OPS - Reconfiguration and Instance Startup in OPS (Doc ID 114566.1)

PURPOSE
-------

The purpose of this document is to explain how reconfiguration occurs
when an Oracle parallel server instance is started, why startup may 
appear to hang during reconfiguration, and how other nodes are affected.

 
SCOPE & APPLICATION
-------------------

This document is intended for parallel server database administrators
that would like to understand a little more of what happens during 
reconfiguration, usually when an Oracle instance registers with the 
cluster.


RECONFIGURATION AND INSTANCE STARTUP IN OPS
-------------------------------------------
 
When starting an Oracle instance, you might notice that 'reconfiguration'
messages appear in the alert log.  At the same time you may notice that
the lmon trace goes through reconfiruation, startup appears to hang 
temporarily and other nodes temporarily hang as well.  This is normal 
behavior and is especially evident when a hashed locking strategy is 
used.  

Reconfiguration can occur when:

- An OPS instance is started.
- An OPS instance dies.
- A node is halted.

What happens during reconfiguration (on all nodes):

1. The lock database is frozen, essentially all global locking activity
comes to a halt.

2. Oracle cleans up all old resources, locks, and groups on all nodes.

3. Oracle re-establishes communication channels across nodes.

4. Oracle redistributes and grants any grantable locks as needed.

5. The DLM database is unfrozen and global locking activity can commence.

We actually do a little more than this but these 5 activities are the
most resource intensive and time consuming.  This can take longer 
depending on the locking strategy implemented, resources, and the 
number of locks that need to be allocated.  A fixed locking strategy 
specifically takes longer because all locks are granted during 
reconfiguration whereas when using a dba or releasable locking strategy
all locks are distributed as needed and are not allocated during 
reconfiguration.