https://issues.apache.org/jira/browse/AMQ-3881 - handy diag script to gather info about the host machine and broker java process and combine into a compressed log, some users have found it valuable in production

git-svn-id: https://svn.apache.org/repos/asf/activemq/trunk@1355055 13f79535-47bb-0310-9956-ffa450edef68
This commit is contained in:
Gary Tully 2012-06-28 15:51:52 +00:00
parent e900fb41db
commit 87b2920927
2 changed files with 221 additions and 0 deletions

View File

@ -68,6 +68,7 @@
<outputDirectory></outputDirectory>
<includes>
<include>bin/activemq</include>
<include>bin/diag</include>
<include>bin/activemq-admin</include>
<include>bin/macosx/activemq</include>
<include>bin/linux-*/activemq</include>

220
assembly/src/release/bin/diag Executable file
View File

@ -0,0 +1,220 @@
#!/bin/sh
#
# diag.sh - a script to run diagnostic commands and save output to a log file
#
# Author: Cal Lott <calvin.lott@ihg.com>
#
PATH=$PATH:$HOME/bin:/bin/:/usr/bin:/sbin:/usr/sbin
# Set a script name string for later use.
SCRIPT_NAME=`basename $0`
ORIGINAL_COMMAND_LINE="$0 $*"
LOCALHOST=`uname -n | awk -F. '{print $1}'`
# Declare some functions.
# Prints argument to stdout, prepended with "[YYYYMMDD-HH:MM:SS]".
dateprint() {
echo "[`date +%Y%m%d-%H:%M:%S`] $*"
}
# Runs argument list as a single shell command.
BAR_SEPARATOR='------------------------------------------------'
run() {
echo
echo $BAR_SEPARATOR
dateprint "START: $*"
executable=`echo $* | awk '{print $1}'`
dateprint "path to executable: `which $executable 2>/dev/null`"
$*
last_exit_code=$?
cumulative_exit_code=`expr $cumulative_exit_code + $last_exit_code`
dateprint "exit $last_exit_code"
dateprint "END: $*"
echo $BAR_SEPARATOR
echo
}
# Print usage information.
usage() {
cat << EOF
$SCRIPT_NAME - run a set of standard diagnostic commands
Usage: $SCRIPT_NAME [-l logfile] [-jz] [-hv] <pid1> <pid2> ... <pidN>
This script runs several diagnostic commands on the local system and saves the
output to a gzipped log file. You may also specify an optional list of pids,
which will cause additional diagnostic commands (lsoft, jstat, etc.) to be run
against those specific processes only.
The default path for the datestamped log file is as follows:
\$HOME/log/$SCRIPT_NAME.\$LOCALHOST.YYYYMMDD-HHMMSS.log
You can override the default file name with the -l option. (See below.)
Options:
-j Also process any pids listed in output of "jps" command.
-l <logfile> Use specified log file instead of default.
-z Do not gzip log file.
Other Options:
-h Print this message and exit.
EOF
}
# Print help information.
help() {
usage
}
# Command line options processing.
while getopts "hjl:vz" OPTION
do
case $OPTION in
h)
help
exit 0
;;
j)
USE_JPS='TRUE'
;;
l)
LOGFILE=$OPTARG
;;
z)
COMPRESSION_OPTION='DISABLED'
;;
?)
usage
exit 1
;;
esac
done
shift `expr $OPTIND - 1` # remove processed arguments from $*
PIDLIST=$* # save list of remaining args as pidlist
# If requested, use jps to get an additional list of java pids.
if [ "$USE_JPS" = 'TRUE' ]
then
# Save the current list of pids.
arg_pids=$PIDLIST
# Use jps to grab additional java pids, filtering the jps pid itself.
jps_pids=`jps 2>/dev/null | awk '$2 !~ /^Jps/ {print $1}'`
# Now obtain an unique sorted list of pids from both groups.
PIDLIST=`echo $arg_pids $jps_pids | sort | uniq`
fi
# Use the default logfile if the user did not specify one.
if [ "$LOGFILE" = '' ]
then
LOGFILE="$HOME/log/$SCRIPT_NAME.$LOCALHOST.`date +%Y%m%d-%H%M%S`.log"
fi
# Create log directory if it does not already exist
mkdir `dirname $LOGFILE` >/dev/null 2>&1
# Print start message to stdout.
dateprint "BEGIN - running diagnostics for host: `uname -n`"
# Complain and die if we cannot create the logfile.
touch $LOGFILE >/dev/null 2>&1
if [ $? -eq 0 ]
then
dateprint "logfile created, running diagnostic commands . . ."
else
dateprint "ERROR: cannot create $LOGFILE, exiting!"
exit 1
fi
# Save stdout and stderr descriptors and redirect output to logfile.
exec 6>&1 # Link file descriptor 6 with stdout (saves stdout)
exec 7>&2 # Link file descriptor 7 with stderr (saves stderr)
exec >$LOGFILE 2>&1 # Redirect stdout and stdin to logfile
##
## Begin diagnostic output
##
dateprint "BEGIN DIAGNOSTIC OUTPUT"
dateprint "running on host: `uname -n`"
dateprint "unprocessed command line: \"$ORIGINAL_COMMAND_LINE\""
dateprint "after options processing: \"$0 $PIDLIST > $LOGFILE\""
# Each of the following shell commands will be run once.
for cmd in \
"crontab -l" \
"id" \
"ifconfig -a" \
"df -kl" \
"env" \
"iostat -x" \
"jps -v" \
"netstat -anp" \
"netstat -r" \
"ps -ef" \
"sar -A" \
"ulimit -a" \
"uname -a" \
"uptime" \
"w" \
; do
run $cmd
done
# As above, but these commands are customized for a specific flavor of Unix.
OS=`uname -s`
case $OS in
Linux)
for cmd in \
"swapon -s" \
"top -b -n1" \
; do
run $cmd
done
;;
SunOS)
for cmd in \
"swap -s" \
"top -b -d1" \
; do
run $cmd
done
;;
*)
dateprint "WARNING: OS \"$OS\" unknown - running no OS-specific commands"
;;
esac
# Each of these commands are run once per each pid argument
for pid in $PIDLIST
do
for cmd in \
"lsof -p $pid" \
"jstat -gcutil $pid" \
"jstack -l $pid" \
; do
run $cmd
done
done
dateprint "END DIAGNOSTIC OUTPUT"
dateprint "cumulative exit code: $cumulative_exit_code"
exec 1>&6 6>&- # Restore stdout and close file descriptor 6.
exec 2>&7 7>&- # Restore stderr and close file descriptor 7.
##
## End diagnostic output
##
# Compress the logfile if requested and end script execution
if [ "$COMPRESSION_OPTION" = 'DISABLED' ]
then
dateprint "FINISH - output stored at: $LOGFILE"
else
nohup gzip $LOGFILE >/dev/null 2>&1 &
dateprint "FINISH - output stored at: $LOGFILE.gz"
fi
exit $cumulative_exit_code