load balancing problem
hi,
i've a proxy with freeradius 1.1.6 in load balacing with two back-end
radius 1.1.6
my proxy is configured like this
realm APPLI1 {
type= radius
authhost= xx.xx.xx.xx:1820
accthost= xx.xx.xx.xx:1821
ldflag = round_robin
secret = mysecret
}
# Server Canna PROD Accellio GPRS de Test avec Realm
realm APPLI1 {
type= radius
authhost= yy.yy.yy.yy:1820
accthost= yy.yy.yy.yy:1821
ldflag = round_robin
secret = mysecret
}
I see the Authentication request and response then Accounting start an
some hours later accounting stop on the proxy
the problem is that I see the Authentication request and response then
Accounting start on the fisrt back-end server and the accounting stop
on the second backend server.
is this a bug or a problem of configuration ?
thanks in advance.
Regards
JPR
-
List info/subscribe/unsubscribe? See http://www.freeradius.org/list/users.html
Re: load balancing problem
EXT / GFI REBOLJ Jean-Pierre wrote: > hi, > i've a proxy with freeradius 1.1.6 in load balacing with two back-end > radius 1.1.6 ... > the problem is that I see the Authentication request and response then > Accounting start on the fisrt back-end server and the accounting stop > on the second backend server. > is this a bug or a problem of configuration ? It's the way load balancing works. It's documented as working this way. Requests get randomly load balanced between different home servers. FreeRADIUS does *not* track individual user sessions for load balancing. That's a *lot* of work. In 2.0, you can do "client-balance", which does load-balancing by hashing the client IP address. So all requests from one client will go to one server. But it's only useful if you have a large number of clients. Alan DeKok. - List info/subscribe/unsubscribe? See http://www.freeradius.org/list/users.html
Re: load balancing problem
On 6/29/07, Alan DeKok <[EMAIL PROTECTED]> wrote: > > Accounting start on the fisrt back-end server and the accounting stop > > on the second backend server. > > is this a bug or a problem of configuration ? > > It's the way load balancing works. It's documented as working this > way. Requests get randomly load balanced between different home servers. in my setup, log dirs live in a shared filesystem, as for accounting.. the SQL server is one. - List info/subscribe/unsubscribe? See http://www.freeradius.org/list/users.html
Re: load balancing problem
inverse wrote: > in my setup, log dirs live in a shared filesystem, NFS mounted? Don't. If NFS goes away, any application using those directories will lock, and be unkillable. Alan DeKok. - List info/subscribe/unsubscribe? See http://www.freeradius.org/list/users.html
Re: load balancing problem
Alan wrote: > NFS mounted? Don't. If NFS goes away, any application using those > directories will lock, and be unkillable. it's part of a red hat cluster, and it's managed by that software suite. If a machine dies a transparent switch occurs. If it fails I'll get angry with red hat --so far it didn't happen (and I didn't bother finding out if it's nfs-based or something else). This feature is not critical, ok. I can live without it. It just works, and simplifies analisys a lot. I've got other reasons to be angry with red had, but this is not the case. Personally I like sshfs much more than nfs, but it's prone to similar problems as those above. So I won't use it. bye, inverse - List info/subscribe/unsubscribe? See http://www.freeradius.org/list/users.html
Re: load balancing problem
"EXT / GFI REBOLJ Jean-Pierre" <[EMAIL PROTECTED]> said: [snip] >ldflag = round_robin [snip] > the problem is that I see the Authentication request and response then > Accounting start on the fisrt back-end server and the accounting stop > on the second backend server. That sounds like round_robin is doing its job. >From the example proxy.conf: # If there is no ldflag attribute, or it is set to 'fail_over', then # the realms are treated as "fail-over". That is, the first matching # realm is used, unless it is down, in which case the realm "fails # over" to the second matching realm. The process continues until an # active matching realm is found, OR the DEFAULT realm is returned. # # If the ldflag attribute is set to 'round_robin', then all active # realms of the same name are put into a pool internally in the # server, and the proxied requests are evenly divided among the # realms in the pool. For this to work, all realms of the same name # MUST have the same value of their 'ldflag' attributes. Mixing up # different types of load balancing schemes for the same realm will # cause problems. If you are trying to truly load balance, rather than just failover, I don't think this can be done just in the realms config. You should probably read doc/load-balance.txt. > JPR -- hugh - List info/subscribe/unsubscribe? See http://www.freeradius.org/list/users.html
