Best Practices - maximum NAS entries in clients.conf

[Christ Schlacta]

On 9/13/2011 00:59, Fajar A. Nugraha wrote:
> On Tue, Sep 13, 2011 at 2:43 PM, Phil Mayers<p.mayers at imperial.ac.uk>  wrote:
>> On 09/12/2011 10:42 PM, Fajar A. Nugraha wrote:
>>> If I understand raddb/sites-available/dynamic-clients correctly, the
>>> only way to store (well, to retrieve actualy) dynamic clients
>>> definition in SQL is to use "%{sql:" expansion. Is there a way to make
>>> it have some level of redundancy? Last time I check, "%{sql:" can't be
>>> used on "virtual" modules (from instantiate or policy section) which
>>> groups multiple sql instance together using "redundant".
>>>
>> You could also use "exec", rlm_perl/python or whatever, all of which can
>> themselves call SQL.
> possible, though not ideal.
>
>> Or, perform an SQL query that MUST return some output, parse the results and
>> call the individual SQL modules directly - like so:
>>
>> update control {
>>   Tmp-String-0 := "%{sql1:select name||','||secret ...}"
>> }
>> if (control:Tmp-String-0 == "") {
>>   update control {
>>     Tmp-String-0 := "%{sql2:...}"
>>   }
>> }
> That's what we currently do (for another purpose, not for dynamic
> client). However:
> - I lost load-balancing feature that comes with redundant-load-balance
> - imagine having to create 8 if-elsif block to properly catch error
> when working with 8 sql nodes, and write the same sql query 8 times in
> the configuration file. Works, but kinda messy.
>
> With current sql module (that only reads nas list from sql during
> startup/HUP) I can use one sql/mysql/*.conf to specify the query, and
> have each sql instance $INCLUDE it. If we can do similar thing with
> "%{sql:" expansion (e.g. store the query in some temporary internal
> variable/attribute) it'd be reduce the measiness greatly, but I
> haven't found out how to do it yet.
>
why not make an arbitrary program that takes the SQL statement as an 
argument, and returns from the first successful connection.  it can take 
a random number between 0 and n-1 on the number of SQL servers you have, 
and start connecting from there.  you get failover and round-robin load 
balancing with the convenience of only having to write your query and 
your series of if-else-if statements once.