package com.bigdata.resources;
/**
* The different actions that we can take to handle an index partition during
* overflow processing.
*
* @author <a href="mailto:thompsonbry@users.sourceforge.net">Bryan Thompson</a>
* @version $Id$
*
* @see OverflowSubtaskEnum
*/
public enum OverflowActionEnum {
/**
* Copy all tuples on the old journal to the new journal during
* synchronous overflow processing.
*/
Copy,
/**
* Copy the tuples from the last commit point on the old journal into an
* index segment, and may incorporate tuples from zero or more additional
* sources in the view. Builds are done both in order to improve read
* performance and to release dependencies on older journals.
* <p>
* The #of sources to use in a build is chosen in order to keep the build
* operation fast while also minimizing the #of sources that are accumulated
* in the view over time and thereby reducing the frequency with which a
* compacting merge must be performed.
* <p>
* If all sources in the view are used, then a build has the same semantics
* as a {@link #Merge}, but it is still reported as a build since actions
* selected as builds tend to be lighter weight even when all sources are
* still used (for example, consider the first overflow event, where there
* is only one source in the view - while that could be called a merge, the
* practice is to call it a build).
*/
Build,
/**
* Compacting merge of the sources for the index partition into a single
* index segment. Compacting merges are done to improve read performance
* and to keep index partition views from including too many distinct
* index segment sources.
*/
Merge,
/**
* Move the index partition to another data service. Note that moves may be
* initiated either to redistribute the load more equitably among the data
* services in the federation or to bring the left/right sibling of an index
* partition onto the same data service as its right/left sibling so that
* they may be joined.
*/
Move,
/**
* Join left- and right- index partition siblings which have underflowed.
*/
Join,
/**
* Split an index partition that has overflowed into 2 or more siblings.
*/
Split,
/**
* Split the first index partition for a new scale-out index into N siblings
* and distribute those siblings across the federation in order to improve
* the data distribution and improve the potential concurrency of the index.
*/
ScatterSplit,
/**
* Split an index partition receiving a lot of writes on the tail of the key
* range into 2 siblings where the left-sibling has most of the key range
* and the right-sibling has the tail of the key range.
*/
TailSplit;
}