Local shared filesystem master slave with geo-redundant pure master slave

[Stevo Slavić <ss...@gmail.com>]

Hello ActiveMQ users,

Imagine 4 nodes, 2 per location, on each node on same location/LAN a
shared filesystem (separate node) used by two local brokers in shared
filesystem master slave (SFSMS) configuration. Can destinations and
messages be shared for HA across the geo-redundant nodes?

Can failover protocol be used for network connectors? Then AMQ brokers
on one location could connect to brokers on other location via
failover protocol. Compared to transport connector failover handling,
if AMQ brokers from one location can not connect to neither of
failover AMQ brokers on other location (e.g. if other location is down
completely, neither of the SFSMS nodes are responding), they should
continue to operate as if nothing happened (slave not responding,
down). When other location is brought back up, before putting it
online one will have to sync the message storage manually, just like
in pure master slave. For each location, other location would be a
slave, in a pure master slave configuration.

Does this make sense? Is it feasible with AMQ 5.5.0?

Regards,
Stevo.

Re: Local shared filesystem master slave with geo-redundant pure master slave

[Stevo Slavić <ss...@gmail.com>]

Thanks Torsten for reply and clarifications!

Obviously I didn't understand well the difference between master/slave
and network of brokers.

I think performance constraints are not too high, having HA is more
important - message received must be consumed if at least one broker &
consumer is running. Will check requirements and get back.

Regards,
Stevo.

On Thu, Sep 29, 2011 at 9:26 AM, Torsten Mielke <to...@fusesource.com> wrote:
>> Can failover protocol be used for network connectors?
>
> Yes, please see http://tmielke.blogspot.com/2011/09/activemq-network-bridge-to-masterslave.html
>
>> Can destinations and
>> messages be shared for HA across the geo-redundant nodes?
> Using a network connector you do not share messages between brokers but you allow messages to travel between brokers.
> I.e. when a msg travels to a different broker it is deleted on the local broker. Msgs only travel to remote brokers within a network of brokers, when there are consumers registered on the remote broker.
> So this will not serve as a master/slave solution.
>
> Master / Slave is typically done on a shared resource (file system or database). This will be difficult to setup between brokers on different geo locations.
> Pure master slave replicates everything but do you really want this over a WAN connection?
>
> Typically users set up master/slave on nodes within one geo location and connect geo location using a network connector.
>
>
> Torsten Mielke
> torsten@fusesource.com
> tmielke@blogspot.com
>
>
> On Sep 28, 2011, at 6:17 PM, Stevo Slavić wrote:
>
>> Hello ActiveMQ users,
>>
>> Imagine 4 nodes, 2 per location, on each node on same location/LAN a
>> shared filesystem (separate node) used by two local brokers in shared
>> filesystem master slave (SFSMS) configuration. Can destinations and
>> messages be shared for HA across the geo-redundant nodes?
>>
>> Can failover protocol be used for network connectors? Then AMQ brokers
>> on one location could connect to brokers on other location via
>> failover protocol. Compared to transport connector failover handling,
>> if AMQ brokers from one location can not connect to neither of
>> failover AMQ brokers on other location (e.g. if other location is down
>> completely, neither of the SFSMS nodes are responding), they should
>> continue to operate as if nothing happened (slave not responding,
>> down). When other location is brought back up, before putting it
>> online one will have to sync the message storage manually, just like
>> in pure master slave. For each location, other location would be a
>> slave, in a pure master slave configuration.
>>
>> Does this make sense? Is it feasible with AMQ 5.5.0?
>>
>> Regards,
>> Stevo.
>
>
>
>
>
>

Re: Local shared filesystem master slave with geo-redundant pure master slave

[Torsten Mielke <to...@fusesource.com>]

> Can failover protocol be used for network connectors? 

Yes, please see http://tmielke.blogspot.com/2011/09/activemq-network-bridge-to-masterslave.html

> Can destinations and
> messages be shared for HA across the geo-redundant nodes?
Using a network connector you do not share messages between brokers but you allow messages to travel between brokers.
I.e. when a msg travels to a different broker it is deleted on the local broker. Msgs only travel to remote brokers within a network of brokers, when there are consumers registered on the remote broker.
So this will not serve as a master/slave solution.

Master / Slave is typically done on a shared resource (file system or database). This will be difficult to setup between brokers on different geo locations.
Pure master slave replicates everything but do you really want this over a WAN connection?

Typically users set up master/slave on nodes within one geo location and connect geo location using a network connector. 


Torsten Mielke
torsten@fusesource.com
tmielke@blogspot.com


On Sep 28, 2011, at 6:17 PM, Stevo Slavić wrote:

> Hello ActiveMQ users,
> 
> Imagine 4 nodes, 2 per location, on each node on same location/LAN a
> shared filesystem (separate node) used by two local brokers in shared
> filesystem master slave (SFSMS) configuration. Can destinations and
> messages be shared for HA across the geo-redundant nodes?
> 
> Can failover protocol be used for network connectors? Then AMQ brokers
> on one location could connect to brokers on other location via
> failover protocol. Compared to transport connector failover handling,
> if AMQ brokers from one location can not connect to neither of
> failover AMQ brokers on other location (e.g. if other location is down
> completely, neither of the SFSMS nodes are responding), they should
> continue to operate as if nothing happened (slave not responding,
> down). When other location is brought back up, before putting it
> online one will have to sync the message storage manually, just like
> in pure master slave. For each location, other location would be a
> slave, in a pure master slave configuration.
> 
> Does this make sense? Is it feasible with AMQ 5.5.0?
> 
> Regards,
> Stevo.