SQL Query question

[tnt at kalik.net]

>I don't use authentication.

I hope this means "I don't use radius authentication."

>I'll give you the full description of my system and maybe you think of a
>solution without modifying the source code:
>A client sends only Accounting Start Requests(the same request for the 2
>servers) to my 2 radius servers (geographically separated). The requests are
>send to separate DBs.

OK. You are independantly processing the same information by two radius
servers storing data in different databases in parallel. No interaction
whatsoever. Unsuprisingly, when things go wrong, it's not going to be
easy to sync them up again. If they were in sync at all.

Considering that freeradius is multithreaded and that it has multiple
threads to the database what chance is there that database info will be
consistant after a while? On low load - good. With high loads or spikes
- slim to none.

Processing information independantly is a bad idea if you want to have
identical copies of data. You can have both redundancy and consistancy
using built-in capability of your devices: NAS should have easily
configurable radius server fail-over; freeradius has configurable sql
module fail-over (see man unlang); sql servers can manage replication
between databases and maintain identical data sets. So, if radius1
fails, radius2 takes over; sql1 will replicate data to sql2; if sql1
fails data will be sent to sql2 and replicated to sql1 whan it comes
back (multimaster replication). If basic capabilities are not good
enough for you, there are high availability solutions that can improve
on fail-over and replication.

Ivan Kalik
Kalik Informatika ISP