clearer names for actions, and infer actions better

[monitor.git] / automate-default.sh

diff --git a/automate-default.sh b/automate-default.sh
index a51144a..d691c66 100755 (executable)
--- a/automate-default.sh
+++ b/automate-default.sh
@@ -11,12 +11,23 @@ set -e
 DATE=`date +%Y-%m-%d-%T`
 MONITOR_PID="${MONITOR_SCRIPT_ROOT}/SKIP"
 
+function send_mail ()
+{
+    subject=$1
+    body=$2
+    mail -s "$subject" $exception_email <<EOF
+$body
+EOF
+}
+
+
 echo "#######################################"; echo "Running Monitor at $DATE"; echo "######################################"
 echo "Performing API test"
 API=$(${MONITOR_SCRIPT_ROOT}/tools/testapi.py)
 if [ "$API" != "ok" ] ; then 
        # NOTE: Do not try to run any commands if the API is obviously broken.
        echo "API IS DOWN : "`date`
+    send_mail "API IS DOWN: canceled monitor run for `date`" "have a nice day..."
        exit 1
 fi
 
@@ -62,6 +73,13 @@ fi
 #TODO: should add a call to ssh-add -l to check if the keys are loaded or not.
 source ${MONITOR_SCRIPT_ROOT}/agent.sh
 
+# CHECK AGENT IS UP AND RUNNING
+count=$( ssh-add -l | wc -l ) 
+if [ $count -lt 3 ] ; then
+    send_mail "ssh-agent is not up and running." "Add keys before monitoring can continue"
+       exit
+fi
+
 ${MONITOR_SCRIPT_ROOT}/commands/syncwithplc.py $DATE || :
 service plc restart monitor
 
@@ -74,9 +92,25 @@ ps ax | grep BatchMode | grep -v grep | awk '{print $1}' | xargs -r kill || :
 ps ax | grep locfg | grep -v grep | awk '{print $1}' | xargs -r kill || :
 
 
+${MONITOR_SCRIPT_ROOT}/commands/repair.py $DATE || :
 ${MONITOR_SCRIPT_ROOT}/commands/policy.py $DATE || :
 curl -s 'http://summer.cs.princeton.edu/status/tabulator.cgi?table=table_nodeview&formatcsv' > /var/lib/monitor/comon/$DATE.comon.csv || :
 
+#${MONITOR_SCRIPT_ROOT}/statistics/add-google-record.py --email email  --password password --database MonitorStats --sheet NodeHistory `${MONITOR_SCRIPT_ROOT}/statistics/get-records.py nodes`
+#${MONITOR_SCRIPT_ROOT}/statistics/add-google-record.py --email email --password password --database MonitorStats --sheet SiteHistory `${MONITOR_SCRIPT_ROOT}/statistics/get-records.py sites`
+
 cp ${MONITOR_SCRIPT_ROOT}/monitor.log ${MONITOR_ARCHIVE_ROOT}/`date +%F-%H:%M`.monitor.log
 service plc restart monitor || :
 rm -f $MONITOR_PID
+
+D=`date +%F-%H:%M`
+
+# NOTE: check log for major sections.
+wc=`grep -E "^(findbad|findbadpcu|nodebad|pcubad|sitebad|apply-policy)$" ${MONITOR_SCRIPT_ROOT}/monitor.log | wc -l`
+if [[ $wc -ge 6 ]] ; then 
+    send_mail "A:finished monitor run for $SD at $D" "Thank you..."
+else
+    send_mail "ERROR finished monitor run for $SD at $D" "Missing some sections:
+        $(grep -E "findbad|findbadpcu|nodebad|pcubad|sitebad|apply-policy" ${MONITOR_SCRIPT_ROOT}/monitor.log)"
+fi
+