buffered-sql problem

[Alan DeKok]

On Feb 7, 2017, at 9:36 AM, Jouni Soini via Freeradius-Users <freeradius-users at lists.freeradius.org> wrote:
> 
>> No, just the sample detail files are fine.  I'll take a look today / tomorrow.
> 
> could you find any problem?

  No.  It works fine for me, and there's no code path I can see where there would be a problem with your scenario.

> I asked developer to rename detail.work to a distinct name each time.
> I noticed the skipping packet is seen in debug right after renaming detail.work say to backup-1000.work
> if packet is received after renaming and deleting detail.work why it exists in backup-1000.work instead of backup-1001.work?

  That just shouldn't happen.

  The detail writer locks the file.  The detail reader renames it, and tries to lock the file.  If it can't get the lock, the reader retries after a while.  It it can get the lock, it knows that the writer is done writing.  And, that the detail file writer is now writing to a *different* file.

  My guess is this:

a) you don't have "locking = yes" in the detail module configuration

or

b) the detail files are on an NFS mount, which is notorious for being bad at locking.

  Alan DeKok.