Update to the problem: the accounting-stop alternate query is actually an INSERT, not an UPDATE, by default, which actually surprises me, as in case of a duplicate packet, an INSERT into a properly unique-indexed table is doomed. I have now simply changed the -alt into an UPDATE query, so it would also not affect any record, but at least will not cause Oracle to balk. My only problem now is how to make sure FreeRADIUS does not freeze upon an SQL error. Cheers, Mike Mother wrote:
Hi all,
I am seeing a strange situation. I receive an accounting-stop request from a NAS, and FreeRADIUS (1.1.7 against Oracle) updates the corresponding radacct record. However, the NAS is not receiving the ack, and thus re-sends the stop request. On the second request, FreeRADIUS tries to do an update query again, and then, an insert query, with the stop message details (i.e. only a stop time, reason idle-timeout, etc.), which fails. After the third request from the NAS (and corresponding update followed by insert), Oracle throws a unique constraint violation error, and the server freezes.
Questions:
1. Why is FreeRADIUS failing to see that this request was already acknowledged, i.e. it has been updated on the database, and just sends an ACK, rather than trying to insert a new record?
2. Why does FreeRADIUS freeze on an SQL error from Oracle? Should it not just log the error and carry on about its business? I am finding my server freezing every few days due to these issues, for example, if a query takes too long to run, or a trigger fails to execute. Is FreeRADIUS against Oracle more fragile than say MySQL?
Cheers,
Mike
- List info/subscribe/unsubscribe? See http://www.freeradius.org/list/users.html