Re: priority servers in an instance

I think we're missing connslots support until the next release (3.16 is mentioned in the archives as the first that's going to have it). Willy must be used to having it from testing the next version :)

switched to dst_conn and gt -- worked great. Thanks Karl!

On Wed, Feb 25, 2009 at 10:47 PM, Karl Pietri <karl#slideshare.com> wrote:
> its the -gt  make it just gt.
>
> this is what i ended up going with:
>
> frontend priority_rails_farm xx.xx.xx.xxx:80
>     mode http
>     option forwardfor
>     acl priority_full dst_conn gt 4
>     use_backend rails_farm if priority_full
>     default_backend priority_rails_farm
>
> the backend priority_rails_farm has 4 servers with maxconn 1 in it.
>
> -Karl
>
> On Wed, Feb 25, 2009 at 10:38 PM, Michael Fortson <mfortson#gmail.com>
> wrote:
>>
>> When trying this, I get:
>> [ALERT] 056/063556 (24031) : parsing [/etc/haproxy/haproxy.cfg:290] :
>> error detected while parsing ACL 'nearly_full'.
>> [ALERT] 056/063556 (24031) : Error reading configuration file :
>> /etc/haproxy/haproxy.cfg
>>
>> (haproxy version 1.3.15.7)
>>
>> source:
>>        acl nearly_full connslots(fast_mongrels) lt 10
>>        use_backend everything if nearly_full
>>
>> also tried:
>>        acl nearly_full connslots(fast_mongrels) -lt 10
>>        use_backend everything if nearly_full
>>
>>
>> hrm...
>>
>> On Sun, Feb 22, 2009 at 12:00 PM, Willy Tarreau <w#1wt.eu> wrote:
>> > On Sun, Feb 22, 2009 at 10:03:22AM -0800, Michael Fortson wrote:
>> >> That's really cool. I've been doing it with weighting, but this is much
>> >> nicer.
>> >
>> > it was proposed and developped by someone on the list (I don't remember
>> > whom right now) for exactly this purpose.
>> >
>> >> Am I right in assuming that in this example, when nearly_full is
>> >> triggered,
>> >> it will switch entirely to that?
>> >
>> > yes, back1 will get traffic only when it's not considered full, and
>> > back2 will get the excess traffic.
>> >
>> >> how does the balance between the two
>> >> backends happen in this instance?
>> >
>> > There's no balance. The second backend only receives overloads. See
>> > that as a cheap vs expensive pool of servers (or local vs remote).
>> >
>> >> Should you just repeat the definition of
>> >> the first backend within the second to go "wide" with the server
>> >> spread?
>> >
>> > Yes, this seems appropriate depending on your workload. Maybe you'll
>> > remove "maxqueue" from the second though.
>> >
>> > Hoping this helps,
>> > Willy
>> >
>> >
>
>
Received on 2009/02/26 07:52