On Wed, Aug 8, 2012 at 4:43 AM, John Horne <user-e95f1ec2f147@xymon.invalid> wrote:
On Tue, 2012-08-07 at 09:52 +0800, Colin Coe wrote:
Hi all
Google knows a lot about $SUBJECT but unfortunately, a lot of the info
is conflicting.
What's the current wisdom for collecting and graphing iostat info on
Linux? Specifically, I'm using RHEL5 and RHEL6.
Hi,
I recently asked about monitoring iostat myself - see the thread here
http://lists.xymon.com/pipermail/xymon/2012-August/035229.html
One reply mentions some scripts on xymonton. I also received a private
reply that there were some scripts on Sourceforge - search for 'Linux
iostat'.
I ended up writing a reasonably short script myself which runs on the
clients. It records the 'avtime', 'svctime' and the percent 'busy'
values. I produce two graphs - one for the percent busy, the other shows
both time values. Both graphs appear in 'trends', and the 'busy' graph
appears on an 'iostats' test status page. (Note, because Xymon already
sort of knows about an 'iostat' test I could not use that name, so used
'iostats' instead.) The script sends a 'status' report (using 'xymon
<server> "status ...") for the 'iostats' test. If the percent busy is
high it sends a yellow status, if it is very high then a red status is
sent. Additionally, a 'data' report (using 'xymon <server> "data ...")
containing the time and busy values is sent back to the Xymon server. On
the server I run a script which is used by 'xymon_rrd' (via the
'--extra-script' and '--extra-tests' options; see the man page and look
in tasks.cfg). The script tells Xymon what RRD files to create and what
the DS entries and values are, from the data received from the client.
(Sorry, that may all sound horrendously complicated. However, it works
very well, and once set up is very easy to maintain. I currently run
three tests like this.)
I would say that monitoring iostat has proved to be useful to us. It is
not something we monitored with Big Brother, but has already shown that
we have two servers which regularly show very high disk I/O. No idea why
yet.
John.
--
John Horne, Plymouth University, UK
Tel: +XX (X)XXXX XXXXXX Fax: +XX (X)XXXX XXXXXX
Thanks John & Alan
I've now looked at a few iostat monitors for xymon including
http://sourceforge.net/projects/xy-alfanoid/files/ and
http://sourceforge.net/projects/xy-alfanoid/files/.
One of the common things I really don't like is the modification of
the TEST2RRD variable within xymonserver.cfg. My guess is this is
unavoidable but it means manually handling after xymon upgrades and I
have more than enough to do now than worry about manual steps after
software upgrades.
Henrik, is there a good way that the above can be handled?
xymonserver.cfg looks like a shell script, does it act like one? e.g.
can we do something like 'TEST2RRD=$TEST2RRD:customtest=a,b,c' in an
include file. Same for GRAPHS. Apologies if this is spelt out
somewhere and I've missed it. BTW, I really like the "*.d"
directories that have come in in very recent xymon versions.
Referring to http://lists.xymon.com/pipermail/xymon/2012-August/035231.html,
what would it take to get the iostat tests up to scratch? (I'm
selfishly referring exclusively to Linux, specifically RHEL5 and
above.)
While I'm on a bit of a whine, in the RHEL/Fedora realm, nagios for
example has heaps of plugin RPMs. I would be great if the best of
breed custom tests were assembled and packaged in a similar manner.
Lastly, this email is not intended to offend, just to try and get the
best solution for monitoring disk stats possible.
Thanks
CC
--
RHCE#805007969328369