updates to improve generalization and auto-installation.

[monitor.git] / automate-default.sh

diff --git a/automate-default.sh b/automate-default.sh
index 8e7be9c..d23cf38 100755 (executable)
--- a/automate-default.sh
+++ b/automate-default.sh
@@ -8,7 +8,7 @@ source $INSTALLPATH/monitorconfig.sh
 cd ${MONITOR_SCRIPT_ROOT}
 set -e
 DATE=`date +%Y-%m-%d-%T`
-MONITOR_PID="$HOME/monitor/SKIP"
+MONITOR_PID="${MONITOR_SCRIPT_ROOT}/SKIP"
 
 echo "Performing API test"
 API=$(./testapi.py)
@@ -41,7 +41,9 @@ if [ ! -f ${MONITOR_SCRIPT_ROOT}/actallsetup.flag ]; then
 fi
 
 
+set +e
 AGENT=`ps ax | grep ssh-agent | grep -v grep`
+set -e
 if [ -z "$AGENT" ] ; then
         echo "starting ssh agent"
         # if no agent is running, set it up.
@@ -71,11 +73,11 @@ cp ${MONITOR_DATA_ROOT}/production.findbadpcus2.pkl ${MONITOR_DATA_ROOT}/product
 # clean up stray 'locfg' processes that hang around inappropriately...
 ps ax | grep locfg | grep -v grep | awk '{print $1}' | xargs kill || :
 
-echo "Generating web data"
+#echo "Generating web data"
 # badcsv.txt
-${MONITOR_SCRIPT_ROOT}/printbadcsv.py  | grep -v loading | tr -d ' ' > badcsv.txt
-cp badcsv.txt /plc/data/var/www/html/monitor/
-${MONITOR_SCRIPT_ROOT}/showlatlon.py | head -9 | awk 'BEGIN {print "<table>"} { print "<tr><td>", $0, "</td></tr>"} END{print "</table>"}'  | sed -e 's\|\</td><td>\g' > /plc/data/var/www/html/monitor/regions.html
+#${MONITOR_SCRIPT_ROOT}/printbadcsv.py  | grep -v loading | tr -d ' ' > badcsv.txt
+#cp badcsv.txt /plc/data/var/www/html/monitor/
+#${MONITOR_SCRIPT_ROOT}/showlatlon.py | head -9 | awk 'BEGIN {print "<table>"} { print "<tr><td>", $0, "</td></tr>"} END{print "</table>"}'  | sed -e 's\|\</td><td>\g' > /plc/data/var/www/html/monitor/regions.html
 
 echo "Performing uptime changes for sites, nodes, and pcus"
 ########################
@@ -91,8 +93,8 @@ ${MONITOR_SCRIPT_ROOT}/pkl2php.py -i findbadpcus2 -o findbadpcus
 ${MONITOR_SCRIPT_ROOT}/pkl2php.py -i act_all -o act_all
 ${MONITOR_SCRIPT_ROOT}/pkl2php.py -i plcdb_hn2lb -o plcdb_hn2lb
 ${MONITOR_SCRIPT_ROOT}/pkl2php.py -i findbad -o findbadnodes
-${MONITOR_SCRIPT_ROOT}/pkl2php.py -i ad_dbTickets -o ad_dbTickets
-${MONITOR_SCRIPT_ROOT}/pkl2php.py -i idTickets -o idTickets
+#${MONITOR_SCRIPT_ROOT}/pkl2php.py -i ad_dbTickets -o ad_dbTickets
+#${MONITOR_SCRIPT_ROOT}/pkl2php.py -i idTickets -o idTickets
 
 echo "Archiving pkl files"
 #########################
@@ -104,15 +106,8 @@ done
 echo "Running grouprins on all dbg nodes"
 ############################
 # 5. Check if there are any nodes in dbg state.  Clean up afterward.
-${MONITOR_SCRIPT_ROOT}/grouprins.py --mail=1 \
-       --nodeselect 'state=DEBUG&&boot_state=(rins|dbg|boot)' \
-       --stopselect 'state=BOOT&&kernel=2.6.22.19-vs2.3.0.34.9.planetlab' \
-       --reboot || :
-${MONITOR_SCRIPT_ROOT}/findbad.py --increment --cachenodes --debug=0 --dbname="findbad" --nodeselect 'state=DEBUG&&boot_state=dbg||state=DEBUG&&boot_state=boot' || :
-
-echo "Collecting RT database dump"
-##########################
-# 6. cache the RT db locally.
-python ${MONITOR_SCRIPT_ROOT}/rt.py
+${MONITOR_SCRIPT_ROOT}/grouprins.py --mail=1 --reboot --nodeselect 'state=DOWN&&boot_state=(boot|rins|dbg|diag)' --stopselect "state=BOOT" || :
+${MONITOR_SCRIPT_ROOT}/grouprins.py --mail=1 --reboot --nodeselect 'state=DEBUG&&boot_state=(rins|dbg|boot)' --stopselect 'state=BOOT' || :
 
+cp ${MONITOR_SCRIPT_ROOT}/monitor.log ${MONITOR_ARCHIVE_ROOT}/`date +%F-%H:%M`.monitor.log
 rm -f $MONITOR_PID