Hi,


With RabbitMQ version 2.8.7 on erlang R15B02 setup as a cluster of 3
machines. Each queue is setup to mirror and durable. Each node is a disc
node. Server 1 is the master and stats during the tests.


I am wondering if the following is expected.
I have a single client communicating to another client both connecting
through Server 1. If I shut down servers 2 and 3 in my cluster and run my
application again the performance is 47 times slower than when I have all 3
servers running. Is this expected? I do not have load balance between my
clients and the rabbit cluster.


I honestly was expecting the single node to actually perform a little better
as it would not have to replicate data to the other two nodes.


-Mark






--
View this message in context: http://rabbitmq.1065348.n5.nabble.com/rabbitmq-cluster-with-all-but-one-node-shutdown-tp23431.html
Sent from the RabbitMQ mailing list archive at Nabble.com.

Search Discussions

  • Mark Ward at Nov 15, 2012 at 4:17 pm
    I have to add one note to this. In my previous testing what I did was just
    shut down the OS without shutting down RabbitMQ first. If I shutdown
    RabbitMQ without shutting down the OS the Server 1 responds just fine during
    my tests when the other two nodes are down.


    I may see if the OS shutdown is repeatable and definitely remember the note
    that in Windows shutdown rabbitMQ first before shutting down the OS.


    -Mark








    --
    View this message in context: http://rabbitmq.1065348.n5.nabble.com/rabbitmq-cluster-with-all-but-one-node-shutdown-tp23431p23432.html
    Sent from the RabbitMQ mailing list archive at Nabble.com.
  • Simon MacMullen at Nov 19, 2012 at 4:52 pm
    No, that's not expected. The single node should perform better.


    However: how are you measuring performance? Are you looking at the msg/s
    value in mgmt, or timing how long it takes to send n messages, or what?


    If the latter, it occurs to me you could be being mislead by the pause
    while Erlang waits for a response from the downed node(s) - this can be
    quite a while (~1min by default) but should only occur once.


    Cheers, Simon

    On 15/11/12 16:01, Mark Ward wrote:
    Hi,

    With RabbitMQ version 2.8.7 on erlang R15B02 setup as a cluster of 3
    machines. Each queue is setup to mirror and durable. Each node is a disc
    node. Server 1 is the master and stats during the tests.

    I am wondering if the following is expected.
    I have a single client communicating to another client both connecting
    through Server 1. If I shut down servers 2 and 3 in my cluster and run my
    application again the performance is 47 times slower than when I have all 3
    servers running. Is this expected? I do not have load balance between my
    clients and the rabbit cluster.

    I honestly was expecting the single node to actually perform a little better
    as it would not have to replicate data to the other two nodes.

    -Mark



    --
    View this message in context: http://rabbitmq.1065348.n5.nabble.com/rabbitmq-cluster-with-all-but-one-node-shutdown-tp23431.html
    Sent from the RabbitMQ mailing list archive at Nabble.com.
    _______________________________________________
    rabbitmq-discuss mailing list
    rabbitmq-discuss at lists.rabbitmq.com
    https://lists.rabbitmq.com/cgi-bin/mailman/listinfo/rabbitmq-discuss



    --
    Simon MacMullen
    RabbitMQ, VMware
  • Mark Ward at Nov 20, 2012 at 4:45 pm
    Hi Simon,


    I have now moved onto testing with RabbitMQ 3. I have performed the same
    test with RabbitMQ 3 and observed what you said. There is an initial pause
    but throughput resumed.


    I'll probably table the issue on my end as it looks like we will work on
    moving forward with RabbitMQ 3. Let me know if you would like me to look
    into this again with the previous version.


    The time was being measured by the client. The client setup i have is able
    to measure message round trip between two clients connected to RabbitMQ.


    -Mark






    --
    View this message in context: http://rabbitmq.1065348.n5.nabble.com/rabbitmq-cluster-with-all-but-one-node-shutdown-tp23431p23528.html
    Sent from the RabbitMQ mailing list archive at Nabble.com.

Related Discussions

Discussion Navigation
viewthread | post
Discussion Overview
grouprabbitmq-discuss @
categoriesrabbitmq
postedNov 15, '12 at 4:01p
activeNov 20, '12 at 4:45p
posts4
users2
websiterabbitmq.com
irc#rabbitmq

2 users in discussion

Mark Ward: 3 posts Simon MacMullen: 1 post

People

Translate

site design / logo © 2017 Grokbase