You are viewing a plain text version of this content. The canonical link for it is here.
Posted to users@activemq.apache.org by spiderman2 <sh...@bridgewatersystems.com> on 2007/02/07 22:12:03 UTC
High Availability and Network Of Brokers
Hi,
I'm a New user. I'd like to use activemq for a Network Of Brokers /
Clustered deployment. The catch is that I can't afford to loose a single
message in the event of failure.
In the Network of Brokers topology, I'm reading this isn't possible:
"At any point in time the message will only exist in one broker's store at
once until its consumed. In the future we will support high availability
brokers using a master-slave protocol where we willl replicate a message on
to a number of slave brokers for hot standby if the master broker were to be
unavailable for a certain period of time."
http://activemq.apache.org/how-do-distributed-queues-work.html
http://activemq.apache.org/how-do-distributed-queues-work.html
Today, is there a way to have High availability with a clustered approach?
Should I just use a Master/Slave instead?
Shawn
--
View this message in context: http://www.nabble.com/High-Availability-and-Network-Of-Brokers-tf3189661s2354.html#a8854295
Sent from the ActiveMQ - User mailing list archive at Nabble.com.
Re: High Availability and Network Of Brokers
Posted by James Strachan <ja...@gmail.com>.
On 2/8/07, spiderman2 <sh...@bridgewatersystems.com> wrote:
>
> I've read about the shared database (known to be slow) and shared file
> system, SAN, (quite expensive HW). But where can I read about replicating
> messages.
The 'Pure Master Slave' is the third option...
http://activemq.apache.org/masterslave.html
> I was under the impression that Brokers currently do *not*
> replicate messages under a Network Of Brokers scenario. Only a M-S scenario.
Exactly. So a Master/Slave cluster acts like a single logical broker,
which depending on the M/S implementation it may use message
replication or may use a shared disk/file system / database
--
James
-------
http://radio.weblogs.com/0112098/
Re: High Availability and Network Of Brokers
Posted by spiderman2 <sh...@bridgewatersystems.com>.
I've read about the shared database (known to be slow) and shared file
system, SAN, (quite expensive HW). But where can I read about replicating
messages. I was under the impression that Brokers currently do *not*
replicate messages under a Network Of Brokers scenario. Only a M-S scenario.
How can I read about this and use it?
Thanks,
Shawn
James.Strachan wrote:
>
> On 2/8/07, James Strachan <ja...@gmail.com> wrote:
>> Yes, Master/Slave is the answer for high availability and to avoid
>> message loss if a broker dies.
>
> Slight correction - killing and restarting a regular broker with
> persistent messaging will not loose messages. Its if you loose the box
> on which the broker is running due to catastrophic hardware failure
> that you might loose messages. i.e. you loose the local hard drive.
>
> So Master/Slave solves this by either reusing a shared network store
> (database or file system) or by replicating messages to different
> local stores.
>
> --
>
> James
> -------
> http://radio.weblogs.com/0112098/
>
>
--
View this message in context: http://www.nabble.com/High-Availability-and-Network-Of-Brokers-tf3189661s2354.html#a8869765
Sent from the ActiveMQ - User mailing list archive at Nabble.com.
Re: High Availability and Network Of Brokers
Posted by James Strachan <ja...@gmail.com>.
On 2/8/07, James Strachan <ja...@gmail.com> wrote:
> Yes, Master/Slave is the answer for high availability and to avoid
> message loss if a broker dies.
Slight correction - killing and restarting a regular broker with
persistent messaging will not loose messages. Its if you loose the box
on which the broker is running due to catastrophic hardware failure
that you might loose messages. i.e. you loose the local hard drive.
So Master/Slave solves this by either reusing a shared network store
(database or file system) or by replicating messages to different
local stores.
--
James
-------
http://radio.weblogs.com/0112098/
Re: High Availability and Network Of Brokers
Posted by James Strachan <ja...@gmail.com>.
On 2/8/07, spiderman2 <sh...@bridgewatersystems.com> wrote:
>
> But I'm concerned with respect to scalability. I'm assuming the Network of
> Brokers was created such that the deployment could scale to a large number
> of consumers.
A single broker can easily handle many thousands of consumers.
> What kind of scalability issues/limits will I have with a
> Master/Slave scenario?
Master/Slave clusters can be networked together if you need them to.
The Master/Slave just defines the cluster which takes ownership of a
single message and ensures HA of those messages. You can network as
many Master/Slave clusters together as you need. (Though its usually
rare folks need more than one logical broker).
--
James
-------
http://radio.weblogs.com/0112098/
Re: High Availability and Network Of Brokers
Posted by spiderman2 <sh...@bridgewatersystems.com>.
But I'm concerned with respect to scalability. I'm assuming the Network of
Brokers was created such that the deployment could scale to a large number
of consumers. What kind of scalability issues/limits will I have with a
Master/Slave scenario?
Thanks!
Shawn
James.Strachan wrote:
>
> Yes, Master/Slave is the answer for high availability and to avoid
> message loss if a broker dies.
>
>
> On 2/7/07, spiderman2 <sh...@bridgewatersystems.com> wrote:
>>
>> Hi,
>>
>> I'm a New user. I'd like to use activemq for a Network Of Brokers /
>> Clustered deployment. The catch is that I can't afford to loose a single
>> message in the event of failure.
>>
>> In the Network of Brokers topology, I'm reading this isn't possible:
>>
>> "At any point in time the message will only exist in one broker's store
>> at
>> once until its consumed. In the future we will support high availability
>> brokers using a master-slave protocol where we willl replicate a message
>> on
>> to a number of slave brokers for hot standby if the master broker were to
>> be
>> unavailable for a certain period of time."
>> http://activemq.apache.org/how-do-distributed-queues-work.html
>> http://activemq.apache.org/how-do-distributed-queues-work.html
>>
>>
>> Today, is there a way to have High availability with a clustered
>> approach?
>> Should I just use a Master/Slave instead?
>>
>>
>> Shawn
>> --
>> View this message in context:
>> http://www.nabble.com/High-Availability-and-Network-Of-Brokers-tf3189661s2354.html#a8854295
>> Sent from the ActiveMQ - User mailing list archive at Nabble.com.
>>
>>
>
>
> --
>
> James
> -------
> http://radio.weblogs.com/0112098/
>
>
--
View this message in context: http://www.nabble.com/High-Availability-and-Network-Of-Brokers-tf3189661s2354.html#a8866669
Sent from the ActiveMQ - User mailing list archive at Nabble.com.
Re: High Availability and Network Of Brokers
Posted by James Strachan <ja...@gmail.com>.
Yes, Master/Slave is the answer for high availability and to avoid
message loss if a broker dies.
On 2/7/07, spiderman2 <sh...@bridgewatersystems.com> wrote:
>
> Hi,
>
> I'm a New user. I'd like to use activemq for a Network Of Brokers /
> Clustered deployment. The catch is that I can't afford to loose a single
> message in the event of failure.
>
> In the Network of Brokers topology, I'm reading this isn't possible:
>
> "At any point in time the message will only exist in one broker's store at
> once until its consumed. In the future we will support high availability
> brokers using a master-slave protocol where we willl replicate a message on
> to a number of slave brokers for hot standby if the master broker were to be
> unavailable for a certain period of time."
> http://activemq.apache.org/how-do-distributed-queues-work.html
> http://activemq.apache.org/how-do-distributed-queues-work.html
>
>
> Today, is there a way to have High availability with a clustered approach?
> Should I just use a Master/Slave instead?
>
>
> Shawn
> --
> View this message in context: http://www.nabble.com/High-Availability-and-Network-Of-Brokers-tf3189661s2354.html#a8854295
> Sent from the ActiveMQ - User mailing list archive at Nabble.com.
>
>
--
James
-------
http://radio.weblogs.com/0112098/