Imprecise total?

[Sylvain Beucler]

Hi,

I was trying to extract the sums (sent&received per day) from
mailgraph's monthly RRD archives, and I got different results.


I think there's a mistake in the graph's CDEF, more exactly when
values are multiplied by $step (CDEF:drecv=recv,UN,0,recv,IF,$step,*).

Step is computed on the assumption that 180 averages are used for the
graph. 180 is defined in mailgraph.cgi as
$xpoints(540)/$points_per_sample(3), that is, a graph of width 540
pixels where each value (RRA row) is 3 pixels wide.

Unfortunately the archive uses 180 averages for 35 days (not 31), as
it's created with a multiple of weekly rows:
$month_steps = $week_steps*5 (cf. /usr/sbin/mailgraph).

So the CDEF is using a $step computed with 3600*24*31/180=14880s,
while a point in the archive is really rrd.step * rra[2].pdp_per_row =
60*280 = 16800s.  So monthly displayed values are about 10% less than
the actual values from the RRA.

This is confirmed by the value shown in rrdtool xport's <step>. An
xport for 31 days will also show 160 rows instead of 180.

Note that this problem does not occur for daily sums, because 180 rows
at 8 points/row are exactly 1 day (180*8*60 = 86400 = 60*60*24).


If you want to check/confirm, here's the xport command I used:
# Use multiples to select the right RRA (see rrdfetch(1))
TIME=$(date +%s)
#RRDRES=$((60*8))   # dayly   = rrd.step * rra[0].pdp_per_row
RRDRES=$((60*280))  # monthly = rrd.step * rra[2].pdp_per_row
# DEF and CDEF are from mailgraph.cgi
rrdtool xport \
  -e $(($TIME/$RRDRES*$RRDRES)) -s e-31d \
  "DEF:sent=mailgraph.rrd:sent:AVERAGE" \
  "CDEF:dsent=sent,UN,0,sent,IF,$RRDRES,*" \
  "CDEF:ssent=PREV,UN,dsent,PREV,IF,dsent,+" \
  "DEF:recv=mailgraph.rrd:recv:AVERAGE" \
  "CDEF:rrecv=recv,60,*" \
  "CDEF:drecv=recv,UN,0,recv,IF,$RRDRES,*" \
  "CDEF:srecv=PREV,UN,drecv,PREV,IF,drecv,+" \
  "DEF:spam=mailgraph_virus.rrd:spam:AVERAGE" \
  "CDEF:dspam=spam,UN,0,spam,IF,$RRDRES,*" \
  "CDEF:sspam=PREV,UN,dspam,PREV,IF,dspam,+" \
  "DEF:virus=mailgraph_virus.rrd:virus:AVERAGE" \
  "CDEF:rvirus=virus,60,*" \
  "CDEF:dvirus=virus,UN,0,virus,IF,$RRDRES,*" \
  "CDEF:svirus=PREV,UN,dvirus,PREV,IF,dvirus,+" \
  "CDEF:recvlegit=srecv,svirus,sspam,+,-"  \
  XPORT:recvlegit:"total_received" \
  XPORT:ssent:"total_sent" \
  XPORT:dsent:"sent"


I had a look at Cacti, and it appears the sum is pre-computed before
graphing, and included in the graph as a comment (see "summation" in
http://docs.cacti.net/node/50).

Maybe mailgraph could use that technique as well?

Cheers,

--
Sylvain

--
Unsubscribe mailto:mailgraph-request@...?subject=unsubscribe
Help        mailto:mailgraph-request@...?subject=help
Archive     http://lists.ee.ethz.ch/mailgraph
WebAdmin    http://lists.ee.ethz.ch/lsg2.cgi

[David Schweikert-2]

Hi Sylvain,

On Mon, May 26, 2008 at 20:17:48 +0200, Sylvain Beucler wrote:
> I think there's a mistake in the graph's CDEF, more exactly when
> values are multiplied by $step (CDEF:drecv=recv,UN,0,recv,IF,$step,*).
> ...
> Unfortunately the archive uses 180 averages for 35 days (not 31), as
> it's created with a multiple of weekly rows:
> $month_steps = $week_steps*5 (cf. /usr/sbin/mailgraph).

You are right! Thanks for the bug report. I will fix it in the next
mailgraph.cgi in the same way as you propose it:

A straightforwarded "fix" is to change the graphs ranges:
my @graphs = (
        { title => 'Last Day',   seconds => 3600*24,        },
        { title => 'Last Week',  seconds => 3600*24*7,      },
#       { title => 'Last Month', seconds => 3600*24*31,     },
#       { title => 'Last Year',  seconds => 3600*24*365, },
        { title => 'Last Month', seconds => 3600*24*7*5,     },
        { title => 'Last Year',  seconds => 3600*24*7*5*12, },
);

Cheers
David

--
Unsubscribe mailto:mailgraph-request@...?subject=unsubscribe
Help        mailto:mailgraph-request@...?subject=help
Archive     http://lists.ee.ethz.ch/mailgraph
WebAdmin    http://lists.ee.ethz.ch/lsg2.cgi