com.bigdata.service.ndx
Class ClientIndexViewRefactor

java.lang.Object
  com.bigdata.service.ndx.AbstractScaleOutClientIndexView
      com.bigdata.service.ndx.AbstractScaleOutClientIndexView2
          com.bigdata.service.ndx.ClientIndexViewRefactor

All Implemented Interfaces:

IAutoboxBTree, IIndex, IRangeQuery, ISimpleBTree, IAsynchronousWriteBufferFactory, IClientIndex, IScaleOutClientIndex, ISplitter

public class ClientIndexViewRefactor
extends AbstractScaleOutClientIndexView2

A client-side view of a scale-out index as of some timestamp.

This view automatically handles the split, join, or move of index partitions within the federation. The IDataService throws back a (sometimes wrapped) StaleLocatorException when it does not have a registered index as of some timestamp. If this exception is observed when the client makes a request using a cached PartitionLocator record then the locator record is stale. The client automatically fetches the locator record(s) covering the same key range as the stale locator record and the re-issues the request against the index partitions identified in those locator record(s). This behavior correctly handles index partition split, merge, and move scenarios. The implementation of this policy is limited to exactly three places in the code: AbstractDataServiceProcedureTask, PartitionedTupleIterator, and DataServiceTupleIterator.

Note that only ITx.UNISOLATED and ITx.READ_COMMITTED operations are subject to stale locators since they are not based on a historical committed state of the database. Historical read and fully-isolated operations both read from historical committed states and the locators are never updated for historical states (only the current state of an index partition is split, joined, or moved - the historical states always remain behind).

Version:

$Id: ClientIndexViewRefactor.java 2265 2009-10-26 12:51:06Z thompsonbry $

Author:

Bryan Thompson

Field Summary

Fields inherited from class com.bigdata.service.ndx.AbstractScaleOutClientIndexView

batchOnly, ERR_ABORT_TX, ERR_NEW_TX, fed, log, name, NON_BATCH_API, readConsistent, taskTimeout, timestamp, WARN

Fields inherited from interface com.bigdata.btree.IRangeQuery

ALL, CURSOR, DEFAULT, DELETED, FIXED_LENGTH_SUCCESSOR, KEYS, NONE, PARALLEL, READONLY, REMOVEALL, REVERSE, VALS

Constructor Summary

ClientIndexViewRefactor(AbstractScaleOutFederation fed, String name, long timestamp, IMetadataIndex metadataIndex)
Create a view on a scale-out index.

Method Summary

protected void

runTasks(boolean parallel, ArrayList<com.bigdata.service.ndx.AbstractDataServiceProcedureTask> tasks) Runs a set of tasks.

Methods inherited from class com.bigdata.service.ndx.AbstractScaleOutClientIndexView2

getRecursionDepth, submit, submit, submit

Methods inherited from class com.bigdata.service.ndx.AbstractScaleOutClientIndexView

contains, contains, getCounter, getCounters, getDataService, getFederation, getIndexMetadata, getMetadataIndex, getMetadataIndexMetadata, getMetadataService, getName, getResourceMetadata, getThreadPool, getTimestamp, getTupleSerializer, insert, insert, locatorScan, lookup, lookup, newWriteBuffer, rangeCount, rangeCount, rangeCountExact, rangeCountExactWithDeleted, rangeIterator, rangeIterator, rangeIterator, remove, remove, splitKeys, splitKeys, staleLocator, submit, submit, submit, toString

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, wait, wait, wait

Constructor Detail

ClientIndexViewRefactor

public ClientIndexViewRefactor(AbstractScaleOutFederation fed,
                               String name,
                               long timestamp,
                               IMetadataIndex metadataIndex)

Create a view on a scale-out index.

Parameters:

fed - The federation containing the index.

name - The index name.

timestamp - A transaction identifier, ITx.UNISOLATED for the unisolated index view, ITx.READ_COMMITTED, or timestamp for a historical view no later than the specified timestamp.

metadataIndex - The IMetadataIndex for the named scale-out index as of that timestamp. Note that the IndexMetadata on this object contains the template IndexMetadata for the scale-out index partitions.

Method Detail

runTasks

protected void runTasks(boolean parallel,
                        ArrayList<com.bigdata.service.ndx.AbstractDataServiceProcedureTask> tasks)

Runs a set of tasks.

Note: If AbstractScaleOutClientIndexView2.getRecursionDepth() evaluates to a value larger than zero then the task(s) will be forced to execute in the caller's thread.

StaleLocatorExceptions are handled by the recursive application of submit(). These recursively submitted tasks are forced to run in the caller's thread by incrementing the AbstractScaleOutClientIndexView2.getRecursionDepth() counter. This is done to prevent the thread pool from becoming deadlocked as threads wait on threads handling stale locator retries. The deadlock situation arises as soon as all threads in the thread pool are waiting on stale locator retries as there are no threads remaining to process those retries.

Specified by:

runTasks in class AbstractScaleOutClientIndexView2

Parameters:

parallel - true iff the tasks MAY be run in parallel.

tasks - The tasks to be executed.