From 893841afee20baff3b9a7f794db6f1ecefc9e7d6 Mon Sep 17 00:00:00 2001
From: Fabrice Le Goff <fabrice.le.goff@cern.ch>
Date: Thu, 28 Apr 2022 19:12:17 +0200
Subject: [PATCH] reorganized ProductionTools for installation of tools
---
DeploymentTest/p1.test.atlascdr.bash | 2 +-
.../stop.all.cs.instances.at.p1.machine.bash | 23 -------------------
ProductionTools/castor.eostimeout.freq | 3 ---
.../{ => installed}/castor.checklogs | 3 +--
.../{ => installed}/castor.greplogs | 2 +-
ProductionTools/{ => installed}/castor.logdir | 0
.../{ => installed}/castor.logsfreq | 2 +-
.../{ => installed}/castor.notrunning | 0
ProductionTools/{ => installed}/castor.ps | 0
.../{ => installed}/castor.pslocal | 0
.../castor.sfotzdb_intr_writer.py} | 1 +
.../castor.sfotzdb_reader.py} | 1 +
.../castor.sfotzdb_writer.py} | 1 +
ProductionTools/installed/castor.signal | 9 ++++++++
.../{ => installed}/castor.starttime | 0
.../{ => installed}/castor.stopall | 2 +-
.../castor.updateconflist} | 0
.../oracle_db_disaster_recovery.py | 3 ++-
point1.deployment.sh | 2 ++
19 files changed, 21 insertions(+), 33 deletions(-)
delete mode 100755 DeploymentTest/stop.all.cs.instances.at.p1.machine.bash
delete mode 100755 ProductionTools/castor.eostimeout.freq
rename ProductionTools/{ => installed}/castor.checklogs (75%)
rename ProductionTools/{ => installed}/castor.greplogs (82%)
rename ProductionTools/{ => installed}/castor.logdir (100%)
rename ProductionTools/{ => installed}/castor.logsfreq (91%)
rename ProductionTools/{ => installed}/castor.notrunning (100%)
rename ProductionTools/{ => installed}/castor.ps (100%)
rename ProductionTools/{ => installed}/castor.pslocal (100%)
rename ProductionTools/{sfotzdb_intr_writer.py => installed/castor.sfotzdb_intr_writer.py} (69%)
mode change 100644 => 100755
rename ProductionTools/{sfotzdb_reader.py => installed/castor.sfotzdb_reader.py} (70%)
mode change 100644 => 100755
rename ProductionTools/{sfotzdb_writer.py => installed/castor.sfotzdb_writer.py} (70%)
mode change 100644 => 100755
create mode 100755 ProductionTools/installed/castor.signal
rename ProductionTools/{ => installed}/castor.starttime (100%)
rename ProductionTools/{ => installed}/castor.stopall (56%)
rename ProductionTools/{generate_instance_lists.sh => installed/castor.updateconflist} (100%)
diff --git a/DeploymentTest/p1.test.atlascdr.bash b/DeploymentTest/p1.test.atlascdr.bash
index 02eee28..5c68661 100755
--- a/DeploymentTest/p1.test.atlascdr.bash
+++ b/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:"
diff --git a/DeploymentTest/stop.all.cs.instances.at.p1.machine.bash b/DeploymentTest/stop.all.cs.instances.at.p1.machine.bash
deleted file mode 100755
index 1e04765..0000000
--- a/DeploymentTest/stop.all.cs.instances.at.p1.machine.bash
+++ /dev/null
@@ -1,23 +0,0 @@
-#!/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
diff --git a/ProductionTools/castor.eostimeout.freq b/ProductionTools/castor.eostimeout.freq
deleted file mode 100755
index ebf039f..0000000
--- a/ProductionTools/castor.eostimeout.freq
+++ /dev/null
@@ -1,3 +0,0 @@
-#!/usr/bin/env bash
-grep "taking to long" /atlas/logs/castor/pc-tdq-sfo-0?/CopyLog.out | awk -F':' '{print $2}' | sort | uniq -c
-
diff --git a/ProductionTools/castor.checklogs b/ProductionTools/installed/castor.checklogs
similarity index 75%
rename from ProductionTools/castor.checklogs
rename to ProductionTools/installed/castor.checklogs
index 336c833..818c70b 100755
--- a/ProductionTools/castor.checklogs
+++ b/ProductionTools/installed/castor.checklogs
@@ -6,11 +6,10 @@
for c in `cat castor.confs`; do
logdir=$(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: $logdir"
diff --git a/ProductionTools/castor.greplogs b/ProductionTools/installed/castor.greplogs
similarity index 82%
rename from ProductionTools/castor.greplogs
rename to ProductionTools/installed/castor.greplogs
index 493b2fc..096f716 100755
--- a/ProductionTools/castor.greplogs
+++ b/ProductionTools/installed/castor.greplogs
@@ -14,7 +14,7 @@ fi
for conf in "$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" $logfile
done
done
diff --git a/ProductionTools/castor.logdir b/ProductionTools/installed/castor.logdir
similarity index 100%
rename from ProductionTools/castor.logdir
rename to ProductionTools/installed/castor.logdir
diff --git a/ProductionTools/castor.logsfreq b/ProductionTools/installed/castor.logsfreq
similarity index 91%
rename from ProductionTools/castor.logsfreq
rename to ProductionTools/installed/castor.logsfreq
index 7ec1fef..b295705 100755
--- a/ProductionTools/castor.logsfreq
+++ b/ProductionTools/installed/castor.logsfreq
@@ -26,7 +26,7 @@ for i in `seq -$(($NB_DAYS - 1)) 0`; do
for conf in "$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}'
diff --git a/ProductionTools/castor.notrunning b/ProductionTools/installed/castor.notrunning
similarity index 100%
rename from ProductionTools/castor.notrunning
rename to ProductionTools/installed/castor.notrunning
diff --git a/ProductionTools/castor.ps b/ProductionTools/installed/castor.ps
similarity index 100%
rename from ProductionTools/castor.ps
rename to ProductionTools/installed/castor.ps
diff --git a/ProductionTools/castor.pslocal b/ProductionTools/installed/castor.pslocal
similarity index 100%
rename from ProductionTools/castor.pslocal
rename to ProductionTools/installed/castor.pslocal
diff --git a/ProductionTools/sfotzdb_intr_writer.py b/ProductionTools/installed/castor.sfotzdb_intr_writer.py
old mode 100644
new mode 100755
similarity index 69%
rename from ProductionTools/sfotzdb_intr_writer.py
rename to ProductionTools/installed/castor.sfotzdb_intr_writer.py
index 3437db7..857da17
--- a/ProductionTools/sfotzdb_intr_writer.py
+++ b/ProductionTools/installed/castor.sfotzdb_intr_writer.py
@@ -1,3 +1,4 @@
+#!/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')
diff --git a/ProductionTools/sfotzdb_reader.py b/ProductionTools/installed/castor.sfotzdb_reader.py
old mode 100644
new mode 100755
similarity index 70%
rename from ProductionTools/sfotzdb_reader.py
rename to ProductionTools/installed/castor.sfotzdb_reader.py
index 0d48441..7eb5ff7
--- a/ProductionTools/sfotzdb_reader.py
+++ b/ProductionTools/installed/castor.sfotzdb_reader.py
@@ -1,3 +1,4 @@
+#!/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')
diff --git a/ProductionTools/sfotzdb_writer.py b/ProductionTools/installed/castor.sfotzdb_writer.py
old mode 100644
new mode 100755
similarity index 70%
rename from ProductionTools/sfotzdb_writer.py
rename to ProductionTools/installed/castor.sfotzdb_writer.py
index 2bb7f6e..d4c4717
--- a/ProductionTools/sfotzdb_writer.py
+++ b/ProductionTools/installed/castor.sfotzdb_writer.py
@@ -1,3 +1,4 @@
+#!/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')
diff --git a/ProductionTools/installed/castor.signal b/ProductionTools/installed/castor.signal
new file mode 100755
index 0000000..e768e1e
--- /dev/null
+++ b/ProductionTools/installed/castor.signal
@@ -0,0 +1,9 @@
+#!/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 $pid
+fi
diff --git a/ProductionTools/castor.starttime b/ProductionTools/installed/castor.starttime
similarity index 100%
rename from ProductionTools/castor.starttime
rename to ProductionTools/installed/castor.starttime
diff --git a/ProductionTools/castor.stopall b/ProductionTools/installed/castor.stopall
similarity index 56%
rename from ProductionTools/castor.stopall
rename to ProductionTools/installed/castor.stopall
index 74ba118..dda4758 100755
--- a/ProductionTools/castor.stopall
+++ b/ProductionTools/installed/castor.stopall
@@ -1,5 +1,5 @@
#!/usr/bin/env bash
for c in `cat castor.machines`; do
echo "$c"
- ssh -x $c "~flegoff/castor_tools/signal-script.sh 12"
+ ssh -x $c "/sw/castor/tools/castor.signal 12"
done
diff --git a/ProductionTools/generate_instance_lists.sh b/ProductionTools/installed/castor.updateconflist
similarity index 100%
rename from ProductionTools/generate_instance_lists.sh
rename to ProductionTools/installed/castor.updateconflist
diff --git a/ProductionTools/oracle_db_disaster_recovery/oracle_db_disaster_recovery.py b/ProductionTools/oracle_db_disaster_recovery/oracle_db_disaster_recovery.py
index 1afb68d..7878d3c 100644
--- a/ProductionTools/oracle_db_disaster_recovery/oracle_db_disaster_recovery.py
+++ b/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'):
diff --git a/point1.deployment.sh b/point1.deployment.sh
index c6064fe..c149139 100755
--- a/point1.deployment.sh
+++ b/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"
--
GitLab