locked
Running out of memory on indexer when we copy fixml from one cluster to another. RRS feed

  • Question

  • Environment:

    FAST ESP 5.3 SP4

    Single Row Four Columns in Archive Mode.  5 Partitions (3 on archive @ 2.7 million & 0 is 50K & 1 is 500K setup to trigger)

    Windows Server 2k8 R2 64 bit 48GB Memory

    Document Size 154 Bytes (in memory) and 14.3 KB (on disk).

    We have been able to successfully feed a cluster through the API with indexing suspended a total of 23.3 million documents and then turn indexing on and everything works.  However, when we copy the fixml to another cluster with the same hardware configuration, we get the following error.

    2013-08-25 17:40:18]   ERROR   indexer   ctcxapif2C2.dmzext.local   15900   systemmsg   index_producer (4): Index producer is unavailable  

     [2013-08-25 17:40:06]   ERROR   index_producer_4   CTCXAPIF2C2.dmzext.local       systemmsg   indexer_holder: Ran out of memory. Shutting down.  

     [2013-08-25 17:40:03]   WARNING   indexer   ctcxapif2C0.dmzext.local   15900   systemmsg   index_worker (4_2): Indexing failed.  

     [2013-08-25 17:40:03]   INFO   indexer   ctcxapif2C0.dmzext.local   15900   systemmsg   index_worker (4_2): Dropping index...  

    This process of copying fixml from one cluster to another worked when our archive partitions were set to 2million each, but has started to fail at 2.7million.  We want to understand what is happening differently in indexing in the case where we copy the fixml versus having it fed through the FAST pipeline and the API.  The process seems similar and if one works we are hopeful we can trick FAST into treating the copied fixml in the same way.  It appears the main difference is it is not indexing in the smaller partitions and triggering the move.  Is there a way to force that behavior?  Trigger mode versus bulk mode.   

    Tuesday, August 27, 2013 5:35 PM

All replies

  • We figured it out.  We had not as part of our copy process done summaryclasses.xml and reapplied the index profile.  After doing so, the memory of the indexer stayed low and the problem went away.
    Wednesday, August 28, 2013 3:16 PM
  • Thank you for posting back  the solution.
    Wednesday, August 28, 2013 6:54 PM