Re: duplicate HTTP requests

From: Anoop Bhatti <anoop.bhatti#gmail.com>
Date: Wed, 5 Nov 2008 13:54:56 -0500


I have a data cluster of machines and I want to send the same request to all of them, if they are available. If a machine has failed, I just want to ignore it. The "if they are available" part is where HAProxy comes in handy. If I put a dedicated HAProxy machine in front of the cluster, I can simply send requests to HAProxy, and let HAProxy send those requests to the available machines. HAProxy does all of that "book keeping." HAProxy is capable of doing a round robin to each available machine in the cluster, but I want something like a "broadcast" to all of them.

I have a long running client that sends these requests to the data cluster, and the probability of having a machine in the data cluster fail is high. So in other words, I'm trying to use HAProxy as a way of introducing fault tolerance into my data cluster --rather than handling it in my code. If a machine in the cluster fails, the client does not need to know --it's sending requests to the dedicated HAProxy machine. I'm trying to push the problem of fault tolerance out onto HAProxy.

It's a good idea, but it seems it's not possible....

Anoop Bhatti

--
Committed to open source technology.



On Wed, Nov 5, 2008 at 1:29 PM, Matt Sanford <matt#twitter.com> wrote:

> Hi Randall & Anoop,
>
> This is not something HAProxy can do as far as I know but I can think of
> two reasons to do this. One is horizontal scaling, but as Randall mentioned
> choosing which result (or worse, assembling composite results) is not an
> easy task. I have seen "run-off" systems where you request from multiple
> back-ends and return the first to complete. That's a pretty seldom used
> architecture and maintaining state for the multiple un-finished back-ends
> add overhead.
> The second reason I can think of is for load testing. I've done this sort
> of "siphon" load testing in the past with apache access logs and some
> scripting. It does ensure representative load for A/B change testing. As
> handy as that's been it seems out of scope for HAProxy to me.
> What type of use to you have in mind?
>
> Thanks;
> — Matt
>
> On Nov 5, 2008, at 10:21 AM, Randall Svancara wrote:
>
>> For what reason may I ask would you ever want to do this?
>>
>> Furthermore, how would HAProxy handle the response? Which response does
>> it return back?
>>
>> Randall
>>
>>
>> On Wed, 2008-11-05 at 12:54 -0500, Anoop Bhatti wrote:
>>>
>>> Hello,
>>>
>>> Is it possible to duplicate a request from one listening server to two
>>> other servers?
>>>
>>> HAProxy is great if you want to configure a request to go to other
>>> servers in a round robin fashion. But I'd like to have a
>>> configuration that sends the request to both (if they're available),
>>> and not do a round robin.
>>>
>>> Is this possible? If so, how?
>>>
>>> Thanks,
>>>
>>> Anoop Bhatti
>>> --
>>> Committed to open source technology.
>>>
>>
>
>
Received on 2008/11/05 19:54

This archive was generated by hypermail 2.2.0 : 2008/11/05 20:00 CET