Florian Lohoff wrote:
For most of my purposes i dont care about systems not available for a longer period as backend systems take care on synchronisation.
Then why replicate via RADIUS? Why not synchronise via the backend?
In the past 15 years i have seen a lot of broken Radius implementations not sending Stop/Start or Accounting On/Off so relying on state by Radius is some kind of guesswork in my eyes.
Yes.
When talking about reliability you talk about fsync after every packet before acknowledge?
No. The detail module doesn't fsync to disk. But it does do it's "best effort" to write to disk.
Store and Forward for Mails is okay - but doing that with some 2000 Packets per seconds sounds wrong.
So submit a patch which implements accounting replication which (a) doesn't write to disk, and (b) is robust in the event of temporary process/system failures. I don't think you can satisfy both requirements at the same time. What you CAN do is to bypass the detail file most of the time. See raddb/sites-available/robust-proxy-accounting Alan DeKok.