Freeradius not sending traffic-shaping attributes randomly (Works when in debug mode)

[Antônio Modesto]

On 15/12/2021 17:59, Alan DeKok wrote:
> On Dec 15, 2021, at 3:31 PM, Antônio Modesto <modesto at hubsoft.com.br> wrote:
>> I am facing a strange problem with one radius server (FreeBSD 12.2 + FreeRADIUS 2.0.21).
>    Which is end of life, dead, and not supported.
>
>> In a random way, Freeradius is not sending some traffic-shaping attributes in the Access-Accept to the NAS (I confirmed it with wireshark). What is puzzling me is that when I run radiusd in debug mode (-X), the problem disappears.
>    Where do these attributes come from?  You've been careful to not give any information about your configuration.
>
>    The server doesn't magically decide to behave differently from day to day.  In 99.999% of these situations, the database is overloaded.  So FreeRADIUS can't get the data it needs, and then things break.
>
>    If the server just uses flat-text files (and no database), then it will *never* suddenly start behaving differently.  It will *always* give exactly the same output for the same input.
>
>> I tried using raddebug but it uses a lot of CPU power, so it is not viable to run it in production. What do you guys suggest?
>    Upgrade.
>
>    And investigate database issues.  Check the radius.log file.  Odds are that there will be LOTS of complaints about "unresponsive child" or "module SQL is blocked".  Fix that.
>
>    Alan DeKok.
>
>
> -
> List info/subscribe/unsubscribe? See http://www.freeradius.org/list/users.htm

I am sorry, I am using Freeradius 3.0.21, not 2.0.21.


But regarding the attributes, they are being read from the database 
(PostgreSQL). What you said about database issues does not make sense to 
me. If freeradius can't read something from the database, it should halt 
the request, not respond a Access-Accept with truncated attributes. As I 
said before, the weird part is that running freeradius in debug-mode 
(single-threaded) the problem didn't happen.

I will try to upgrade to the last build and see what happens.


Thanks.


> l
-- 
Att, *Antônio Modesto*