com.bigdata.bfs.FileVersionDeleter Maven / Gradle / Ivy
package com.bigdata.bfs;
import com.bigdata.btree.IIndex;
import com.bigdata.btree.ITuple;
import com.bigdata.btree.filter.Advancer;
import com.bigdata.btree.filter.TupleUpdater;
import com.bigdata.btree.keys.KeyBuilder;
import com.bigdata.sparse.IRowStoreConstants;
import com.bigdata.sparse.KeyDecoder;
import com.bigdata.sparse.TimestampChooser;
import com.bigdata.sparse.ValueType;
import com.bigdata.sparse.TPS.TPV;
import com.bigdata.util.Bytes;
/**
* A procedure that performs a key range scan, marking all non-deleted
* versions within the key range as deleted (by storing a null property
* value for the {@link FileMetadataSchema#VERSION}).
*
* @author Bryan Thompson
* @version $Id$
*/
public class FileVersionDeleter extends TupleUpdater {
private static final long serialVersionUID = -3084247827028423921L;
/**
* The timestamp specified to the ctor.
*/
private final long timestamp;
/**
* The timestamp chosen on the server (not serialized).
*/
transient private long choosenTimestamp;
/** false until we choose the timestamp (not serialized) */
transient private boolean didInit = false;
/** used to generate keys on the server (not serialized). */
transient private KeyBuilder keyBuilder;
/**
*
* @param timestamp
* A valid timestamp or {@link IRowStoreConstants#AUTO_TIMESTAMP}
* or {@link IRowStoreConstants#AUTO_TIMESTAMP_UNIQUE}.
*/
public FileVersionDeleter(long timestamp) {
this.timestamp = timestamp;
}
/**
* Only visits the {@link FileMetadataSchema#VERSION} columns.
*/
@Override
protected boolean isValid(ITuple tuple) {
final KeyDecoder keyDecoder = new KeyDecoder(tuple.getKey());
final String name = keyDecoder.getColumnName();
if(!name.equals(FileMetadataSchema.VERSION)) return false;
return true;
}
/**
* Appends a new tuple into the index whose key uses the
* {@link #choosenTimestamp} and whose value is an encoded null.
* This is interepreted as a "deleted" file version tuple.
*
* Note: The old tuple is not removed so you can continue to access the
* historical version explicitly.
*
* @todo unit tests to verify that old versions are left behind on overflow
* (per the appropriate policy).
*
* @todo unit test to verify continued access to the historical version.
*
* FIXME Unit test to verify that the iterator correctly visits all file
* versions in the range. In particular, we will be seeing each file version
* tuple in the key range. There may be more than one of these, right?
* Therefore we need to use the {@link Advancer} pattern to skip beyond the
* last possible such tuple (which will in fact be the one that we write on
* here!).
*
* One way to approach this is a logical row scan with a column name filter
* that only visits the "version" column and an {@link TupleUpdater} and writes a
* new tuple per the code below. We just need to make sure that we either do
* not then visit the new tuple or that we ignore a "version" row whose
* value is [null].
*/
protected void update(final IIndex ndx, final ITuple tuple) {
final byte[] key = tuple.getKey();
if (!didInit) {
// choose timestamp for any writes.
choosenTimestamp = TimestampChooser.chooseTimestamp(ndx,
timestamp);
keyBuilder = new KeyBuilder(key.length);
didInit = true;
}
// // remove the old tuple.
// ndx.remove(key);
// copy everything from the key except the old timestamp.
keyBuilder.reset().append(key, 0/* off */,
key.length - Bytes.SIZEOF_LONG);
// append the new timestamp.
keyBuilder.append(choosenTimestamp);
/*
* insert a new tuple using the new timestamp. the [null] will be
* interpreted as a deleted file version.
*/
ndx.insert(key, ValueType.encode(null));
}
}