#! /bin/sh
# PCP QA Test No. 023
# Test PMCD's PMDA timeout behaviour
#
# Copyright (c) 1995-2002 Silicon Graphics, Inc.  All Rights Reserved.
#

seq=`basename $0`
echo "QA output created by $seq"

# get standard filters
. ./common.product
. ./common.filter
. ./common.check

_get_libpcp_config
rm -f $seq.out
if [ "$ipv6" = "true" ]; then
    ln $seq.out.2 $seq.out || exit 1
else
    ln $seq.out.1 $seq.out || exit 1
fi

BINFMT=unknown-obj
which pmobjstyle >/dev/null 2>&1 && BINFMT=`pmobjstyle`

_filter_log()
{
    sleep 3
    _filter_pmcd_log <./pmcd.log \
    | sed \
	-e '/^DATA: from client/d' \
	-e 's/Cannot open 000000660066: No such file or directory//' \
	-e 's/^$//' \
	-e '/^00[08]:/d' \
	-e '/started PMDA ora_/d' \
	-e '/cmd=\/usr\/pcp\/lib\/pmdaoracle7/d' \
	-e '/started PMDA watch/d' \
	-e '/cmd=\/usr\/pcp\/lib\/pmdawatch/d' \
	-e '/pmGetPDU/{
s/\[[0-9][0-9]*]/[PID]/
s/from=.*/from=PID/
}' \
	-e '/_pmHaveMoreInput/{
s/\[[0-9][0-9]*]/[PID]/
s/0x[^ ]*\([^ ][^ ][^ ]\)/ADDR...\1/
}' \
	-e '/Piggy-back/{
s/0x[^ ]*\([^ ][^ ][^ ]\)/ADDR...\1/
s/from=.*/from=PID/
}' \
	-e '/pmXmitPDU/s/\[[0-9][0-9]*]/[PID]/' \
	-e '/pmResult/s/ .* numpmid/ ... numpmid/' \
	-e '/value /{
s/value [0-9][0-9]*/value INTEGER/
}' \
	-e "s;$PCP_PMCDCONF_PATH;\$PCP_PMCDCONF_PATH;g" \
	-e 's#/usr/pcp/lib/mips#mips#' \
	-e 's#/usr/lib/pcp/mips#mips#' \
	-e "s/$BINFMT/ISA/g" \
	-e '/access violation from host/d' \
	-e '/endclient client.* No permission/d' \
	-e 's/ while attempting to read 12 bytes out of 12 in .*on fd/ on fd/' \
	-e 's/fd=[0-9][0-9]*/fd=FD/g' \
	-e "s;$here/src/;;" \
	-e '/UNIX_DOMAIN_SOCKET/d' \
	-e '/unix:/d' \
    | $PCP_AWK_PROG '
$3 ~ /^[0-9][0-9]*$/	{ $3 = "A_PID"; }
/Host access list:/	{ localHostPrinted = 0 }
$1 == "y" && $2 = "y" && $8 == "localhost" {
			  if (! localHostPrinted) {
				localHostPrinted = 1
				$5 = "IP-addr-in-hex"; $6 = "Host-mask"
			  }
			  else next
			}
/pipe cmd=/		{ $4 = "FD"; $5 = "FD" }
			{ print	}' \
    | sed -e '/^pmcd 2 A_PID /s/ A_PID .*/ A_PID .../'
}

CONFIG=$PCP_PMCDCONF_PATH
TMP_CONFIG=$tmp/conf.tmp
SAVE_CONFIG=$tmp/pmcd.conf.save
[ -z "$PCP_PMLOGGERCONTROL_PATH" ] && \
	PCP_PMLOGGERCONTROL_PATH="$PCP_SYSCONF_DIR/pmlogger/control"
LOGCONTROL="$PCP_PMLOGGERCONTROL_PATH"
SAVE_LOGCONTROL=$tmp/control.save
signal=$PCP_BINADM_DIR/pmsignal
_needclean=true

rm -rf $tmp $tmp.*
mkdir $tmp
chmod ugo+rwx $tmp
cd $tmp

SAVE_LOGGER=`_get_config pmlogger`
[ -f $LOGCONTROL ] && $sudo mv $LOGCONTROL $SAVE_LOGCONTROL

cleanup()
{
    echo >>$here/$seq.full
    echo "pmcd.log at cleanup" >>$here/$seq.full
    cat pmcd.log >>$here/$seq.full
    cd $here
    if $_needclean
    then
	_needclean=false
	$sudo $signal -a -s TERM pmcd

        [ -f $SAVE_CONFIG ] && $sudo mv $SAVE_CONFIG $CONFIG
	echo "=== restored $CONFIG ===" >>$here/$seq.full
	cat $CONFIG >>$here/$seq.full
	[ -f $SAVE_LOGCONTROL ] && $sudo mv $SAVE_LOGCONTROL $LOGCONTROL
	echo "=== restored $LOGCONTROL ===" >>$here/$seq.full
	cat $LOGCONTROL >>$here/$seq.full
	[ ! -z "$SAVE_LOGGER" ] && _change_config pmlogger $SAVE_LOGGER
	echo "Restart and ping pmcd ..."
	unset PMCD_PORT # don't worry about preserving just get rid of it
	unset PMCD_SOCKET
	$sudo $PCP_RC_DIR/pcp restart \
	| tee -a $here/$seq.full \
	| _filterall_pcp_start \
	| sed -e '/Waiting for pmcd to terminate/d'
	_wait_for_pmcd
	_wait_for_pmlogger
	pmprobe pmcd.control.debug
	sleepers=`ps $PCP_PS_ALL_FLAGS | grep '[d]umb_pmda' | $PCP_AWK_PROG '$3 == 1 { print $2 }'`
	[ ! -z "$sleepers" ] && $sudo $signal -s TERM $sleepers
    fi
    rm -rf $tmp $tmp.*
}

trap "cleanup; exit \$status" 0 1 2 3 15

rm -f $here/$seq.full

# real QA test starts here

# Get a domain number for the 'fake_kernel' pmda
FAKE_DOM=`pminfo -m hinv.ncpu | $PCP_AWK_PROG '{split ($3,d,"."); print d[1];}'`

# pick a tcp port that is not in use
#
port=`_get_port tcp 4340 4350`
if [ -z "$port" ]
then
    echo "Arrgh ... no free TCP port in the range 4340 ... 4350"
    exit 1
fi

$sudo $PCP_RC_DIR/pcp stop | _filter_pcp_stop
_change_config pmlogger off
$sudo $signal -a -s TERM pmie 2>/dev/null
$sudo $signal -a -s TERM pmlogger 2>/dev/null

# Create a fake namespace to force pminfo to query the fake agent
FAKE_NS=$tmp/$seq.fake_ns

echo 'root { hinv } ' > $FAKE_NS
echo 'hinv { ncpu 1:0:1 }' >> $FAKE_NS

# Make our own version of pmcd.conf with dummy agents that will time out.
# Copy the pmcd PMDA from the original pmcd.conf so that we can check pmcd's
# timeouts by storing into pmcd.control.timeout
# NOTE: none of the domains should clash with the pmcd PMDA (domain 2) or the
# sample agent (domain 254).  These agents will be appended to the file.
#
cat <<EOF >$TMP_CONFIG
# Created by QA $seq
fake_irix 1 pipe binary $here/src/dumb_pmda -d 1 fake_irix
fake_proc 3 pipe binary $here/src/dumb_pmda -d 3 fake_proc
fake_cisco 5 pipe binary $here/src/dumb_pmda -d 5 fake_cisco
fake_six 6 pipe binary $here/src/dumb_pmda -d 6 fake_six
EOF
sed -n <$CONFIG >>$TMP_CONFIG \
    -e '/^#/d' \
    -e '/pmda_pmcd/s/[ 	][ 	]*/ /gp' \
    -e '/pmdapmcd/s/[ 	][ 	]*/ /gp'
cat <<EOF >>$TMP_CONFIG
[access]
allow localhost : all;
disallow * : all;
EOF

$sudo mv $CONFIG $SAVE_CONFIG
$sudo mv $TMP_CONFIG $CONFIG

# very odd inode update problem on kenj-pc with one variant of the Linux
# 2.6.11 kernel ... this should be 100% benign for everyone else
#
sync

echo '$PCP_PMCDCONF_PATH contains:' | sed -e "s;$PCP_VAR_DIR;\$PCP_VAR_DIR;"
echo '<BEGIN>'
sed <$CONFIG \
    -e '/^pmcd 2 /s/2 .*/2 .../' \
    -e "s;$here/src/;;"
echo '<END>'

# Note: start pmcd with -f so that its PID stays the same (no daemon)
#
PMCD_PORT=$port
PATH=$PATH:$here/src
PMCD_SOCKET=$tmp/pmcd.socket
export PMCD_PORT PATH PMCD_SOCKET
$PCP_PMCD_PROG -f -t 2 &
pmcd=$!
if [ -d $PCP_RUN_DIR ]
then
    echo "$pmcd" >$tmp.tmp
    $sudo cp $tmp.tmp $PCP_RUN_DIR/pmcd.pid
fi

# Sleep 3 sec for each fake pmda
sleep 15 
_wait_for_pmcd
_filter_log
echo "pmcd.log after pmcd first started" >>$here/$seq.full
cat pmcd.log >>$here/$seq.full

# Get pmcd to stop the fake_kernel PMDA by asking for a metric from it.
# Sleep won't do anything, so it'll be timed out returning an IPC protocol
# failure.  The next fetch for the domain will return a "No agent for
# domain..." error.
#
echo
echo "Restart dead agent test"
echo "Expect \"IPC protocol failure\" then \"No agent for domain...\":"
pminfo -n $FAKE_NS -d hinv.ncpu
pminfo -n $FAKE_NS -d hinv.ncpu

pminfo -f pmcd.agent.status
echo

# Check the functionality of pmcd.control.timeout.  agenttimeout sets the pmcd
# agenttimeout for a specified domain to a given timeout, then fetches from
# the domain, expecting the agent to timeout and finally checks the timeout
# against the elapsed time for the fetch.
# Usage: agenttimeout domain timeout
#
echo
echo "pmcd.control.timeout tests"
$here/src/agenttimeout 5 3
$here/src/agenttimeout 6 8
#
# Be nice and set it back to 2 so that subsequent timeouts happen more quickly
#
pmstore pmcd.control.timeout 2

pminfo -f pmcd.agent.status
echo

# Get PMCD to restart any deceased agents.  (The config file hasn't changed)
#
sleep 1
$signal -s HUP $pmcd

# Sleep 3 sec for each fake pmda in the config
sleep 15
_wait_for_pmcd
_filter_log
echo >>$here/$seq.full
echo "pmcd.log after pmcd SIGHUP" >>$here/$seq.full
cat pmcd.log >>$here/$seq.full

pminfo -f pmcd.agent.status
echo

# If agent not restarted then both messages will be "no agent for domain..."
#
echo "Expect \"IPC protocol failure\" then \"No PMCD agent...\":"
pminfo -n $FAKE_NS -d hinv.ncpu
pminfo -n $FAKE_NS -d hinv.ncpu

pminfo -f pmcd.agent.status
echo

status=0