com.bigdata.mdi.MetadataIndexView Maven / Gradle / Ivy
/*
Copyright (C) SYSTAP, LLC DBA Blazegraph 2006-2016. All rights reserved.
Contact:
SYSTAP, LLC DBA Blazegraph
2501 Calvert ST NW #106
Washington, DC 20008
[email protected]
This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation; version 2 of the License.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
along with this program; if not, write to the Free Software
Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
*/
/*
* Created on Feb 20, 2008
*/
package com.bigdata.mdi;
import org.apache.log4j.Logger;
import com.bigdata.btree.AbstractBTree;
import com.bigdata.btree.DelegateIndex;
import com.bigdata.btree.ILinearList;
import com.bigdata.btree.IRangeQuery;
import com.bigdata.btree.ITuple;
import com.bigdata.btree.ITupleIterator;
import com.bigdata.btree.IndexMetadata;
import com.bigdata.cache.LRUCache;
import com.bigdata.mdi.MetadataIndex.MetadataIndexMetadata;
/**
* The extension semantics for the {@link IMetadataIndex} are implemented by
* this class.
*
* @author Bryan Thompson
* @version $Id$
*/
public class MetadataIndexView extends DelegateIndex implements IMetadataIndex {
protected static final Logger log = Logger.getLogger(MetadataIndexView.class);
// protected static final boolean INFO = log.isInfoEnabled();
// protected static final boolean DEBUG = log.isDebugEnabled();
private final AbstractBTree delegate;
/**
* true iff this is a read-only view. this is used to
* conditionally enable caching of de-serialized objects in
* {@link #getLocatorAtIndex(int)}. We can't cache those objects if
* the view is mutable!
*/
final private boolean readOnly;
public MetadataIndexView(final AbstractBTree delegate) {
super(delegate);
this.delegate = delegate;
this.readOnly = delegate.isReadOnly();
}
public MetadataIndexMetadata getIndexMetadata() {
return (MetadataIndexMetadata) super.getIndexMetadata();
}
public IndexMetadata getScaleOutIndexMetadata() {
return getIndexMetadata().getManagedIndexMetadata();
}
public PartitionLocator get(byte[] key) {
/*
* Note: The cast of the key to Object triggers automatic
* de-serialization using the ITupleSerializer.
*/
return (PartitionLocator) delegate.lookup((Object) key);
}
/**
* The method is used to discover the locator for the index partition within
* which the key would be found.
*/
public PartitionLocator find(byte[] key) {
return find_with_indexOf(key);
// return find_with_iterator(key);
}
/**
* The implementation uses an iterator with a capacity of ONE (1) and a
* {@link IRangeQuery#REVERSE} scan. This approach can be used with a
* key-range partitioned metadata index.
*
* @todo test this variant and keep it on hand for the key-range partitioned
* metadata index (it does not quite work yet).
*/
private PartitionLocator find_with_iterator(byte[] key) {
final ITupleIterator itr = delegate.rangeIterator(
null/* fromKey */, key/* toKey */, 1/* capacity */,
IRangeQuery.VALS | IRangeQuery.REVERSE, null/* filter */);
if (!itr.hasNext()) {
/*
* There are no index partitions defined since none span the key.
*
* Note: This is never a good sign. There should always be at least
* one index partition having an empty byte[] as its left separator
* and a null as its right separator.
*/
log.warn("No index partitions defined? name="
+ getIndexMetadata().getName());
return null;
}
return itr.next().getObject();
}
/**
* This implementation depends on the {@link ILinearList} API and therefore
* can not be used with a key-range partitioned metadata index.
*/
private PartitionLocator find_with_indexOf(final byte[] key) {
final long index;
if (key == null) {
// use the index of the last partition.
index = delegate.getEntryCount() - 1;
} else {
// locate the index partition for that key.
index = findIndexOf(key);
}
if(index == -1) {
return null;
}
if (readOnly && index < Integer.MAX_VALUE)
return getAndCacheLocatorAtIndex((int) index);
else
return getLocatorAtIndex(index);
}
/**
* Uses a cache to reduce {@link PartitionLocator} de-serialization costs.
*
* @param index
*
* @return
*/
private PartitionLocator getAndCacheLocatorAtIndex(final int index) {
final Integer key = Integer.valueOf(index);
synchronized (locatorCache) {
PartitionLocator locator = locatorCache.get(key);
if (locator == null) {
locator = getLocatorAtIndex(index);
locatorCache.put(key, locator, false/* dirty */);
}
return locator;
}
}
/**
* {@link #find(byte[])} was a hot spot with the costs being primarily the
* the de-serialization of the {@link PartitionLocator} from the
* {@link ITuple} so I setup this {@link LRUCache}. The keys are the index
* position in the B+Tree. The values are de-serialized
* {@link PartitionLocator}s. That seems to do the trick for now, but a
* different approach may be required if {@link #find(byte[])} is changed to
* use an iterator.
*
* Note: This exploits the fact that the indexOf the tuple in the metadata
* index is also its partition identifier. This is true because the
* partition identifier is based on a one-up index local counter. This means
* that the index partition tuples are created in strict order.
*/
private LRUCache locatorCache = new LRUCache(
1000);
/**
* Remove the locator from the {@link #locatorCache}. It will be re-read on
* demand from the {@link #delegate}.
*/
public void staleLocator(PartitionLocator locator) {
locatorCache.remove(locator.getPartitionId());
}
/**
* Looks up and de-serializes the {@link PartitionLocator} at the given
* index.
*
* @param index
*
* @return
*/
private PartitionLocator getLocatorAtIndex(final long index) {
final ITuple tuple = delegate.valueAt(index,
delegate.getLookupTuple());
return tuple.getObject();
}
/**
* Find the index of the partition spanning the given key. It is only used
* by {@link #find_with_indexOf(byte[])} and does not scale-out because of a
* dependency on the {@link ILinearList} API.
*
* @return The index of the partition spanning the given key or
* -1 iff there are no partitions defined.
*
* @exception IllegalStateException
* if there are partitions defined but no partition spans the
* key. In this case the {@link MetadataIndex} lacks an entry
* for the key new byte[]{}.
*/
private long findIndexOf(final byte[] key) {
long pos = delegate.indexOf(key);
if (pos < 0) {
/*
* the key lies between the partition separators and represents the
* insert position. we convert it to an index and subtract one to
* get the index of the partition that spans this key.
*/
pos = -(pos+1);
if(pos == 0) {
if(delegate.getEntryCount() != 0) {
throw new IllegalStateException(
"Partition not defined for empty key.");
}
return -1;
}
pos--;
return pos;
} else {
/*
* exact hit on a partition separator, so we choose the entry with
* that key.
*/
return pos;
}
}
}