fixmlindex unable to bind to nameservice

Symptom

The following messages are recorded in the index_producer logs:

Cause

Adding additional columns, partitions, or processes in general, increases the number of connections to the nameservice, which by default, has a limited number of dedicated connections that stay allocated to particular components.

It is also possible to see this during simultaneous process restarts, such as starting up all indexers at once, or by applying an index-profile.


Resolution

To resolve this issue, change the nameservice settings in %FASTSEARCH%\etc\omniorb.cfg on the admin node as follows:

threadPerConnectionPolicy = 0
maxServerThreadPoolSize = 200
These settings make each request use a pool of common threads (200 in total), after which time the thread becomes available for any other request from any process rather than staying reserved.

For the configuration changes to take effect, the nameservice on the admin node should be restarted, followed by the affected processes (for example, indexer) on the remote nodes.
Propiedades

Id. de artículo: 2390936 - Última revisión: 5 oct. 2011 - Revisión: 1

Comentarios