I know sar was somewhat challenging to implement here.

First, the lowest interval I could achieve was 2 seconds (but thats just my craziness) so this would apply even to 5 minute increments. One thing I would love by the way is still to be able to customize the

One note: I had to change the perms on sar to allow hobbit to run it (/usr/lib/sa/sa1 and sa2 on AIX)

The first piece was a client script like this:

while read aline; do CPUNUM=echo $aline | awk '{print $1}' PUSR=echo $aline | awk '{print $2}' PSYS=echo $aline | awk '{print $3}' PWIO=echo $aline | awk '{print $4}' PIDL=echo $aline | awk '{print $5}'

echo "cpu"$CPUNUM"pcntusr: $PUSR" >> /tmp/hobbit_sar"$CPUNUM".msg echo "cpu"$CPUNUM"pcntsys: $PSYS" >> /tmp/hobbit_sar"$CPUNUM".msg echo "cpu"$CPUNUM"pcntwio: $PWIO" >> /tmp/hobbit_sar"$CPUNUM".msg echo "cpu"$CPUNUM"pcntidl : $PIDL" >> /tmp/hobbit_sar"$CPUNUM".msg

$BB $BBDISP "status $MACHINE.sar green date cat /tmp/hobbit_sar"$CPUNUM".msg " rm /tmp/hobbit_sar"$CPUNUM".msg done < /tmp/hobbit_sar.tmp rm /tmp/hobbit_sar.tmp

The second piece was the server side script:

if [ "$TESTNAME" = "sar" ] ; then

     NUM=`cat $FNAME | grep cpu | head -1 | cut -c 4`
     if [ $NUM -eq 1 ] ; then
        NEXT=`cat $FNAME | grep cpu | head -1 | cut -c 5`
        if [ $NEXT != "p" ] ; then
           NUM=`cat $FNAME | grep cpu | head -1 | cut -c 4-5`
        fi
     fi

     USR=`grep pcntusr $FNAME | awk -F: '{print $2}'`
     SYS=`grep pcntsys $FNAME | awk -F: '{print $2}'`
     WIO=`grep pcntwio $FNAME | awk -F: '{print $2}'`
     IDL=`grep pcntidl $FNAME | awk -F: '{print $2}'`
     echo "DS:cpu"$NUM"pcntusr:GAUGE:3:0:U"
     echo "DS:cpu"$NUM"pcntsys:GAUGE:3:0:U"
     echo "DS:cpu"$NUM"pcntwio:GAUGE:3:0:U"
     echo "DS:cpu"$NUM"pcntidl:GAUGE:3:0:U"
     if [ $NUM -eq 0 ] ; then
        NUM=00
     fi
     if [ $NUM -eq 1 ] ; then
        NUM=01
     fi
     if [ $NUM -eq 2 ] ; then
        NUM=02
     fi
     if [ $NUM -eq 3 ] ; then
        NUM=03
     fi
     if [ $NUM -eq 4 ] ; then
        NUM=04
     fi
     if [ $NUM -eq 5 ] ; then
        NUM=05
     fi
     if [ $NUM -eq 6 ] ; then
        NUM=06
     fi
     if [ $NUM -eq 7 ] ; then
        NUM=07
     fi
     if [ $NUM -eq 8 ] ; then
        NUM=08
     fi
     if [ $NUM -eq 9 ] ; then
        NUM=09
     fi
     echo "sar,cpu"$NUM".rrd"
     echo "$USR:$SYS:$WIO:$IDL"

fi

The reason for the number conversion had to do with the way the charts appeared. sar,cpu1.rrd, sar,cpu10.rrd acts differently than sar,cpu01.rrd and sar,cpu10.

Then i needed a hobbit graph definition:

    FNPATTERN sar,(.*).rrd
    TITLE CPU sar
    YAXIS %
    DEF:usr at RRDIDX@=@RRDFN@:cpu at RRDIDX@pcntusr:AVERAGE
    DEF:sys at RRDIDX@=@RRDFN@:cpu at RRDIDX@pcntsys:AVERAGE
    DEF:wio at RRDIDX@=@RRDFN@:cpu at RRDIDX@pcntwio:AVERAGE
    DEF:idl at RRDIDX@=@RRDFN@:cpu at RRDIDX@pcntidl:AVERAGE
    -u 100
    -l 0
    -r
    -E
    -Y
    AREA:usr at RRDIDX@#0000FF:cpu at RRDIDX@pcntusr
    GPRINT:usr at RRDIDX@:LAST: \: %5.1lf (cur)
    GPRINT:usr at RRDIDX@:MAX: \: %5.1lf (max)
    GPRINT:usr at RRDIDX@:MIN: \: %5.1lf (min)
    GPRINT:usr at RRDIDX@:AVERAGE: \: %5.1lf (avg)\n
    STACK:sys at RRDIDX@#FF0000:cpu at RRDIDX@pcntsys
    GPRINT:sys at RRDIDX@:LAST: \: %5.1lf (cur)
    GPRINT:sys at RRDIDX@:MAX: \: %5.1lf (max)
    GPRINT:sys at RRDIDX@:MIN: \: %5.1lf (min)
    GPRINT:sys at RRDIDX@:AVERAGE: \: %5.1lf (avg)\n
    STACK:wio at RRDIDX@#800080:cpu at RRDIDX@pcntwio
    GPRINT:wio at RRDIDX@:LAST: \: %5.1lf (cur)
    GPRINT:wio at RRDIDX@:MAX: \: %5.1lf (max)
    GPRINT:wio at RRDIDX@:MIN: \: %5.1lf (min)
    GPRINT:wio at RRDIDX@:AVERAGE: \: %5.1lf (avg)\n
    STACK:idl at RRDIDX@#008000:cpu at RRDIDX@pcntidl
    GPRINT:idl at RRDIDX@:LAST: \: %5.1lf (cur)
    GPRINT:idl at RRDIDX@:MAX: \: %5.1lf (max)
    GPRINT:idl at RRDIDX@:MIN: \: %5.1lf (min)
    GPRINT:idl at RRDIDX@:AVERAGE: \: %5.1lf (avg)\n

Then I added "sar" to TEST2RRD in hobbitserver.cfg, as well as: "sar::1" in the graphs.

Because I wanted things at a certain interval, I hand-created the rrd's for each server (depending on the number of CPU's):

/usr/local/rrdtool/bin/rrdtool create sar,cpu00.rrd --step 2
DS:cpu0pcntwio:GAUGE:3:0:U
DS:cpu0pcntsys:GAUGE:3:0:U
DS:cpu0pcntusr:GAUGE:3:0:U
DS:cpu0pcntidl:GAUGE:3:0:U
RRA:AVERAGE:0.5:1:86400
RRA:AVERAGE:0.5:6:86400
RRA:AVERAGE:0.5:24:86400
RRA:AVERAGE:0.5:288:86400 <etc etc etc... for each cpu>

That's what I had to do to get sar stuff working here. Of course, I probably did it some wacky way because of the --step I needed.

-Jeff

On 7/18/06, Henrik Stoerner <henrik at hswn.dk> wrote:

On Mon, Jul 17, 2006 at 09:48:17PM -0500, Craig Cook wrote:

...

I have noticed in 4.2 RC there are iostat commands (hobbitclient-sunos.sh -> [iostatcpu] & [iostatdisk])

But there is no definition listed in hobbitgraph.cfg...

Is this a "work-in-progress"?

Yes. I added it to the client scripts because that way it would be easy to use once I get the server-end adapted to use it.

Doing so will require some new logic in Hobbit to decide which of the many sources of load-measurements to use - e.g. vmstat or sar data could provide some of those same data. We need to rethink how to collect and present these data.

Regards, Henrik

To unsubscribe from the hobbit list, send an e-mail to hobbit-unsubscribe at hswn.dk

Just a request (not for 4.2 that is in RC so no new feature i think :P) .... It will be possible to get a inode column for systems with the same logic that there's for disk check? We are having some problems witha filesystem that at the moment get out of inode while having a good chunk of space and then monitoring of inode shuld not be too long to implement in the client/server side but could be a nice feature too have.

Thanks Francesco Duranti

Frédéric Mangeant a écrit :

I second this : on a few systems (like the Hobbit server itself), I install bb-inode.sh from Deadcat, but it needs a 'df' command which supports inode reporting.

For systems that tend to use a lot of inodes (mail server, hobbit), it's best to use something like reiserfs, that as no inodes at all.

But the feature would be nice (I needed it on my BB server).

-- Charles Goyard - cgoyard at cvf.fr - (+33) 1 45 38 01 31