Hi all!
I've just added some code to OpenBSC which keeps counters for various events, such as the number of channel requests, how often we were out of channels, the number of delivered/submitted sms, paging statistics, etc.
Those counters are available through the VTY for evaluation at runtime.
There is now a "Counters" table in our database schema. It stores those counters in a generic way. The counters are stored to the database every minute. Each counter is identified by a string like "net.sms.delivered"
The counters increment all the time, until OpenBSC terminates, when they will re-start from zero. I did not want to make the counters persistent (i.e. load them from the database) as it would cofuse users of the VTY.
What's still missing now is some kind of script that uses those counters to generate something like nice charts. Given the way how the counters are stored in the database, it should be relatively easy to write a fairly generic tool that does not need any modification even if we add new counters.
What I'm thinking is something like specifying the specific counters one is interested in at the command line, maybe also providing a start and end time.
The graphs should then generally show the increments of the counters between two counter stores in the database. So if e.g. one minute ago the number of delivered SMS was '10', and now it is '15, then the actual value plotted in the graph should only be '5'.
The only "difficulty" is to handle restarts of OpenBSC... but that's quite easy. All counters are monotonically incrementing, i.e. whenever you see a smaller counter, you can safely assume that openbsc was restarted and that we simply continue with the next diff, ignoring the negative one.
If anyone is interested in working in such a tool (preferrably in python or perl, using their respective sqlite3 bindings), it would be greatly appreciated.
However, please do coordinate on this mailing list... your time is too precious to be wasted by code duplication.
Thanks in advance, Harald
Is someone already working on this ?
It should be pretty straightforward to pipe this into a RRD database (rrdtool) and from there, do pretty much any graph you want ... It will handle wrap around / reset to 0 / differential all by itself.
I'll have a look at this tomorrow if nobody manifests itself as already working on it.
Sylvain
On Thu, Dec 24, 2009 at 01:11:46AM +0100, Sylvain Munaut wrote:
Is someone already working on this ?
not that I'm aware of.
It should be pretty straightforward to pipe this into a RRD database (rrdtool) and from there, do pretty much any graph you want ... It will handle wrap around / reset to 0 / differential all by itself.
great idea.
I'll have a look at this tomorrow if nobody manifests itself as already working on it.
thanks. But please, SMS stuff has higher priority. You can still work on the rrdtool integration after you've arrived at the event..
Hi,
I also thought about RRD, I'm writing on a small python script that gets the new values from the db (since the last update of the target rrd file) and updates the rrd file.
At the moment it only supports one data source for a single rrd file, but I'll add commandline options for the file and counter name. I think this is the easiest solution, adding a new counter is just creating a new rrd file and adding a cronjob.
It would be great if someone could post some sample data to do some tests.
cu, Clemens
Am Donnerstag 24 Dezember 2009 09:27:49 schrieb Harald Welte:
On Thu, Dec 24, 2009 at 01:11:46AM +0100, Sylvain Munaut wrote:
Is someone already working on this ?
not that I'm aware of.
It should be pretty straightforward to pipe this into a RRD database (rrdtool) and from there, do pretty much any graph you want ... It will handle wrap around / reset to 0 / differential all by itself.
great idea.
I'll have a look at this tomorrow if nobody manifests itself as already working on it.
thanks. But please, SMS stuff has higher priority. You can still work on the rrdtool integration after you've arrived at the event..
Hi Clemens,
On Fri, Dec 25, 2009 at 01:41:06AM +0100, Clemens Hopfer wrote:
At the moment it only supports one data source for a single rrd file, but I'll add commandline options for the file and counter name. I think this is the easiest solution, adding a new counter is just creating a new rrd file and adding a cronjob.
well, it's also easy to generate a list of the counters from the sql database by using something like "select unique(name) from counters" and then iterate over that list...
It would be great if someone could post some sample data to do some tests.
I cannot send you the full hlr.sqlite3 of my test machine, sorry. but I've attached the output of 'sqlite3 -csv hlr.sqlite3 "select * from counters;"'. The counter values are very low as I've been playing around with few phones and few events, but it should give you a feeling for it.
btw: Will you be at 26C3?
Am Freitag 25 Dezember 2009 03:07:34 schrieb Harald Welte:
Hi Clemens,
On Fri, Dec 25, 2009 at 01:41:06AM +0100, Clemens Hopfer wrote:
At the moment it only supports one data source for a single rrd file, but I'll add commandline options for the file and counter name. I think this is the easiest solution, adding a new counter is just creating a new rrd file and adding a cronjob.
well, it's also easy to generate a list of the counters from the sql database by using something like "select unique(name) from counters" and then iterate over that list...
hmm, I'm not very familiar with rrdtool, but afaics there is no direct option in rrdtool to add a data source after creating a rrd file. However the Perl RRD bindings actually got a function to do that and there are several suggestions to dump the rrd file to xml, change that file and recreate the rrd file but I'm not sure if this is a really good idea... That's one reason (apart from simplicity) why i only used one data source for now. It would be also possible to use a dedicated directory and create RRDs of new counters on-the-fly.
I'll have a look at the Perl api, although I don't like Perl that much... ;-)
It would be great if someone could post some sample data to do some tests.
I cannot send you the full hlr.sqlite3 of my test machine, sorry. but I've attached the output of 'sqlite3 -csv hlr.sqlite3 "select * from counters;"'. The counter values are very low as I've been playing around with few phones and few events, but it should give you a feeling for it.
thanks, that's everything I wanted :-)
btw: Will you be at 26C3?
yep, I'll arrive at Saturday afternoon.
Am Freitag 25 Dezember 2009 03:56:37 schrieb Clemens Hopfer:
[..] hmm, I'm not very familiar with rrdtool, but afaics there is no direct option in rrdtool to add a data source after creating a rrd file. However the Perl RRD bindings actually got a function to do that and there are several suggestions to dump the rrd file to xml, change that file and recreate the rrd file but I'm not sure if this is a really good idea...
ok, the perl api is also just dumping the whole rrd file to xml, changing it and recreating the rrd file. The xml file seems to be about the factor 10 bigger than the rrd file, i expected 100 or more. I'll have a deeper look into the xml structure tomorrow.
Clemens
Hi,
there is one problem with the sqlite db: Afaik sqlite3 doesn't really know about datatypes apart from the standard types (like integer, text etc..) and uses a dynamic type system.
So the query for inserting a new counter value to the database does not insert the timestamp as unix time as expected but as string ( db.c:1037 datetime('now') ).
Something like strftime('%s','now') would make much more sense, rrd also uses unix time for updating rrd files.
Am Freitag 25 Dezember 2009 03:07:34 schrieb Harald Welte:
Hi Clemens,
On Fri, Dec 25, 2009 at 01:41:06AM +0100, Clemens Hopfer wrote:
At the moment it only supports one data source for a single rrd file, but I'll add commandline options for the file and counter name. I think this is the easiest solution, adding a new counter is just creating a new rrd file and adding a cronjob.
well, it's also easy to generate a list of the counters from the sql database by using something like "select unique(name) from counters" and then iterate over that list...
It would be great if someone could post some sample data to do some tests.
I cannot send you the full hlr.sqlite3 of my test machine, sorry. but I've attached the output of 'sqlite3 -csv hlr.sqlite3 "select * from counters;"'. The counter values are very low as I've been playing around with few phones and few events, but it should give you a feeling for it.
btw: Will you be at 26C3?
-
Clemens Hopfer -
Harald Welte -
Sylvain Munaut