"
 
 declare -A nodes
 nodes=( \
-  [8.8.8.8]="google1" \
+  [8.8.8.8]="google1 member" \
   [8.8.4.4]="google2" \
   )
 
 # local variables
-cntNotify=0
 status="|"
+cntNotify=0
 tmpLog=$(mktemp)
 tmpPing=$(mktemp)
+cntNodes=0
+cntNodesDown=0
+cntMembers=0
+cntMembersDown=0
+declare -a items
 
 # preload output
 date > ${tmpLog}
 
 # loop through nodes and test
 for node in  ${!nodes[*]}; do
+
+  ((cntNodes++))
+
+  # split out node details
+  # 0: alias/name
+  # 1: optional 'member' for determining edge outage
+  info=${nodes[${node}]}
+  items[1]="none"
+  ix=0
+  for arg in ${info}; do
+    items[ix]=${arg}
+    ((ix++))
+    done
+
+  name="${items[0]}"
+
   echo -e ${txtSeparator} >> ${tmpLog};
-  echo "checking node ${nodes[${node}]}:" >> ${tmpLog};
+  echo "checking node ${name}:" >> ${tmpLog};
   echo "" >> ${tmpLog}
 
   ping -W 1 -c ${nAttempts} ${node} > ${tmpPing}
   cat ${tmpPing} >> ${tmpLog}
 
   value=$(grep transmitted ${tmpPing} | cut -d ' ' -f 4)
-  if [[ nTrigger -ge value ]]; then flagNxt="dn"
-  else flagNxt="up"; fi
+  if [[ nTrigger -ge value ]]; then
+    flagNxt="dn"
+    ((cntNodesDown++))
+  else
+    flagNxt="up"
+    fi
+
+  if test "member" == "${items[1]}"; then
+    ((cntMembers++))
+    if test "dn" = "${flagNxt}"; then ((cntMembersDown++)); fi
+    fi
 
   flagPrv="na"
 
@@ -44,35 +80,45 @@ for node in  ${!nodes[*]}; do
     flagPrv="dn"
     if test "up" = "${flagNxt}"; then
       rm /tmp/pl.dn.${node}
+      fi
     fi
-  fi
 
   if [[ -f /tmp/pl.up.${node} ]]; then
     flagPrv="up"
     if test "dn" = "${flagNxt}"; then
       rm /tmp/pl.up.${node}
+      fi
     fi
-  fi
 
   if test "${flagPrv}" != "${flagNxt}"; then
     touch /tmp/pl.${flagNxt}.${node}
     ((cntNotify++));
     mtr -w -b --report ${node} >> ${tmpLog};
-    status="${status} ${nodes[${node}]} ${flagPrv}>${flagNxt} |"
+    status="${status} ${name} ${flagPrv}>${flagNxt} |"
     echo "" >> ${tmpLog}
     echo "Above State Change: ${flagPrv}>${flagNxt}" >> ${tmpLog}
-  fi
+    fi
 
-done
+  done
 
 # footer
-echo -e ${txtSeparator} >> ${tmpLog};
+echo -e ${txtSeparator} >> ${tmpLog}
 date >> ${tmpLog}
 
 # notify on failure
-if [[ cntNotify -gt 0 ]]; then
-  cat ${tmpLog} | mail -s "${txtSubject}:${status}" ${txtEmail};
-fi
+if [[ cntNodes -eq cntNodesDown ]]; then
+  logger "path-loss - all nodes unreachable"
+else
+  # if something to notify
+  if [[ cntNotify -gt 0 ]]; then
+    # need a pagerduty alert if all important members are down
+    if [[ cntMembers -eq cntMembersDown ]]; then
+      response=$(cat ${tmpLog} | ./pd-trigger.sh -L -s "${txtPdServiceKey}" -d "${status}")
+      fi
+    # attempt an email
+    cat ${tmpLog} | mail -s "${txtSubject}:${status}" ${txtEmail};
+    fi
+  fi
 
 # clean up
 rm ${tmpLog}