19:31:46 <TheSnide> #startmeeting
19:31:46 <MeetBot> Meeting started Tue Nov 18 19:31:46 2014 UTC.  The chair is TheSnide. Information about MeetBot at http://wiki.debian.org/MeetBot.
19:31:46 <MeetBot> Useful Commands: #action #agreed #help #info #idea #link #topic.
19:31:49 <TheSnide> hi all
19:32:18 <TheSnide> #topic 2.1
19:33:19 <TheSnide> graph-sql is now mostly able to handle simple plugins, that don't have any cdef.
19:36:37 <TheSnide> the biggest time taker is rrd graphing.
19:37:30 <TheSnide> ... which is very sane ;-)
19:40:14 <RoyK> rrd is rather I/O intensive, though
19:40:31 <RoyK> could this be solved with a DBMS of sorts?
19:41:45 <RoyK> we're using a VM with storage on a raid-1+0 shelf and it's eating some 200iops constantly (as measured from linux)
19:44:12 <TheSnide> royk: no this is solved via rrdcached
19:44:14 <Knorrie> are you using rrdcached? it helps. http://oss.oetiker.ch/rrdtool/doc/rrdcached.en.html
19:45:16 <TheSnide> http://guide.munin-monitoring.org/en/latest/master/rrdcached.html
19:45:18 <RoyK> TheSnide: imo that's a workaround, not a solution
19:45:43 <TheSnide> using a rdms is much worse usually.
19:45:49 <TheSnide> for timeseries.
19:46:29 <RoyK> does rrd use sync writes? looks like it does...
19:47:59 <TheSnide> nope. is uqes posix advises to preload
19:48:05 <TheSnide> but sync, no.
19:48:32 <TheSnide> it uses flush, when you want to graph something
19:49:38 <TheSnide> nah, rrdcached is the best thibg here. but is difficult to install
19:50:21 <TheSnide> note that rrd offers a libdbi interface, if you *really* want to use.rdms.
19:51:11 <RoyK> has anyone tried to use a postgresql db for this thing?
19:51:18 <TheSnide> munin tried hard to avoid fsync().
19:52:30 <RoyK> or any other (R)DBMSes?
19:52:41 <TheSnide> and no. but i'm highly interested by eventual results.
19:53:10 <TheSnide> http://oss.oetiker.ch/rrdtool/doc/rrdgraph_libdbi.en.html is the api. beware that it is quite eye-bleeding.
19:53:11 <RoyK> would an object database be better, such as mongodb?
19:53:36 <TheSnide> munin cares about its data. so no mongodb ;-)
19:53:46 * RoyK is not objecting, just asking
19:54:16 <RoyK> what's wrong with mongodb?
19:54:18 <TheSnide> was joking. i'd like to try elasticsearch
19:54:33 <TheSnide> but for now, i'm focussing on rrd.
19:56:08 <TheSnide> royk, it is the php/mysql of nosql. i don't know much about it, but is mich more dev oriented than ops, i'm hearing.
19:56:42 <TheSnide> anyway, i don't want to reimplement all the cdef, vdef, ... part of rrd.
19:57:15 <RoyK> I don't touch mysql unless I have to (which I obviously have to during my dayjob)
19:57:46 <RoyK> regrettably
19:58:37 <RoyK> but I may do some testing on postgresql
19:58:55 <RoyK> it's far better on transaction-based databases (and most stuff) than mysql
20:00:26 <RoyK> I guess if munin was rewritten to use a transaction-based model, summing up all data from all sources into a single transaction with postgresql, it could be rather neat and efficient
20:01:56 <TheSnide> it does with metadata.
20:02:08 <TheSnide> with timz series, not for now.
20:02:37 <RoyK> well, the time series are the main I/O hog
20:04:08 <TheSnide> ok. that's all for me.
20:04:44 <TheSnide> #endmeeting