#!/bin/sh
# PCP QA Test No. 941
# pmcd health check
#
# With a single --check option, is silent except if there is a problem
# and runs just the integrity check.  This could be used from check or
# check.callback
#
# Copyright (c) 2017 Ken McDonell.  All Rights Reserved.
#

seq=`basename $0`

check=false
if [ $# -ge 1 -a "$1" = "--check" ]
then
    check=true
    # if 2nd arg present then use this as $seq ... we're being called from
    # check.callback more than likely
    #
    [ $# -ge 2 -a "$2" != "" ] && seq="$2"
fi

$check || echo "QA output created by $seq"

# get standard environment, filters and checks
. ./common.product
. ./common.filter
. ./common.check

_cleanup()
{
    cd $here
    $sudo rm -rf $tmp $tmp.*
}

status=0	# success is the default!
$check || $sudo rm -f $seq.full
$sudo rm -rf $tmp $tmp.*
trap "_cleanup; exit \$status" 0 1 2 3 15

# real QA test starts here

_get_pids_by_name pmcd >$tmp.pids
num_pmcds=`cat $tmp.pids | wc -l | sed -e 's/ //g'`
if $check
then
    echo "--- running 941 --check @ `date` ---"
    echo "_get_pids_by_name pmcd ..."
    cat $tmp.pids
    $PCP_PS_PROG $PCP_PS_ALL_FLAGS >$tmp.ps.out
    echo "$PCP_PS_PROG $PCP_PS_ALL_FLAGS | grep pmcd ..."
    grep -E '[P]PID|/[p]mcd( |$)' <$tmp.ps.out
    echo "num_pmcds=$num_pmcds (should be 1)"
fi
case "$num_pmcds"
in
    0)
    	echo "Error: no pmcd running!"
	status=1
	exit
	;;
    1)
	$check || echo "OK: one pmcd running"
    	;;
    *)
    	echo "Error: $num_pmcds pmcd processes running"
	$PCP_PS_PROG $PCP_PS_ALL_FLAGS >$tmp.ps
	for pid in `cat $tmp.pids`
	do
	    $PCP_AWK_PROG <$tmp.ps '$2 == '$pid' { print }' >$tmp.out
	    if [ -s $tmp.out ]
	    then
		grep PPID $tmp.ps
		cat $tmp.out
	    else
		echo "PID $pid has gone missing from ps(1) output?"
	    fi
	    for log in pmcd.log pmcd.log.prev
	    do
		if [ -f $PCP_LOG_DIR/pmcd/$log ]
		then
		    if grep "pmcd: PID = $pid," $PCP_LOG_DIR/pmcd/$log >/dev/null
		    then
			echo "==> start matching $PCP_LOG_DIR/pmcd/$log file ==>"
			cat $PCP_LOG_DIR/pmcd/$log
			echo "<== end matching $PCP_LOG_DIR/pmcd/$log file <=="
		    fi
		fi
	    done
	done
	status=1
	exit
	;;
esac

pid=`cat $tmp.pids`
if [ -f $PCP_LOG_DIR/pmcd/pmcd.log ]
then
    if grep "pmcd: PID = $pid," $PCP_LOG_DIR/pmcd/pmcd.log >/dev/null
    then
	$check || echo "OK: pmcd.log is for running pmcd"
    else
	echo "Error: $PCP_LOG_DIR/pmcd/pmcd.log not for running pmcd (PID=$pid)"
	echo "==> start $PCP_LOG_DIR/pmcd/pmcd.log file ==>"
	cat $PCP_LOG_DIR/pmcd/pmcd.log
	echo "<= end $PCP_LOG_DIR/pmcd/pmcd.log file <=="
	$check && status=1
    fi
else
    echo "Error: $PCP_LOG_DIR/pmcd/pmcd.log missing!"
    $check && status=1
fi

if [ -f $PCP_LOG_DIR/pmcd/pmcd.log.prev ]
then
    if grep "pmcd: PID = $pid," $PCP_LOG_DIR/pmcd/pmcd.log.prev >/dev/null
    then
	echo "Error: $PCP_LOG_DIR/pmcd/pmcd.log.prev for running pmcd (PID=$pid)"
	echo "==> start $PCP_LOG_DIR/pmcd/pmcd.log.prev file ==>"
	cat $PCP_LOG_DIR/pmcd/pmcd.log.prev
	echo "<= end $PCP_LOG_DIR/pmcd/pmcd.log.prev file <=="
	$check && status=1
    fi
fi

if [ -f $PCP_RUN_DIR/pmcd.pid ]
then
    if [ "$pid" = "`cat $PCP_RUN_DIR/pmcd.pid`" ]
    then
	$check || echo "OK: run file has pmcd's pid"
    else
	echo "Error: $PCP_RUN_DIR/pmcd.pid -> PID `cat $PCP_RUN_DIR/pmcd.pid` not pmcd (PID=$pid)"
	status=1
    fi
else
    echo "Error: $PCP_RUN_DIR/pmcd.pid missing"
    status=1
fi

marker='PCP QA test'
# Did a QA test leave behind a modified pmcd.conf file?
#
if grep "$marker" $PCP_PMCDCONF_PATH >/dev/null
then
    echo "Error: $PCP_PMCDCONF_PATH changed for QA, may not be correct"
    grep "$marker" $PCP_PMCDCONF_PATH
    status=1
else
    $check || echo "OK: PMCDCONF_PATH not from a QA test"
fi

# Did a QA test leave behind a modified pmcd.options file?
#
if grep "$marker" $PCP_PMCDOPTIONS_PATH >/dev/null
then
    echo "Error: $PCP_PMCDOPTIONS_PATH changed for QA, may not be correct"
    grep "$marker" $PCP_PMCDOPTIONS_PATH
    status=1
else
    $check || echo "OK: PMCDOPTIONS_PATH not from a QA test"
fi

# success, all done
exit
