com.bigdata.sparse.SparseRowStore 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
*/
package com.bigdata.sparse;
import java.text.RuleBasedCollator;
import java.util.Iterator;
import java.util.LinkedList;
import java.util.List;
import java.util.Map;
import org.apache.log4j.Logger;
import com.bigdata.bfs.BigdataFileSystem;
import com.bigdata.btree.IIndex;
import com.bigdata.btree.IRangeQuery;
import com.bigdata.btree.ITuple;
import com.bigdata.btree.IndexMetadata;
import com.bigdata.btree.keys.CollatorEnum;
import com.bigdata.btree.keys.IKeyBuilder;
import com.bigdata.journal.IIndexManager;
import com.bigdata.journal.ITimestampService;
import com.bigdata.journal.Journal;
import com.bigdata.btree.AbstractBTree;
import com.bigdata.journal.TimestampUtility;
import com.bigdata.rdf.store.AbstractTripleStore;
import com.bigdata.relation.RelationSchema;
import com.bigdata.service.ndx.IClientIndex;
import cutthecrap.utils.striterators.Resolver;
import cutthecrap.utils.striterators.Striterator;
/**
* A client-side class that knows how to use an {@link IIndex} to provide an
* efficient data model in which a logical row is stored as one or more entries
* in the {@link IIndex}. Operations are provided for atomic read and write of
* logical row. While the scan operations are always consistent (they will never
* reveal data from a row that undergoing concurrent modification), they do NOT
* cause concurrent atomic row writes to block. This means that rows that would
* be visited by a scan MAY be modified before the scan reaches those rows and
* the client will see the updates.
*
* The {@link SparseRowStore} requires that you declare the {@link KeyType} for
* primary key so that it may impose a consistent total ordering over the
* generated keys in the index.
*
* There is no intrinsic reason why column values must be strongly typed.
* Therefore, by default column values are loosely typed. However, column values
* MAY be constrained by a {@link Schema}.
*
* This class builds keys using the sparse row store design pattern. Each
* logical row is modeled as an ordered set of index entries whose keys are
* formed as:
*
*
*
*
* [schemaName][primaryKey][columnName][timestamp]
*
*
*
*
*
* and the values are the value for a given column for that primary key.
*
*
*
*
*
* Timestamps are either generated by the application, in which case they define
* the semantics of a write-write conflict, or on write by the index. In the
* latter case, write-write conflicts never arise. Regardless of how timestamps
* are generated, the use of the timestamp in the key requires that
* applications specify filters that are applied during row scans to limit the
* data points actually returned as part of the row. For example, only returning
* the most recent column values no later than a given timestamp for all columns
* for some primary key.
*
*
*
*
*
* For example, assuming records with the following columns
*
*
* - Id
* - Name
* - Employer
* - DateOfHire
*
*
* would be represented as a series of index entries as follows:
*
*
*
*
*
* [employee][12][DateOfHire][t0] : [4/30/02]
* [employee][12][DateOfHire][t1] : [4/30/05]
* [employee][12][Employer][t0] : [SAIC]
* [employee][12][Employer][t1] : [SYSTAP]
* [employee][12][Id][t0] : [12]
* [employee][12][Name][t0] : [Bryan Thompson]
*
*
*
*
*
* In order to read the logical row whose last update was t0,
* the caller would specify t0 as the toTime of interest.
* The values read in this example would be {<DateOfHire, t0, 4/30/02>,
* <Employer, t0, SAIC>, <Id, t0, 12>, <Name, t0, Bryan
* Thompson>}.
*
*
* Likewise, in order to read the logical row whose last update was
* <code>t1</code> the caller would specify
* <code>t1</code> as the toTime of interest. The values
* read in this example would be {<DateOfHire, t1, 4/30/05>, <Employer,
* t0, SYSTAP>, <Id, t0, 12>, <Name, t0, Bryan Thompson>}. Notice
* that values written at <code>t0</code> and not overwritten or
* deleted by <code>t1</code> are present in the resulting logical
* row.
*
*
* Note: Very large objects should be stored in the {@link BigdataFileSystem}
* (distributed, atomic, versioned, chunked file system) and the identifier for
* that object can then be stored in the row store.
*
*
* @author Bryan Thompson
* @version $Id$
*
* FIXME write a REST service using Json to interchange data with the
* {@link SparseRowStore}. A caching layer in the web app could be used to
* reduce any hotspots.
*
* @author Bryan Thompson
* @version $Id$
*/
public class SparseRowStore implements IRowStoreConstants {
protected static final Logger log = Logger.getLogger(SparseRowStore.class);
// /**
// * True iff the {@link #log} level is INFO or less.
// */
// final protected boolean INFO = log.isInfoEnabled();
//
// /**
// * True iff the {@link #log} level is DEBUG or less.
// */
// final protected boolean DEBUG = log.isDebugEnabled();
static final String UTF8 = "UTF-8";
private final IIndex ndx;
/**
* The backing index.
*/
public IIndex getIndex() {
return ndx;
}
/**
* Create a client-side abstraction that treats an {@link IIndex} as a
* {@link SparseRowStore}.
*
* Note: When creating the backing index you MUST specify the split handler
* to ensure that dynamic sharding does not break logical rows, e.g.:
*
*
* md.setSplitHandler(LogicalRowSplitHandler.INSTANCE);
*
*
* Note: The JDK {@link RuleBasedCollator} embeds nul bytes in
* the Unicode sort keys. This makes them unsuitable for the row store which
* can not locate the start of the column name if there are embedded
* nuls in the primaryKey. Therefore, if you are using the
* {@link CollatorEnum#JDK} as your default collator, then you MUST override
* the {@link IndexMetadata} for the row store to use either an ASCII
* collator or the ICU collator. In general, the ICU collator is superior to
* the JDK collator and will be used by default. The ASCII collator is not
* ideal since non-ascii distinctions will be lost, but it is better than
* being unable to decode the data in the row store.
*
* @param ndx
* The index.
*/
public SparseRowStore(final IIndex ndx) {
if (ndx == null)
throw new IllegalArgumentException();
this.ndx = ndx;
}
// /**
// * Used to encode and decode tuples for the {@link SparseRowStore} index.
// * Each tuple corresponds to a {@link ITPV timestamped property value} for
// * some {@link Schema}.
// *
// * @todo there needs to be some way to lookup the {@link Schema} from the
// * schema name as encoded in the key. One possibility is to register
// * the known {@link Schema} against a static factory. Another is to
// * have the known {@link Schema} registered in the
// * {@link IndexMetadata} for the index backing the
// * {@link SparseRowStore} (much like an extSer integration). The
// * schema can be resolved using its encoded bytes as the key and the
// * Unicode text of the schema name can be persisted in the
// * {@link Schema}'s data. See the TPSTupleSerializer also.
// *
// * @author Bryan Thompson
// * @version $Id$
// */
// public class TPVTupleSerializer implements ITupleSerializer {
//
// /**
// * De-serialization ctor.
// */
// public TPVTupleSerializer() {
//
// }
//
// public byte[] serializeKey(TPV t) {
//
// if(t == null) throw new IllegalArgumentException();
//
// IKeyBuilder keyBuilder = getKeyBuilderFactory().getKeyBuilder();
//
// final byte[] key = t.getSchema().getKey(keyBuilder, t.primaryKey, t.getName(), t.getTimestamp());
//
// return key;
// }
//
// /**
// * De-serializes as much of the key as possible.
// *
// * @see KeyDecoder
// */
// public KeyDecoder deserializeKey(ITuple tuple) {
//
// return new KeyDecoder(tuple.getKey());
//
// }
//
// public byte[] serializeVal(TPV t) {
//
// return ValueType.encode(t.getValue());
//
// }
//
// public ITPV deserialize(ITuple tuple) {
//
// final KeyDecoder keyDecoder = new KeyDecoder(tuple.getKey());
//
// final Schema schema = resolveSchema(keyDecoder.getSchemaBytes());
//
// final Object value = ValueType.decode(tuple.getValue());
//
// final TPV t = new TPV(schema, keyDecoder.getColumnName(), keyDecoder
// .getTimestamp(), value);
//
// return t;
//
// }
//
// }
/**
* Verifies the given arguments.
*/
final static void assertArgs(final Schema schema, final Object primaryKey,
final long fromTime, final long toTime) {
if (schema == null)
throw new IllegalArgumentException("schema");
if (primaryKey == null)
throw new IllegalArgumentException("primaryKey");
if (fromTime == CURRENT_ROW) {
throw new IllegalArgumentException(
"fromTime MAY NOT be 'CURRENT_ROW'");
}
if (fromTime < MIN_TIMESTAMP) {
throw new IllegalArgumentException("fromTime less than MIN_TIMESTAMP");
}
if (toTime != CURRENT_ROW) {
if (fromTime >= toTime) {
throw new IllegalArgumentException("from/to time out of order");
}
}
}
/**
* Verifies the writeTime.
*
* @param writeTime
*/
final static void assertWriteTime(long writeTime) {
if (writeTime == AUTO_TIMESTAMP)
return;
if (writeTime == AUTO_TIMESTAMP_UNIQUE)
return;
if (writeTime < MIN_TIMESTAMP)
throw new IllegalArgumentException();
}
/**
* Validates the column name productions
*/
final static void assertPropertyNames(final Map propertySet) {
if (propertySet == null)
throw new IllegalArgumentException();
final Iterator itr = propertySet.keySet().iterator();
while (itr.hasNext()) {
final String col = itr.next();
// validate the column name production.
NameChecker.assertColumnName(col);
}
}
/**
* Return the current binding for the named property.
*
* @param schema
* The {@link Schema} governing the logical row.
* @param primaryKey
* The primary key that identifies the logical row.
* @param name
* The property name.
* @return The current binding -or- null iff the property is
* not bound.
*
* @todo this can be optimized and should use its own stored procedure. See
* {@link AbstractAtomicRowReadOrWrite#getCurrentValue(IIndex, Schema, Object, String)}
*/
public Object get(final Schema schema, final Object primaryKey, final String name) {
final TPS tps = (TPS) read(schema, primaryKey, MIN_TIMESTAMP,
CURRENT_ROW, new SingleColumnFilter(name));
if (tps == null) {
return null;
}
return tps.get(name).getValue();
}
/**
* Read the most recent logical row from the index.
*
* @param schema
* The {@link Schema} governing the logical row.
* @param primaryKey
* The primary key that identifies the logical row.
*
* @return The data for the current state of that logical row -or-
* null IFF there are no property values for that
* logical row (including no deleted property values, no property
* values that are excluded due to their timestamps, and no property
* values that are excluded due to a property name filter). A
* null return is a strong guarantee that NO data
* existed in the row store and that time of the read for the given
* schema and primaryKey.
*/
public Map read(final Schema schema, final Object primaryKey) {
final TPS tps = (TPS) read(schema, primaryKey, MIN_TIMESTAMP,
CURRENT_ROW, null/* filter */);
if (tps == null) {
return null;
}
return tps.asMap();
}
/**
* Read the most recent logical row from the index.
*
* @param schema
* The {@link Schema} governing the logical row.
* @param primaryKey
* The primary key that identifies the logical row.
* @param filter
* An optional filter.
*
* @return The data for the current state of that logical row -or-
* null IFF there are no property values for that
* logical row (including no deleted property values, no property
* values that are excluded due to their timestamps, and no property
* values that are excluded due to a property name filter). A
* null return is a strong guarantee that NO data
* existed in the row store and that time of the read for the given
* schema and primaryKey.
*/
public Map read(final Schema schema,
final Object primaryKey, final INameFilter filter) {
final TPS tps = (TPS) read(schema, primaryKey, MIN_TIMESTAMP,
CURRENT_ROW, filter);
if (tps == null) {
return null;
}
return tps.asMap();
}
/**
* Read a logical row from the index.
*
* @param schema
* The {@link Schema} governing the logical row.
* @param primaryKey
* The primary key that identifies the logical row.
* @param fromTime
* The first timestamp for which timestamped property values will
* be accepted.
* @param toTime
* The first timestamp for which timestamped property values will
* NOT be accepted -or- {@link IRowStoreConstants#CURRENT_ROW} to
* accept only the most current binding whose timestamp is GTE
* fromTime.
* @param filter
* An optional filter that may be used to select values for
* property names accepted by the filter.
*
* @return The data in that row -or- null IFF there are no
* property values for that logical row (including no deleted
* property values, no property values that are excluded due to
* their timestamps, and no property values that are excluded due to
* a property name filter). A null return is a strong
* guarantee that NO data existed in the row store and that time of
* the read for the given schema and primaryKey.
*
* @throws IllegalArgumentException
* if the schema is null.
* @throws IllegalArgumentException
* if the primaryKey is null.
* @throws IllegalArgumentException
* if the fromFrom and or toTime are invalid.
*
* @see ITimestampPropertySet#asMap(), return the most current bindings.
* @see ITimestampPropertySet#asMap(long)), return the most current bindings
* as of the specified timestamp.
*
* @see IRowStoreConstants#CURRENT_ROW
* @see IRowStoreConstants#MIN_TIMESTAMP
* @see IRowStoreConstants#MAX_TIMESTAMP
*/
public ITPS read(final Schema schema, final Object primaryKey,
final long fromTime, final long toTime, final INameFilter filter) {
assertArgs(schema, primaryKey, fromTime, toTime);
if (log.isInfoEnabled()) {
String ts = "N/A";
if (getIndex() instanceof IClientIndex) {
ts = TimestampUtility.toString(((IClientIndex) getIndex())
.getTimestamp());
} else if (getIndex() instanceof AbstractBTree) {
ts = TimestampUtility.toString(((AbstractBTree) getIndex())
.getLastCommitTime());
}
log.info("ts=" + ts + ", schema=" + schema.getName()
+ ", primaryKey=" + primaryKey + ", fromTime=" + fromTime
+ ", toTime=" + toTime + ", filter="
+ (filter == null ? "N/A" : filter.getClass().getName()));
}
final AtomicRowRead proc = new AtomicRowRead(schema, primaryKey,
fromTime, toTime, filter);
final byte[] key = schema.fromKey(
ndx.getIndexMetadata().getKeyBuilder(), primaryKey).getKey();
// Submit the atomic read operation.
return (TPS) ndx.submit(key, proc);
}
/**
* Atomic write with atomic read-back of the post-update state of the
* logical row.
*
* Note: In order to cause a column value for row to be deleted you MUST
* specify a null column value for that column.
*
* Note: the value of the primaryKey is written each time the
* logical row is updated and timestamp associate with the value for the
* primaryKey property tells you the timestamp of each row revision.
*
* @param schema
* The {@link Schema} governing the logical row.
*
* @param propertySet
* The column names and values for that row.
*
* @return The result of an atomic read on the post-update state of the
* logical row. Only the most current bindings will be present for
* each property.
*/
public Map write(final Schema schema,
final Map propertySet) {
return write(schema, propertySet, AUTO_TIMESTAMP_UNIQUE, null/* filter */,
null/* precondition */).asMap();
}
/**
* Atomic write with atomic read-back of the post-update state of the
* logical row.
*
* @param schema
* The {@link Schema} governing the logical row.
* @param propertySet
* The column names and values for that row.
* @param writeTime
* The timestamp to use for the row -or-
* {@link IRowStoreConstants#AUTO_TIMESTAMP} if the timestamp
* will be generated by the server -or-
* {@link IRowStoreConstants#AUTO_TIMESTAMP_UNIQUE} if a
* federation-wide unique timestamp will be generated by the
* server.
*
* @return The result of an atomic read on the post-update state of the
* logical row. Only the most current bindings will be present for
* each property.
*/
public Map write(final Schema schema,
final Map propertySet, final long writeTime) {
return write(schema, propertySet, writeTime, null/* filter */, null/* precondition */)
.asMap();
}
/**
* Atomic write with atomic read of the then current post-condition state of
* the logical row.
*
* Note: In order to cause a column value for row to be deleted you MUST
* specify a null column value for that column. A
* null will be written under the key for the column value
* with a new timestamp. This is interpreted as a deleted property value
* when the row is simplified as a {@link Map}. If you examine the
* {@link ITPS} you can see the {@link ITPV} with the null
* value and the timestamp of the delete.
*
* Note: the value of the primaryKey is written each time the
* logical row is updated and timestamp associate with the value for the
* primaryKey property tells you the timestamp of each row revision.
*
* Note: If the caller specified a timestamp, then that timestamp is
* used by the atomic read. If the timestamp was assigned by the server,
* then the server assigned timestamp is used by the atomic read.
*
* Note: You can verify pre-conditions for the logical row on the server.
* Among other things this could be used to reject an update if someone has
* modified the logical row since you last read some value.
*
* @param schema
* The {@link Schema} governing the logical row.
* @param propertySet
* The column names and values for that row. The primaryKey as
* identified by the {@link Schema} MUST be present in the
* propertySet.
* @param writeTime
* The timestamp to use for the row -or-
* {@link IRowStoreConstants#AUTO_TIMESTAMP} if the timestamp
* will be generated by the server -or-
* {@link IRowStoreConstants#AUTO_TIMESTAMP_UNIQUE} if a
* federation-wide unique timestamp will be generated by the
* server.
* @param filter
* An optional filter used to select the property values that
* will be returned (this has no effect on the atomic write).
* @param precondition
* When present, the pre-condition state of the row will be read
* and offered to the {@link IPrecondition}. If the
* {@link IPrecondition} fails, then the atomic write will NOT be
* performed and the pre-condition state of the row will be
* returned. If the {@link IPrecondition} succeeds, then the
* atomic write will be performed and the post-condition state of
* the row will be returned. Use {@link TPS#isPreconditionOk()}
* to determine whether or not the write was performed.
*
* @return The result of an atomic read on the post-update state of the
* logical row -or- null iff there is no data for the
* primaryKey (per the contract for an atomic read).
*
* If an optional {@link IPrecondition} was specified and the
* {@link IPrecondition} was NOT satisfied, then the write
* operation was NOT performed and the result is the pre-condition
* state of the logical row (which, again, will be null
* IFF there is NO data for the primaryKey).
*
* @see ITPS#getWriteTimestamp()
*/
public TPS write(final Schema schema,
final Map propertySet, final long writeTime,
final INameFilter filter, final IPrecondition precondition) {
return write(schema, propertySet, MIN_TIMESTAMP, CURRENT_ROW,
writeTime, filter, precondition);
}
/**
* Atomic write with atomic read of the post-condition state of the logical
* row.
*
* Note: In order to cause a column value for row to be deleted you MUST
* specify a null column value for that column. A
* null will be written under the key for the column value
* with a new timestamp. This is interpreted as a deleted property value
* when the row is simplified as a {@link Map}. If you examine the
* {@link ITPS} you can see the {@link ITPV} with the null
* value and the timestamp of the delete.
*
* Note: the value of the primaryKey is written each time the
* logical row is updated and timestamp associate with the value for the
* primaryKey property tells you the timestamp of each row revision.
*
* Note: If the caller specified a timestamp, then that timestamp is
* used by the atomic read. If the timestamp was assigned by the server,
* then the server assigned timestamp is used by the atomic read.
*
* Note: You can verify pre-conditions for the logical row on the server.
* Among other things this could be used to reject an update if someone has
* modified the logical row since you last read some value.
*
* @param schema
* The {@link Schema} governing the logical row.
* @param propertySet
* The column names and values for that row. The primaryKey as
* identified by the {@link Schema} MUST be present in the
* propertySet.
* @param fromTime
* During pre-condition and post-condition reads, the
* first timestamp for which timestamped property values will be
* accepted.
* @param toTime
* During pre-condition and post-condition reads, the
* first timestamp for which timestamped property values will NOT
* be accepted -or- {@link IRowStoreConstants#CURRENT_ROW} to
* accept only the most current binding whose timestamp is GTE
* fromTime.
* @param writeTime
* The timestamp to use for the row -or-
* {@link IRowStoreConstants#AUTO_TIMESTAMP} if the timestamp
* will be generated by the server -or-
* {@link IRowStoreConstants#AUTO_TIMESTAMP_UNIQUE} if a
* federation-wide unique timestamp will be generated by the
* server.
* @param filter
* An optional filter used to select the property values that
* will be returned (this has no effect on the atomic write).
* @param precondition
* When present, the pre-condition state of the row will be read
* and offered to the {@link IPrecondition}. If the
* {@link IPrecondition} fails, then the atomic write will NOT be
* performed and the pre-condition state of the row will be
* returned. If the {@link IPrecondition} succeeds, then the
* atomic write will be performed and the post-condition state of
* the row will be returned. Use {@link TPS#isPreconditionOk()}
* to determine whether or not the write was performed.
*
* @return The result of an atomic read on the post-update state of the
* logical row, which will be null IFF there is NO
* data for the primaryKey.
*
* If an optional {@link IPrecondition} was specified and the
* {@link IPrecondition} was NOT satisified, then the write
* operation was NOT performed and the result is the pre-condition
* state of the logical row (which, again, will be null
* IFF there is NO data for the primaryKey).
*
* @throws UnsupportedOperationException
* if a property has an auto-increment type and the
* {@link ValueType} of the property does not support
* auto-increment.
* @throws UnsupportedOperationException
* if a property has an auto-increment type but there is no
* successor in the value space of that property.
*
* @see ITPS#getWriteTimestamp()
*
* @todo the atomic read back may be overkill. When you need the data is
* means that you only do one RPC rather than two. When you do not
* need the data it is just more network traffic and more complexity
* in this method signature. You can get pretty much the same result
* by doing an atomic read after the fact using the timestamp assigned
* by the server to the row (pretty much in the sense that it is
* possible for another write to explicitly specify the same timestamp
* and hence overwrite your data).
*
* @todo the timestamp could be an {@link ITimestampService} with an
* implementation that always returns a caller-given constant, another
* that uses the local system clock, another that uses the system
* clock but ensures that it never hands off the same timestamp twice
* in a row, and another than resolves the global timestamp service.
*
* it is also possible that the timestamp behavior should be defined
* by the {@link Schema} and therefore factored out of this method
* signature.
*/
public TPS write(final Schema schema,
final Map propertySet, final long fromTime,
final long toTime, final long writeTime, final INameFilter filter,
final IPrecondition precondition) {
// check before extracting the primary key.
if (schema == null)
throw new IllegalArgumentException();
// check before extracting the primary key.
if (propertySet == null)
throw new IllegalArgumentException();
// extract the primary key.
final Object primaryKey = propertySet.get(schema.getPrimaryKeyName());
// verify args.
assertArgs(schema, primaryKey, fromTime, toTime);
if (log.isInfoEnabled())
log.info("schema=" + schema.getName() + ", primaryKey="
+ primaryKey + ", timestamp=" + writeTime + ", filter="
+ (filter == null ? "N/A" : filter.getClass().getName())+
", precondition="
+ (precondition == null ? "N/A" : precondition.getClass()
.getName()));
final AtomicRowWriteRead proc = new AtomicRowWriteRead(schema,
propertySet, fromTime, toTime, writeTime, filter, precondition);
final byte[] key = schema.fromKey(
ndx.getIndexMetadata().getKeyBuilder(), primaryKey).getKey();
return (TPS) ndx.submit(key, proc);
}
/**
* Atomic delete of all property values for the current logical row.
*
* @param schema
* The schema.
* @param primaryKey
* The primary key for the logical row.
*
* @return The deleted property values.
*/
public ITPS delete(Schema schema, Object primaryKey) {
return delete(schema, primaryKey, MIN_TIMESTAMP, CURRENT_ROW,
AUTO_TIMESTAMP_UNIQUE, null/* filter */);
}
/**
* Atomic delete of all property values for the logical row. The property
* values are read atomically, each property value that is read is then
* overwritten with a null, and the read property values are
* returned.
*
* @param schema
* The schema.
* @param primaryKey
* The primary key for the logical row.
* @param fromTime
* During pre-condition and post-condition reads, the
* first timestamp for which timestamped property values will be
* accepted.
* @param toTime
* During pre-condition and post-condition reads, the
* first timestamp for which timestamped property values will NOT
* be accepted -or- {@link IRowStoreConstants#CURRENT_ROW} to
* accept only the most current binding whose timestamp is GTE
* fromTime.
* @param writeTime
* The timestamp that will be written into the "deleted" entries
* -or- {@link IRowStoreConstants#AUTO_TIMESTAMP} if the
* timestamp will be generated by the server -or-
* {@link IRowStoreConstants#AUTO_TIMESTAMP_UNIQUE} if a
* federation-wide unique timestamp will be generated by the
* server.
* @param filter
* An optional filter used to select the property values that
* will be deleted.
*
* @return The property values that were read from the store before they
* were deleted. The {@link ITPS#getWriteTimestamp()} will report
* the timestamp assigned to the deleted entries used to overwrite
* these property values in the store.
*
* @todo add optional {@link IPrecondition}.
*
* @todo unit tests.
*/
public ITPS delete(final Schema schema, Object primaryKey,
final long fromTime, final long toTime, final long writeTime,
final INameFilter filter) {
assertArgs(schema, primaryKey, fromTime, toTime);
if (log.isInfoEnabled())
log.info("schema=" + schema + ", primaryKey=" + primaryKey
+ ", timestamp=" + writeTime + ", filter="
+ (filter == null ? "N/A" : filter.getClass().getName()));
final AtomicRowDelete proc = new AtomicRowDelete(schema, primaryKey,
fromTime, toTime, writeTime, filter);
final byte[] key = schema.fromKey(
ndx.getIndexMetadata().getKeyBuilder(), primaryKey).getKey();
return (TPS) ndx.submit(key, proc);
}
/**
* A logical row scan. Each logical row will be read atomically. Only the
* current bindings for property values will be returned.
*
* @param schema
* The {@link Schema} governing the logical row.
*
* @return An iterator visiting each logical row in the specified key range.
*/
public Iterator rangeIterator(final Schema schema) {
return rangeIterator(schema, null/* fromKey */, null/* toKey */,
0/* capacity */, MIN_TIMESTAMP, CURRENT_ROW, null/* filter */);
}
/**
* A logical row scan. Each logical row will be read atomically. Only the
* current bindings for property values will be returned.
*
* @param schema
* The {@link Schema} governing the logical row.
* @param fromKey
* The value of the primary key for lower bound (inclusive) of
* the key range -or- null iff there is no lower
* bound.
* @param toKey
* The value of the primary key for upper bound (exclusive) of
* the key range -or- null iff there is no lower
* bound.
*
* @return An iterator visiting each logical row in the specified key range.
*/
public Iterator rangeIterator(final Schema schema,
final Object fromKey, final Object toKey) {
return rangeIterator(schema, fromKey, toKey, 0/* capacity */,
MIN_TIMESTAMP, CURRENT_ROW, null/* filter */);
}
/**
* A logical row scan. Each logical row will be read atomically. Only the
* current bindings for property values will be returned.
*
* @param schema
* The {@link Schema} governing the logical row.
* @param fromKey
* The value of the primary key for lower bound (inclusive) of
* the key range -or- null iff there is no lower
* bound.
* @param toKey
* The value of the primary key for upper bound (exclusive) of
* the key range -or- null iff there is no lower
* bound.
* @param filter
* An optional filter.
*
* @return An iterator visiting each logical row in the specified key range.
*/
public Iterator rangeIterator(final Schema schema,
final Object fromKey, final Object toKey, final INameFilter filter) {
return rangeIterator(schema, fromKey, toKey, 0/* capacity */,
MIN_TIMESTAMP, CURRENT_ROW, filter);
}
/**
* A logical row scan. Each logical row will be read atomically.
*
* @param schema
* The {@link Schema} governing the logical row.
* @param fromKey
* The value of the primary key for lower bound (inclusive) of
* the key range -or- null iff there is no lower
* bound.
* @param toKey
* The value of the primary key for upper bound (exclusive) of
* the key range -or- null iff there is no lower
* bound.
* @param capacity
* When non-zero, this is the maximum #of logical rows that will
* be read atomically. This is only an upper bound. The actual
* #of logical rows in an atomic read depends on a variety of
* factors.
* @param fromTime
* The first timestamp for which timestamped property values will
* be accepted.
* @param toTime
* The first timestamp for which timestamped property values will
* NOT be accepted -or- {@link IRowStoreConstants#CURRENT_ROW} to
* accept only the most current binding whose timestamp is GTE
* fromTime.
* @param nameFilter
* An optional filter used to select the property(s) of interest.
*
* @return An iterator visiting each logical row in the specified key range.
*/
@SuppressWarnings("unchecked")
public Iterator rangeIterator(final Schema schema,
Object fromKey, Object toKey, final int capacity,
final long fromTime, final long toTime, final INameFilter nameFilter) {
assertArgs(schema, Boolean.TRUE/* fake */, fromTime, toTime);
if (log.isInfoEnabled())
log.info("schema="
+ schema
+ ", fromKey="
+ fromKey
+ ", toKey="
+ toKey
+ ", capacity="
+ capacity
+ ", fromTime="
+ fromTime
+ ", toTime="
+ toTime
+ ", filter="
+ (nameFilter == null ? "N/A" : nameFilter.getClass()
.getName()));
final IKeyBuilder keyBuilder = ndx.getIndexMetadata().getKeyBuilder();
if (fromKey != null) {
// convert to an unsigned byte[].
fromKey = schema.fromKey(keyBuilder, fromKey).getKey();
}
if (toKey != null) {
// convert to an unsigned byte[].
toKey = schema.fromKey(keyBuilder, toKey).getKey();
}
/*
* If the primary key type has a fixed length (int, long, etc), then the
* successor for continuation queries must be formed by adding one to
* the last key visited. Otherwise an unsigned nul byte is appended
* (ASCII, Unicode).
*/
final boolean fixedLengthSuccessor = schema.getPrimaryKeyType().isFixedLength();
final int flags = IRangeQuery.DEFAULT
| IRangeQuery.READONLY
| (fixedLengthSuccessor ? IRangeQuery.FIXED_LENGTH_SUCCESSOR
: 0);
/*
* Setup an iterator that visits the timestamp-property-value tuples and
* a filter that aggregates logical rows into chunks.
*/
return new Striterator(ndx.rangeIterator(//
(byte[]) fromKey, //
(byte[]) toKey, //
capacity, // max #of rows to fetch at a time.
flags, //
new AtomicRowFilter(schema, fromTime, toTime, nameFilter)))
.addFilter(new Resolver() {
private static final long serialVersionUID = 1L;
@Override
protected Object resolve(Object obj) {
// resolve visited TPS from tuple.
final ITuple tuple = (ITuple) obj;
if (log.isInfoEnabled()) {
log.info("resolving TPS: " + tuple.getVisitCount());
}
return tuple.getObject();
}
});
}
/**
* Options for the {@link SparseRowStore}.
*
* @author Bryan
* Thompson
*/
public interface Options {
/**
* The schema name was originally written using a Unicode sort key.
* However, the JDK can generate Unicode sort keys with embedded nuls
* which in turn will break the logic to detect the end of the schema
* name in the key. In order to accommodate this behavior, the schema
* name is now encoded as UTF8 which also has the advantage that we can
* decode the schema name. Standard prefix compression on the B+Tree
* should make up for the larger representation of the schema name in
* the B+Tree.
*
* This change was introduced on 7/29/2010 in the trunk. When this
* property is true it breaks compatibility with earlier
* revisions of the {@link SparseRowStore}. This flag may be set to
* false for backward compatibility.
*
* @see #DEFAULT_SCHEMA_NAME_UNICODE_CLEAN
*/
String SCHEMA_NAME_UNICODE_CLEAN = Schema.class.getName()
+ ".schemaName.unicodeClean";
/**
* @see https://sourceforge.net/apps/trac/bigdata/ticket/107
*/
String DEFAULT_SCHEMA_NAME_UNICODE_CLEAN = "true";
/**
* The primary key was originally written using a Unicode sort key.
* However, the JDK generates Unicode sort keys with embedded nuls and
* that broke the logic to detect the end of the Unicode primary keys.
* In order to accommodate this behavior, the Unicode primary key is now
* encoded as UTF8 which also has the advantage that we can decode
* Unicode primary keys. Standard prefix compression on the B+Tree
* should make up for the larger representation of the Unicode primary
* key in the B+Tree.
*
* This change was introduced on 7/15/2010 in the trunk and breaks
* compatibility with earlier revisions of the {@link SparseRowStore}.
* This flag may be set to false for backward
* compatibility.
*
* @see Options#DEFAULT_PRIMARY_KEY_UNICODE_CLEAN
*/
String PRIMARY_KEY_UNICODE_CLEAN = Schema.class.getName()
+ ".primaryKey.unicodeClean";
/**
* @see https://sourceforge.net/apps/trac/bigdata/ticket/107
*/
String DEFAULT_PRIMARY_KEY_UNICODE_CLEAN = "true";
}
/**
* This is a global option since it was always false for
* historical stores.
*
* @see Options#SCHEMA_NAME_UNICODE_CLEAN
*/
final static transient boolean schemaNameUnicodeClean = Boolean
.valueOf(System.getProperty(
SparseRowStore.Options.SCHEMA_NAME_UNICODE_CLEAN,
SparseRowStore.Options.DEFAULT_SCHEMA_NAME_UNICODE_CLEAN));
/**
* This is a global option since it was always false for
* historical stores.
*
* @see Options#PRIMARY_KEY_UNICODE_CLEAN
*/
final static transient boolean primaryKeyUnicodeClean = Boolean
.valueOf(System.getProperty(
SparseRowStore.Options.PRIMARY_KEY_UNICODE_CLEAN,
SparseRowStore.Options.DEFAULT_PRIMARY_KEY_UNICODE_CLEAN));
/**
* List of namespaces, defined in the row store.
* @param tx The transaction identifier -or- timestamp if the
* {@link IIndexManager} is not a {@link Journal}.
* @return List of namespaces
*/
public List getNamespaces(long tx) {
// the triple store namespaces.
final List namespaces = new LinkedList();
// scan the relation schema in the global row store.
@SuppressWarnings("unchecked")
final Iterator itr = (Iterator)
rangeIterator(RelationSchema.INSTANCE);
while (itr.hasNext()) {
// A timestamped property value set is a logical row with
// timestamped property values.
final ITPS tps = itr.next();
// If you want to see what is in the TPS, uncomment this.
// System.err.println(tps.toString());
// The namespace is the primary key of the logical row for the
// relation schema.
final String namespace = (String) tps.getPrimaryKey();
// Get the name of the implementation class
// (AbstractTripleStore, SPORelation, LexiconRelation, etc.)
final String className = (String) tps.get(RelationSchema.CLASS)
.getValue();
if (className == null) {
// Skip deleted triple store entry.
continue;
}
try {
final Class cls = Class.forName(className);
if (AbstractTripleStore.class.isAssignableFrom(cls)) {
// this is a triple store (vs something else).
namespaces.add(namespace);
}
} catch (ClassNotFoundException e) {
log.error(e, e);
}
}
return namespaces;
}
}