Return-Path: X-Original-To: apmail-flume-user-archive@www.apache.org Delivered-To: apmail-flume-user-archive@www.apache.org Received: from mail.apache.org (hermes.apache.org [140.211.11.3]) by minotaur.apache.org (Postfix) with SMTP id 937FBDEE8 for ; Thu, 6 Sep 2012 14:35:55 +0000 (UTC) Received: (qmail 45110 invoked by uid 500); 6 Sep 2012 14:35:55 -0000 Delivered-To: apmail-flume-user-archive@flume.apache.org Received: (qmail 45048 invoked by uid 500); 6 Sep 2012 14:35:54 -0000 Mailing-List: contact user-help@flume.apache.org; run by ezmlm Precedence: bulk List-Help: List-Unsubscribe: List-Post: List-Id: Reply-To: user@flume.apache.org Delivered-To: mailing list user@flume.apache.org Received: (qmail 45036 invoked by uid 99); 6 Sep 2012 14:35:54 -0000 Received: from nike.apache.org (HELO nike.apache.org) (192.87.106.230) by apache.org (qpsmtpd/0.29) with ESMTP; Thu, 06 Sep 2012 14:35:54 +0000 X-ASF-Spam-Status: No, hits=1.5 required=5.0 tests=FSL_RCVD_USER,HTML_MESSAGE,RCVD_IN_DNSWL_LOW,SPF_PASS X-Spam-Check-By: apache.org Received-SPF: pass (nike.apache.org: domain of cwneal@gmail.com designates 209.85.212.173 as permitted sender) Received: from [209.85.212.173] (HELO mail-wi0-f173.google.com) (209.85.212.173) by apache.org (qpsmtpd/0.29) with ESMTP; Thu, 06 Sep 2012 14:35:47 +0000 Received: by wibhm6 with SMTP id hm6so5126509wib.8 for ; Thu, 06 Sep 2012 07:35:27 -0700 (PDT) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gmail.com; s=20120113; h=mime-version:in-reply-to:references:date:message-id:subject:from:to :content-type; bh=hTyUGtACbAWq0aqXTRkNkIWP/2C4f7k2C2emdI/We2c=; b=mefE6UEhZV0Jpprn5bXImXq7YicqEJDeP8PHkOQN7eFBsSx5b1EWqHhYkCoazPNfhR SUVsmYm+t30tpq0J7HoZGQspBOpzpDEKpVQWad8U8Zh40W+WsBJigH68qed2ncRgoFnL QK1A9oqPyXkaUci1Cz/u+wAktBAr6A9E5+szWsur8cqxWFoBHrJAx8Rtr1YVyuIAcuYV 3eExklqwDv2AyHN1bSoohF9dtkXkUHaytEAMLeaAt2PgpGYkRgBZAr/3h02pSwxQvRt9 NaMnOEe7EaDNUsVW7HEW3fTxFL/oVR9iefuGKCRMz2o2VUmKSIrIqbm466jVJIj5jbRs 5vZQ== MIME-Version: 1.0 Received: by 10.180.84.104 with SMTP id x8mr46177370wiy.20.1346942126793; Thu, 06 Sep 2012 07:35:26 -0700 (PDT) Received: by 10.216.244.133 with HTTP; Thu, 6 Sep 2012 07:35:26 -0700 (PDT) In-Reply-To: <5048257B.6090602@cyberagent.co.jp> References: <5045DA44.8000400@cyberagent.co.jp> <5048257B.6090602@cyberagent.co.jp> Date: Thu, 6 Sep 2012 09:35:26 -0500 Message-ID: Subject: Re: Failover Processor + Load Balanced Processor? From: Chris Neal To: user@flume.apache.org Content-Type: multipart/alternative; boundary=f46d043bd75800234004c909676f --f46d043bd75800234004c909676f Content-Type: text/plain; charset=ISO-8859-1 Nice! Thanks :) Will take a look. On Wed, Sep 5, 2012 at 11:24 PM, Juhani Connolly < juhani_connolly@cyberagent.co.jp> wrote: > Since there was no response to this, I set up a separate ticket at > https://issues.apache.org/jira/browse/FLUME-1541 and implemented it as a > SinkSelector for the LoadBalancingSinkProcessor. > > Review can be found at https://reviews.apache.org/r/6939/ > > Chris: if you're interested you may want to give this a poke, see if it > fulfills your needs. The only change in configuration needed is to change > the selector type from "round_robin" to "round_robin_backoff" > > > On 09/04/2012 07:39 PM, Juhani Connolly wrote: > > I'm thinking of working on this(adding backoff semantics to the load > balancing processor) > > The ticket FLUME-1488 however refers to the load balancing rpc client(or > is it just poorly worded/unclear?). If it is in fact a separate ticket I'll > file one for this > > Anyway, I was interested in hearing thoughts on approach. I'd have liked > to do it within the framework of the LoadBalancingSinkProcessor by adding a > new Selector, however as it is now, it the processor provides no feedback > to the selectors about whether sinks are working or not, so this can't work. > > This leaves two choices: write a new SinkProcessor or modify the > SinkSelector interface to give it a couple of callbacks that the processor > calls to inform the selector of trouble. This shouldn't really be a problem > even if people have written their own selectors so long as they are > extending AbstractSinkSelector which can stub the callbacks. > > Thoughts? > > On 08/18/2012 02:01 AM, Arvind Prabhakar wrote: > > Hi, > > FYI - the load balancing sink processor does support simple failover > semantics. The way it works is that if a sink is down, it will proceed to > the next sink in the group until all sinks are exhausted. The failover sink > processor on the other hand does complex failure handling and back-off such > as blacklisting sinks that repeatedly fail etc. The issue [1] tracks > enhancing this processor to support backoff semantics. > > The one issue with your configuration that I could spot by a quick > glance is that you are adding your active sinks to both the sink groups. > This does not really work and the configuration subsystem simply flags the > second inclusion as a problem and ignores it. By design, a sink can either > be on its own or in one explicit sink group. > > [1] https://issues.apache.org/jira/browse/FLUME-1488 > > Regards, > Arvind Prabhakar > > On Fri, Aug 17, 2012 at 8:59 AM, Chris Neal wrote: > >> Hi all. >> >> The User Guide talks about the various types of Sink Processors, but >> doesn't say whether they can be aggregated together. A Failover Processor >> that moves between 1..n sinks is great, as is a Load Balancer Processor >> that moves between 1..n sinks, but what is the best would be an agent that >> can utilize both a Failover Processor AND a Load Balancer Processor! >> >> I've created a configuration which I believe supports this, and the >> Agent starts up and processes events, but I wanted to ping this group to >> make sure that this configuration is really doing what I think it is doing >> behind the scenes. >> >> Comments? >> >> # Define the sources, sinks, and channels for the agent >> agent.sources = avro-instance_1-source avro-instance_2-source >> agent.channels = memory-agent-channel >> agent.sinks = avro-hdfs_1-sink avro-hdfs_2-sink >> agent.sinkgroups = failover-sink-group lb-sink-group >> >> # Bind sources to channels >> agent.sources.avro-instance_1-source.channels = memory-agent-channel >> agent.sources.avro-instance_2-source.channels = memory-agent-channel >> >> # Define sink group for failover >> agent.sinkgroups.failover-sink-group.sinks = avro-hdfs_1-sink >> avro-hdfs_2-sink >> agent.sinkgroups.failover-sink-group.processor.type = failover >> agent.sinkgroups.failover-sink-group.processor.priority.avro-hdfs_1-sink >> = 5 >> agent.sinkgroups.failover-sink-group.processor.priority.avro-hdfs_2-sink >> = 10 >> agent.sinkgroups.failover-sink-group.processor.maxpenalty = 10000 >> >> # Define sink group for load balancing >> agent.sinkgroups = lb-sink-group >> agent.sinkgroups.group1.sinks = avro-hdfs_1-sink avro-hdfs_2-sink >> agent.sinkgroups.group1.processor.type = load_balance >> agent.sinkgroups.group1.processor.selector = round_robin >> >> # Bind sinks to channels >> agent.sinks.avro-hdfs_1-sink.channel = memory-agent-channel >> agent.sinks.avro-hdfs_2-sink.channel = memory-agent-channel >> >> # avro-instance_1-source properties >> agent.sources.avro-instance_1-source.type = exec >> agent.sources.avro-instance_1-source.command = tail -F /somedir/Trans.log >> agent.sources.avro-instance_1-source.restart = true >> agent.sources.avro-instance_1-source.batchSize = 100 >> >> # avro-instance_2-source properties >> agent.sources.avro-instance_2-source.type = exec >> agent.sources.avro-instance_2-source.command = tail -F >> /somedir/UDXMLTrans.log >> agent.sources.avro-instance_2-source.restart = true >> agent.sources.avro-instance_2-source.batchSize = 100 >> >> # avro-hdfs_1-sink properties >> agent.sinks.avro-hdfs_1-sink.type = avro >> agent.sinks.avro-hdfs_1-sink.hostname = hdfshost1.domin.com >> agent.sinks.avro-hdfs_1-sink.port = 10000 >> >> # avro-hdfs_2-sink properties >> agent.sinks.avro-hdfs_2-sink.type = avro >> agent.sinks.avro-hdfs_2-sink.hostname = hdfshost2.domain.com >> agent.sinks.avro-hdfs_2-sink.port = 10000 >> >> # memory-agent-channel properties >> agent.channels.memory-agent-channel.type = memory >> agent.channels.memory-agent-channel.capacity = 20000 >> agent.channels.memory-agent-channel.transactionCapacity = 100 >> >> Thanks! >> > > > > --f46d043bd75800234004c909676f Content-Type: text/html; charset=ISO-8859-1 Content-Transfer-Encoding: quoted-printable Nice! =A0Thanks :) =A0Will take a look.

O= n Wed, Sep 5, 2012 at 11:24 PM, Juhani Connolly <juhani_con= nolly@cyberagent.co.jp> wrote:
=20 =20 =20
Since there was no response to this, I set up a separate ticket=A0 at =20 https://issues.apache.org/jira/browse/FLUME-1541 and implemented it as a SinkSelector for the LoadBalancingSinkProcessor.

Review can be found at https://reviews.apache.org/r/6939/

Chris: if you're interested you may want to give this a poke, see if it fulfills your needs. The only change in configuration needed is to change the selector type from "round_robin" to "round_robin_backoff"


On 09/04/2012 07:39 PM, Juhani Connolly wrote:
=20
I'm thinking of working on this(adding backoff semantics to the load balancing processor)

The ticket FLUME-1488 however refers to the load balancing rpc client(or is it just poorly worded/unclear?). If it is in fact a separate ticket I'll file one for this

Anyway, I was=A0 interested in hearing thoughts on approach. I'= d have liked to do it within the framework of the LoadBalancingSinkProcessor by adding a new Selector, however as it is now, it the processor provides no feedback to the selectors about whether sinks are working or not, so this can't work.

This leaves two choices: write a new SinkProcessor or modify the SinkSelector interface to give it a couple of callbacks that the processor calls to inform the selector of trouble. This shouldn't really be a problem even if people have written their own selectors so long as they are extending AbstractSinkSelector which can stub the callbacks.

Thoughts?

On 08/18/2012 02:01 AM, Arvind Prabhakar wrote:
Hi,

FYI - the load balancing sink processor does support simple failover semantics. The way it works is that if a sink is down, it will proceed to the next sink in the group until all sinks are exhausted. The failover sink processor on the other hand does complex failure handling and back-off such as blacklisting sinks that repeatedly fail etc. The issue [1] tracks enhancing this processor to support backoff semantics.

The one issue with your configuration that I could spot by a quick glance is that you are adding your active sinks to both the sink groups. This does not really work and the configuration subsystem simply flags the second inclusion as a problem and ignores it. By design, a sink can either be on its own or in one explicit sink group.=A0


Regards,
Arvind Prabhakar

On Fri, Aug 17, 2012 at 8:59 AM, Chris Neal <cwneal@gmail.com> wrote:
Hi all.

The User Guide talks about the various types of Sink Processors, but doesn't say whether they can be aggregated together. =A0A Failover Processor that moves between 1..n sinks is great, as is a Load Balancer Processor that moves between 1..n sinks, but what is the best would be an agent that can utilize both a Failover Processor AND a Load Balancer Processor!

I've created a configuration which I believe support= s this, and the Agent starts up and processes events, but I wanted to ping this group to make sure that this configuration is really doing what I think it is doing behind the scenes.

Comments?

# Define the sources, sinks, and channels for the agent
agent.sources =3D avro-instance_1-source avro-instance_2-source
agent.channels =3D memory-agent-channel
agent.sinks =3D avro-hdfs_1-sink avro-hdfs_2-sink
agent.sinkgroups =3D failover-sink-group lb-sink-group

# Bind sources to channels
agent.sources.avro-instance_1-source.channels =3D memory-agent-channel
agent.sources.avro-instance_2-source.channels =3D memory-agent-channel

# Define sink group for failover
agent.sinkgroups.failover-sink-group.sinks =3D avro-hdfs_1-sink avro-hdfs_2-sink
agent.sinkgroups.failover-sink-group.processor.type =3D failover
agent.sinkgroups.failover-sink-group.processor.priorit= y.avro-hdfs_1-sink =3D 5
agent.sinkgroups.failover-sink-group.processor.priorit= y.avro-hdfs_2-sink =3D 10
agent.sinkgroups.failover-sink-group.processor.maxpena= lty =3D 10000

# Define sink group for load balancing
agent.sinkgroups =3D lb-sink-group
agent.sinkgroups.group1.sinks =3D avro-hdfs_1-sink avro-hdfs_2-sink
agent.sinkgroups.group1.processor.type =3D load_balance
agent.sinkgroups.group1.processor.selector =3D round_robin

# Bind sinks to channels
agent.sinks.avro-hdfs_1-sink.channel =3D memory-agent-channel
agent.sinks.avro-hdfs_2-sink.channel =3D memory-agent-channel

# avro-instance_1-source properties
agent.sources.avro-instance_1-source.type =3D exec
agent.sources.avro-instance_1-source.command =3D tail -F /somedir/Trans.log
agent.sources.avro-instance_1-source.restart =3D true<= /div>
agent.sources.avro-instance_1-source.batchSize =3D 100

# avro-instance_2-source properties
agent.sources.avro-instance_2-source.type =3D exec
agent.sources.avro-instance_2-source.command =3D tail -F /somedir/UDXMLTrans.log
agent.sources.avro-instance_2-source.restart =3D true<= /div>
agent.sources.avro-instance_2-source.batchSize =3D 100

# avro-hdfs_1-sink properties
agent.sinks.avro-hdfs_1-sink.type =3D avro
agent.sinks.avro-hdfs_1-sink.hostname =3D hdfshost1.domin.com
agent.sinks.avro-hdfs_1-sink.port =3D 10000

# avro-hdfs_2-sink properties
agent.sinks.avro-hdfs_2-sink.type =3D avro
agent.sinks.avro-hdfs_2-sink.hostname =3D hdfshost2.domain.com
agent.sinks.avro-hdfs_2-sink.port =3D 10000

# memory-agent-channel properties
agent.channels.memory-agent-channel.type =3D memory
agent.channels.memory-agent-channel.capacity =3D 20000
agent.channels.memory-agent-channel.transactionCapacit= y =3D 100

Thanks!




--f46d043bd75800234004c909676f--