reorganized ProductionTools for installation of tools

DeploymentTest/p1.test.atlascdr.bash

@@ -107,7 +107,7 @@ while [ "x$LISTING_OUTPUT" != x ]; do
 done
 
 echo "Stopping CastorScript"
-~/signal-script.sh 12
+/sw/castor/tools/castor.signal 12
 
 if [[ $COPY_ENABLED == True && $REMOTE_DIR == /eos/* ]]; then
   echo "Contents of remote directory:"

DeploymentTest/stop.all.cs.instances.at.p1.machine.bash

-#!/usr/bin/env bash
-
-if [ -z $1 ]; then
-	echo "usage: $0 {test host}"
-	exit 1
-fi
-
-TESTHOST=$1
-
-
-KEYTAB_FILE="/daq_area/castor/$TESTHOST/atlascdr/atlascdr.keytab"
-if ! ssh $TESTHOST "sudo -u atlascdr -i cat $KEYTAB_FILE &>/dev/null"; then
-	echo "error: keytab file not readble by atlascdr ($KEYTAB_FILE)"
-	exit 3
-else
-	echo "OK: readable keytab file: $KEYTAB_FILE"
-fi
-
-SOURCE_DIR=`pwd`/..
-TOOLS_DIR=`pwd`/../../flegoff/castor_tools
-
-echo "sending signal 12 to CastorScripts at $TESTHOST"
-ssh $TESTHOST sudo -u atlascdr -i $TOOLS_DIR/signal-script.sh 12

ProductionTools/castor.eostimeout.freq

-#!/usr/bin/env bash
-grep "taking to long" /atlas/logs/castor/pc-tdq-sfo-0?/CopyLog.out | awk -F':' '{print $2}' | sort | uniq -c
-

ProductionTools/castor.checklogs → ProductionTools/installed/castor.checklogs

@@ -6,11 +6,10 @@
 for c in `cat castor.confs`; do
 	logdir=$(grep LogDir $c | awk '{print $NF}' | tr -d "'")
 	logdir=${logdir%/} # remove trailing slash
-	lsl=$(ls -l `echo "$logdir/*.out"` 2>/dev/null)
+	lsl=$(ls -l `echo "$logdir/*.log"` 2>/dev/null)
 	if [ $? -eq 0 ]; then
 		# Special case for ManagerLog.out which always contain 80/81 bytes
 		# because it outputs its thread ID even if logged are limited to >info
-		#echo "$lsl" | awk '/.*ManagerLog.out$/{if ($5 > 82) print $NF ": " $5} ! /.*ManagerLog.out$/ {if ($5 > 0) print $NF ": " $5}'
 		echo "$lsl" | awk '{if ($5 > 0) print $NF ": " $5}'
 	else
 		echo "No log files for $c: $logdir"

ProductionTools/castor.greplogs → ProductionTools/installed/castor.greplogs

@@ -14,7 +14,7 @@ fi
 
 for conf in "$CASTOR_CONFS"; do
 	logdir=`grep LogDir $conf | awk '{print $NF}' | tr -d "'"`
-	for logfile in `find $logdir -maxdepth 1 -name '*.out'`; do #-maxdepth 1
+	for logfile in `find $logdir -maxdepth 1 -name '*.log.*'`; do #-maxdepth 1
 		grep "$TOGREP" $logfile 
 	done
 done

ProductionTools/castor.logsfreq → ProductionTools/installed/castor.logsfreq

@@ -26,7 +26,7 @@ for i in `seq -$(($NB_DAYS - 1)) 0`; do
 
 	for conf in "$CASTOR_CONFS"; do
 		logdir=`grep LogDir $conf | awk '{print $NF}' | tr -d "'"`
-		for logfile in `find $logdir -name '*.out'`; do #-maxdepth 1
+		for logfile in `find $logdir -name '*.log.*'`; do #-maxdepth 1
 			grep "$TOGREP" $logfile | grep -c $DATE
 		done
 	done | awk 'BEGIN {s=0} {s+=$1} END {print s}'

ProductionTools/sfotzdb_intr_writer.py → ProductionTools/installed/castor.sfotzdb_intr_writer.py

+#!/sw/atlas/sw/lcg/releases/LCG_101/Python/3.9.6/x86_64-centos7-gcc11-opt/bin/python -i
 import cx_Oracle
 import coral_auth
 user,pwd,dbn = coral_auth.get_connection_parameters_from_connection_string('oracle://int8r/ATLAS_SFO_T0')

ProductionTools/sfotzdb_reader.py → ProductionTools/installed/castor.sfotzdb_reader.py

+#!/sw/atlas/sw/lcg/releases/LCG_101/Python/3.9.6/x86_64-centos7-gcc11-opt/bin/python -i
 import cx_Oracle
 import coral_auth
 user,pwd,dbn = coral_auth.get_connection_parameters_from_connection_string('oracle://atonr_conf/ATLAS_SFO_T0_R')

ProductionTools/sfotzdb_writer.py → ProductionTools/installed/castor.sfotzdb_writer.py

+#!/sw/atlas/sw/lcg/releases/LCG_101/Python/3.9.6/x86_64-centos7-gcc11-opt/bin/python -i
 import cx_Oracle
 import coral_auth
 user,pwd,dbn = coral_auth.get_connection_parameters_from_connection_string('oracle://atonr_conf/ATLAS_SFO_T0')

ProductionTools/installed/castor.signal

+#!/bin/bash
+
+pid=`ps -elf | grep Castor | grep python | grep -v grep | awk '{print \$4}'`
+if [ "x$pid" == x ] ; then
+    echo "cannot send signal: castor script is not running"
+    exit -1
+else
+    kill -s $1 $pid
+fi

ProductionTools/castor.stopall → ProductionTools/installed/castor.stopall

 #!/usr/bin/env bash
 for c in `cat castor.machines`; do
 	echo "$c"
-	ssh -x $c "~flegoff/castor_tools/signal-script.sh 12"
+	ssh -x $c "/sw/castor/tools/castor.signal 12"
 done

ProductionTools/oracle_db_disaster_recovery/oracle_db_disaster_recovery.py

@@ -103,7 +103,8 @@ if not args.dryrun:
                     f'/mnt/daq_area_rw/castor/pc-tdq-sfo-{i}/atlascdr/prod.stopped')
             if args.verbose: print(f'disabled CS on sfo-{i}')
         if castorscript_is_running(f'pc-tdq-sfo-{i}', 'prod.cfg'):
-            stop_cs_cmd = ['ssh', '-x', f'pc-tdq-sfo-{i}', '~/signal-script.sh', '12']
+            stop_cs_cmd = ['ssh', '-x', f'pc-tdq-sfo-{i}',
+                        '/sw/castor/tools/castor.signal', '12']
             subprocess.run(stop_cs_cmd, check=True)
             if args.verbose: print(f'sent stop signal to CS on sfo-{i}: waiting for CS to stop')
             while castorscript_is_running(f'pc-tdq-sfo-{i}', 'prod.cfg'):

point1.deployment.sh

@@ -32,6 +32,8 @@ fi
 
 # DB disaster recovery are deployed as part of the CastorScript
 mv ProductionTools/oracle_db_disaster_recovery . || err "cannot mv oracle_db_disaster_recovery"
+# operation tools
+mv ProductionTools/installed tools || err "cannot mv installed tools"
 
 # remove from local directory what we don't want to be deployed
 rm -rf Configs DeploymentTest ProductionTools UnitTests .git || err "cannot delete"