#! /bin/sh
#
# check PCP QA setup
#
# Copyright (c) 1997-2002 Silicon Graphics, Inc.  All Rights Reserved.
#
# $Id$
#

tmp=/var/tmp/$$
trap "rm -f $tmp.*; exit 0" 0 1 2 3 15

_fail()
{
    echo "... check failed!"
    exit 1
}

_check_pmcd()
{
    # 5 seconds default seems like a reasonble max time to establish
    # contact
    rm -f $tmp.ok
    _can_wait=5
    _host=$1
    _i=0
    _dead=true
    while [ $_i -lt $_can_wait ]
    do
	_sts=`pmprobe -n $PCP_VAR_DIR/pmns/root_pmcd -h $_host pmcd.numclients 2>/dev/null | $PCP_AWK_PROG '{print $2}'`
	if [ "${_sts:-0}" -gt 0 ]
	then
	    # numval really > 0, we're done
	    #
	    _dead=false
	    break
	fi
	sleep 1
	_i=`expr $_i + 1`
    done
    if $_dead
    then
	:
    else
	touch $tmp.ok
    fi
}

# files in the PCP distribution
#
fail=false
for file in /etc/pcp.env
do
    if [ ! -f "$file" ]
    then
	echo "$0: Error: cannot file \"$file\" from the PCP distribution"
	fail=true
    fi
done

# files in the PCP QA distribution
#
for file in mk.localconfig common.rc common.setup common.product \
    common.check common.config
do
    if [ ! -f "$file" ]
    then
	echo "$0: Error: cannot file \"$file\" from the PCP QA distribution"
	fail=true
    fi
done

# no point proceeding if any of these files are missing
#
$fail && _fail

if [ ! -f localconfig ]
then
    # need localconfig
    #
    if ./mk.localconfig
    then
	make setup | egrep -v 'Entering|Leaving|Nothing'
    else
	echo "$0: Error: need to fix failure in mk.localconfig"
	_fail
    fi
    if [ ! -f localconfig ]
    then
	echo "$0: Error: mk.localconfig failed to create the file \"localconfig\""
	_fail
    fi
fi

# source hierarchy
#
# common.product
# 	common.setup
#		common.rc
#			/etc/pcp.env
#	localconfig
# common.check
# common.config
#	localconfig
#
. ./common.product
. ./common.check
. ./common.config

# things we care about from /etc/pcp.env
#

# files
#
for var in PCP_PMCDCONF_PATH PCP_PMCDOPTIONS_PATH PCP_MAGIC_FILE
do
    eval file=\$$var
    if [ -z "$file" ]
    then
	echo "$0: Error: variable \$$var is not set from /etc/pcp.env"
	fail=true
    else
	if [ ! -f "$file" ]
	then
	    echo "$0: Error: cannot find file \"$file\" (\$$var from /etc/pcp.env)"
	    fail=true
	fi
    fi
done

# programs
#
for var in PCP_AWK_PROG PCP_XCONFIRM_PROG
do
    eval prog=\$$var
    if [ -z "$prog" ]
    then
	echo "$0: Error: variable \$$var is not set from /etc/pcp.env"
	fail=true
    else
	if which $prog >/dev/null 2>&1
	then
	    :
	else
	    echo "$0: Error: cannot find executable program \"$prog\" (\$$var from /etc/pcp.env)"
	    fail=true
	fi
    fi
done

# directories
#
# these ones should be in /etc/pcp.conf but are not currently used in
# the QA scripts, but are tested here for completeness
#
xtra="PCP_SYSCONFIG_DIR PCP_LIB_DIR PCP_LIB_COMPAT_DIR PCP_INC_DIR PCP_MAN_DIR"
#
for var in PCP_RC_DIR PCP_PMDAS_DIR PCP_VAR_DIR PCP_DEMOS_DIR PCP_LOG_DIR \
	PCP_BIN_DIR PCP_BINADM_DIR PCP_TMP_DIR PCP_SHARE_DIR PCP_DOC_DIR \
	$xtra
do
    eval dir=\$$var
    if [ -z "$dir" ]
    then
	echo "$0: Error: variable \$$var is not set from /etc/pcp.env"
	fail=true
    else
	if [ ! -d "$dir" ]
	then
	    echo "$0: Error: cannot find directory \"$prog\" (\$$var from /etc/pcp.env)"
	    fail=true
	fi
    fi
done

# common.rc will unconditionally set these ones, else fail ... so no need to
# check
#	$PCP_PMCD_PROG
#	$PCP_PMCDLOG_PATH

# localconfig will unconditionally set these ones ... no real sane checking
# I can do here!
#	$PCP_PLATFORM	(redundant, copied from /etc/pcp.conf)
#	$PCP_VER
#	$PCP_EOE_VER
#	$PCP_EOE_NOSHIP_VER

# files from common.product
#
for var in PCP_PMCD_PATH
do
    eval file=\$$var
    if [ -z "$file" ]
    then
	echo "$0: Error: variable \$$var is not set from common.product"
	fail=true
    else
	if [ ! -f "$file" ]
	then
	    echo "$0: Error: cannot find file \"$file\" (\$$var from common.product)"
	    fail=true
	fi
    fi
done

# no point proceeding if any of these files are missing
#
$fail && _fail

# down to the common.config things ... these are the most contentious
# and the ones that need local tuning ...
#
#	$PCPQA_CLOSE_X_SERVER is set here, but is not used in the open
#	source version of the PCP QA and is hard to check, so skip that
#	one.
#
for var in PCPQA_FAR_PMCD PCPQA_HYPHEN_HOST
do
    eval host=\$$var
    if [ -z "$host" ]
    then
	echo "$0: Warning: variable \$$var is not set from common.config"
    else
	# try to contact pmcd ...
	#
	echo -n "Contacting pmcd at $host ... "
	_check_pmcd $host
	if [ -f $tmp.ok ]
	then
	    echo "OK"
	else
	    echo "no response (warning)"
	fi
    fi
done

# no point proceeding if any of these variables are not set
#
$fail && _fail

# the other pmcd's we think we can talk to ...
#
make qa_hosts >/dev/null 2>&1
if [ ! -f qa_hosts ]
then
    echo "$0: Error: make failed to create the file \"qa_hosts\""
    _fail
fi

# failures here are not fatal ... some of the hosts may be off with the
# pixies
#
echo "hullo world" >$tmp.hullo
for host in `cat qa_hosts`
do
    # try to contact remote pmcd ...
    #
    echo -n "Contacting remote pmcd at $host ... "
    _check_pmcd $host
    if [ -f $tmp.ok ]
    then
	echo "OK"
	# if the host is there, then the login for pcpqa needs to be
	# setup and and rsh/rcp working without passwords
	echo -n "Trying rcp/rsh for login pcpqa at $host ... "
	echo "== rsh 1st rm ==" >$tmp.out
	rsh -l pcpqa $host rm -f chk.hullo >>$tmp.out 2>&1
	echo "== rcp -> $host ==" >>$tmp.out
	rcp $tmp.hullo pcpqa@$host:chk.hullo >>$tmp.out 2>&1
	echo "== rsh cat ==" >>$tmp.out
	rsh -l pcpqa $host cat chk.hullo >$tmp.ans 2>>$tmp.out
	echo "== rsh 2nd rm ==" >>$tmp.out
	rsh -l pcpqa $host rm -f chk.hullo >>$tmp.out 2>&1
	if [ "`cat $tmp.ans`" = "`cat $tmp.hullo`" ]
	then
	    echo "OK"
	else
	    echo "failed"
	    fail=true
	    echo "Expected output: \"`cat $tmp.hullo`\""
	    echo "Actual output: \"`cat $tmp.ans`\""
	    echo "Diagnostic Log:"
	    cat $tmp.out
	fi
    else
	echo "no response (warning)"
    fi
done

# down to pmcd on the local host
#
for host in localhost `hostname` `_get_fqdn`
do
    # try to contact pmcd ...
    #
    echo -n "Contacting local pmcd at $host ... "
    _check_pmcd $host
    if [ -f $tmp.ok ]
    then
	echo "OK"
    else
	echo "no response (fatal)"
	fail=true
    fi
done

# no point proceeding if any of these cannot be contacted
#
$fail && _fail

echo "Check the sanity of these:"
echo -n "  \$PCP_PLATFORM=$PCP_PLATFORM [y] ? "
read ans
[ ! -z "$ans" -a "$ans" != y ] && _fail
echo -n "  \$PCP_VERSION=$PCP_VERSION [y] ? "
read ans
[ ! -z "$ans" -a "$ans" != y ] && _fail

exit 0