I believe rebalancing should be triggered by the user, using the command "storm rebalance" or its API equivalent. The rest is about right :)

Good luck,

Ahmed

On 01/30/2015 03:46 PM, Manpreet Singh wrote:
Thanks a lot Ahmed.

It really gave me a direction. So you are saying that:

  * A program should check metrics at regular intervals.
  * Based on that, program can add/remove VMs to cluster. Storm will
    automatically recognize the change in VMs
  * Depending on VM addition or removal, program will trigger
    re-balance command to change no. of worker processes or executors.

Did I get it right?

Thanks,
Manpreet.

On Thu, Jan 29, 2015 at 2:50 PM, Ahmed El Rheddane <[email protected] <mailto:[email protected]>> wrote:

    Hi all,

    I have investigated ways of auto scaling Storm topologies recently
    and it is perfectly feasible.
    - For monitoring, you can use either the Thrift interface or the
    new MetricsConsumer bolt, it's probably possible using the REST
    API as well, but I haven't tried that out yet.
    - For scaling, Storm automatically discovers the addition of new
    nodes (VMs) to the cluster and allows you to change the number of
    executors (threads) per component on the fly.
    - You can as well plug in a custom scheduler, which is called upon
    each rebalancing request, to place the executors of your topology
    as you see fits.

    I hope this could help.

    Ahmed


    On 01/28/2015 09:55 PM, Parth Brahmbhatt wrote:

        Not sure how much progress has been made on
        https://issues.apache.org/jira/browse/STORM-594 but that is
        the only
        initiative that I am aware of.

        Thanks
        Parth

        On 1/28/15, 12:42 PM, "Margus Roo" <[email protected]
        <mailto:[email protected]>> wrote:

            I have not heard or read about it from the documentation.

            Margus (margusja) Roo
            http://margus.roo.ee
            skype: margusja
            +372 51 480

            On 28/01/15 20:59, Manpreet Singh wrote:

                Hi,

                I am a newbie to Apache Storm and analyzing it for a
                product. I am
                comparing it with Gearman from parallel processing
                aspect. So far
                Apache Storm seems to be winning on most of the
                aspects. But one of
                the important aspect which I got stuck is
                'auto-scale'. Auto-scale (up
                as well as down) of storm resources doesn't look
                possible. This is
                critical for my product as stream data could vary a
                lot through out
                the day. Having auto-scale will save me cost as well
                as monitoring the
                usage every now and then.

                Are there any workarounds to this problem with Apache
                Storm?

                Thanks in advance.

                -Manpreet.




Reply via email to