Hi Nosy,
On Wed, Jun 27, 2012 at 5:42 AM, Marlen Caemmerer
<marlen.caemmerer(a)wikimedia.de> wrote:
I sometimes get complaints that toolserver was slow.
This regards web
applications and it is not always slow but sometimes.
I cannot always really correlate this to certain servers  being slow at the
time in question.
That is the reason I really would like to have a reference measurement
application that graphs me how long it took.
So the idea would be to have an application that is called via web server,
makes a database query on every host, records, how long it took and measures
how long the application took to complete.
The measurements could be written out to rrd files so it is easy to graph
afterwards.
Do you think it would be sufficiant to measure the time of a simple select
and an insert or would a more complex query be sensible?
Do you already a tool that does this?
What do you think about this?
For slow queries you could just use the built in mysql slow query log.
(the threshold for "slow" can be tweaked. apparently at least to
1/100th of a sec) or also add pt-query-digest and do a periodic sample
like at WMF cluster. or maybe you're interested in pt-stalk?
https://gerrit.wikimedia.org/r/gitweb?p=operations%2Fpuppet.git&a=searc…
https://ishmael.wikimedia.org/ <-- requires a labs account. you really
should get one! in the meantime any of the wikidata team can show you
what it looks like. and the code is at
https://github.com/asher/ishmael (with some relevant bits in puppet.
just search for ishmael)
http://www.percona.com/doc/percona-toolkit/2.1/pt-stalk.html
Also you could set up something like
http://lognormal.github.com/boomerang/doc/ ?
And finally you can of course use graphite.
https://graphite.wikimedia.org/ <-- also requires labs account
https://gdash.wikimedia.org/ <-- open to the world
http://www.aosabook.org/en/graphite.html <-- mediawiki has a chapter too ;)
-Jeremy