[rabbitmq-discuss] RabbitMQ Clustering
simon at rabbitmq.com
Thu Dec 1 13:03:37 GMT 2011
Depends what you mean by "scale as expected". In general, mirrored
queues *do not add scalability* - as all the mirrors have to do the same
amount of work anyway.
In practice mirrored queues are quite a bit slower than unmirrored ones,
due to the amount of coordination needed across nodes.
On 27/11/11 02:32, Venkat wrote:
> Good evening.
> Following is our clustering scenario:
> - Using RabbitMQ 2.6.1 with spring amqp 1.0 using RabbitMQ mirrored
> - We have two nodes on which queue consumers are running: q_c_node1
> and q_c_node2.
> - We have two rabbitmq nodes rmq_node1 and rmq_node2.
> - Using HA Proxy to load balance.
> - When q_c_node1 is started assume that there are two queue consumers
> q_c1 using queue q1 and q_c2 using queue q2 are started. Assume that
> HA Proxy routed the q1 creation to rmq_node1 and q2 creation to
> - When q_c_node2 is started, q_c1 and q_c2 will be pointing to q1 on
> rmq_node1 and q2 on rmq_node2 as q1 and q2 are already created (please
> correct me if I am wrong).
> - Assume that rmq_node1 is down. Since mirrored queues are used, we
> have rmq_node2 as back up. The q_c1 consumer on q_c_node1 and
> q_c_node2 will be pointed to rmq_node2 using HA Proxy and spring-
> amqp's reconnecting mechanism. At this point two queue consumers
> runnning on q_c_node1 and q_c_node2 are using rmq_node2 (as rmq_node1
> is down). If a brand new rabbitmq node rmq_new_node joins the cluster
> (using ha-x-policy-all, therefore this node also is mirrored) will it
> scale as expected?. My concern is, q_c_node1 and q_c_node2 are
> pointing to rmq_node2. Whenever requests go to rmq_new_node I am
> assuming that they will be simply passed on to rmq_node1. How will the
> CPU Utilization and I/O activity will be affected on rmq_node2 and
> rabbitmq-discuss mailing list
> rabbitmq-discuss at lists.rabbitmq.com
More information about the rabbitmq-discuss