HAProxy has a TCP mode where it round robins TCP connections. Does it need to understand something specific about the wire protocol used by Flume?
________________________________ From: Brock Noland <[email protected]> To: [email protected]; Rahul Ravindran <[email protected]> Sent: Wednesday, November 14, 2012 6:20 PM Subject: Re: Flume hops behind HAProxy > It would be round robin but not sticky sessions( so each request could Goto > any random flume hop) Does HAProxy understand the protocol? To round robin requests like this it needs to understand the communication protocol, which I suppose would work if you were using the HTTPSource. On Wed, Nov 14, 2012 at 4:46 PM, Rahul Ravindran <[email protected]> wrote: > HAProxy is widely deployed already in our deployment and Ops is familiar > with dealing with it for hosts which go down etc. > > ________________________________ > From: "Camp, Roy" <[email protected]> > To: "[email protected]" <[email protected]> > Sent: Wednesday, November 14, 2012 2:15 PM > Subject: RE: Flume hops behind HAProxy > > Out of curiosity, what is the use case vs using the built in load balancing? > > > > -----Original Message----- > From: Rahul Ravindran [mailto:[email protected]] > Sent: Wednesday, November 14, 2012 1:49 PM > To: [email protected] > Cc: [email protected] > Subject: Re: Flume hops behind HAProxy > > It would be round robin but not sticky sessions( so each request could Goto > any random flume hop) > > Sent from my phone.Excuse the terseness. > > On Nov 14, 2012, at 1:33 PM, Brock Noland <[email protected]> wrote: > >> I assume it would be connection based round robin? Might work just >> fine, but probably best to the use built-in support. >> >> On Wed, Nov 14, 2012 at 2:46 PM, Rahul Ravindran <[email protected]> >> wrote: >>> Resending given I sent it during off-hours. >>> >>> ________________________________ >>> From: Rahul Ravindran <[email protected]> >>> To: "[email protected]" <[email protected]> >>> Sent: Tuesday, November 13, 2012 5:52 PM >>> Subject: Flume hops behind HAProxy >>> >>> Hi, >>> Before I try it, I wanted to check if there were any known issues >>> with this. We will have multiple flume agents sending an Avro stream >>> each to a smaller set of intermediate flume hops. Are there any >>> issues/concerns around having the flume agents send their streams to >>> an HAProxy which will round robin between the different flume hops. >>> Any issue around the transaction mechanism with this setup? >>> >>> I know that there is a selector mechanism in Flume to do this, but >>> our operations extensively use HAProxy, and are most familiar with it. >>> >>> Thanks, >>> ~Rahul. >> >> >> >> -- >> Apache MRUnit - Unit testing MapReduce - >> http://incubator.apache.org/mrunit/ > > -- Apache MRUnit - Unit testing MapReduce - http://incubator.apache.org/mrunit/
