On Mon 23 Jul 2007, Phil Mayers wrote: > On Mon, 2007-07-23 at 10:11 -0500, Hugh Messenger wrote: > > While we are on the subject of schemas (schemi?) ... > > > > Would it be worth adding some indexing to the basic sqlippool table? At > > the moment, only the 'id' is indexed, by virtue of being the primary > > key. > > > > I should imagine that even a medium sized provider could end up with > > quite large radippool tables, at a thousand or so entries for every four > > class C's. I'm just a corner shop ISP, but even I'll end up with about > > 10k entries by the time I'm done migrating everything from Funk. > > > > With 3 queries per login, 1 each for every stop, start and update, plus > > whatever work the backend provisioning is doing to check and update the > > pools, that's a lot of extra work for the db with no indexing. > > > > Also, unless there is some bizarre circumstance I haven't thought of > > where the same IP would appear twice, make FrameIPAddress a 'unique' > > index, to prevent accidentally adding the same IP twice when > > provisioning the table. > > VRFs with overlapping address space are one possible use I can think of; > handing out special IPs which redirect to "you are banned" or null route > is another. > > Don't make it unique. Do index it.
I disagree. It DOES need to be unique and indexed. In the case where you are virtualising things with VRF (Which I do in production) you need to add an extra VRF key to the queries (In my case I use calledstationid) and you would adjust the index/constraints accordingly. -- Peter Nixon http://peternixon.net/ - List info/subscribe/unsubscribe? See http://www.freeradius.org/list/users.html
