Today, Steven Wagner wrote forth saying...

> matt massie wrote:
> > steve-
> > 
> > does that mean that the C version of gmetad works on solaris if you 
> > compile rrdtool 32-bit?
> > 
> 
> Well, at the time that I wrote that, I hadn't tried this yet*.  But I did 
> just now, and it does work (despite the fact that my 32-bit compiler's a 
> bit old).
> 
> A bit of a pain in the butt to have to compile gmetad in a totally
> different environment, though.

You should only have to compile the rrd library 32-bit right?  What 
happens when you compile the monitor-core 64-bit and link the 32-bit rrd 
library.  Does that not work?  If it worked, then you would only need to 
deal with this headache whenever you upgraded rrdtool.  To be honest 
though, I don't know about 32-bit 64-bit linking issues so this could a 
completely useless sentence to write (except maybe this last sentence).

> I'm now running it with debug output.  And it seems to be storing RRD
> values.
>
> Mostly.
> 
> ---[ one log snippet from my 2.5.0-running Solaris boxen ]---
> 
> RRD_update: expected 1 data source readings (got 2) from N:1591.600000:22:...
> save_to_rrd() couldn't parse the XML and data to RRD for [fs]
> Updated rrd /www/gmetad/rrds/fileservers/SOME_SERVER/cpu_user.rrd with 
> value N:0.1

this is an error that i don't completely understand but i'm going to 
squash this bug today (assuming it is a bug.. it might just be a 
too-generic error message problem which might read "unable to contact data 
source to get XML in the first place).  i'll find out.

> ---[and one that includes errors from both, including my 2.4.1 Linux 
> cluster ]---
> RRD_update: expected 1 data source readings (got 2) from N:1591.600000:22:...
> save_to_rrd() couldn't parse the XML and data to RRD for [fs]
> RRD_update: expected 1 data source readings (got 2) from N:1591.600000:22:...
> RRD_update: expected 1 data source readings (got 2) from N:159.000000:91:...
> save_to_rrd() couldn't parse the XML and data to RRD for [SOME_CLUSTER]

i know what is causing this bug.  in order to play nice, the new gmetad 
doesn't require a 2.5.0 gmond.  this makes it impossible on pre-2.5.0 
systems to know the SLOPE attribute.  this means that some non-numerical 
metrics slip through.. i bet one of your non-numeric metrics on a 
pre-2.5.0 gmond had a ':' in it?  please check your colon and get back 
with us.

> Of course there are screenloads of what appear to be real updates.  
> But I'm not seeing the graphs update on the frontend, which admittedly
> is running on a hacked-up 0.1.0 codebase.  My updated incarnation is
> having trouble rendering graphs.  Maybe the RRD layout has changed
> between versions?

yes.  you need to check out the latest gmetad-webfrontend from CVS.  the 
databases for the hosts are the same BUT the summary information is stored 
in a more generic and flexible manner so that we can get both the average 
and sum on demand.

something to watch for.. the web frontend assume rrdtool is installed in 
/usr/bin/rrdtool.  check your conf.php and make sure it all looks good.

> When I run basic vanilla "rrdtool dump," though, I don't see any new 
> values.  But the files *are* being written to (the timestamps are all 
> updating)... except for the summaries...

an easier way to check the data in a database is
# rrdtool fetch proc_total.rrd AVERAGE

it'll just print the timestamp and the values.

-matt


Reply via email to