com.bigdata.rdf.rules.RDFJoinNexus 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 Jun 25, 2008
*/
package com.bigdata.rdf.rules;
import java.util.ArrayList;
import java.util.Iterator;
import java.util.List;
import java.util.concurrent.Callable;
import java.util.concurrent.ExecutionException;
import java.util.concurrent.Future;
import org.apache.log4j.Logger;
import com.bigdata.bop.Constant;
import com.bigdata.bop.IBindingSet;
import com.bigdata.bop.IConstant;
import com.bigdata.bop.IPredicate;
import com.bigdata.bop.IVariable;
import com.bigdata.bop.IVariableOrConstant;
import com.bigdata.bop.Var;
import com.bigdata.bop.joinGraph.IEvaluationPlanFactory;
import com.bigdata.bop.joinGraph.IRangeCountFactory;
import com.bigdata.btree.keys.IKeyBuilder;
import com.bigdata.btree.keys.ISortKeyBuilder;
import com.bigdata.btree.keys.KeyBuilder;
import com.bigdata.journal.IIndexManager;
import com.bigdata.rdf.inf.Justification;
import com.bigdata.rdf.internal.IV;
import com.bigdata.rdf.internal.VTE;
import com.bigdata.rdf.internal.impl.TermId;
import com.bigdata.rdf.lexicon.LexiconRelation;
import com.bigdata.rdf.model.BigdataValue;
import com.bigdata.rdf.relation.rule.BindingSetSortKeyBuilder;
import com.bigdata.rdf.spo.SPO;
import com.bigdata.rdf.spo.SPORelation;
import com.bigdata.rdf.spo.SPOSortKeyBuilder;
import com.bigdata.rdf.store.AbstractTripleStore;
import com.bigdata.relation.IMutableRelation;
import com.bigdata.relation.IRelation;
import com.bigdata.relation.accesspath.IAccessPath;
import com.bigdata.relation.accesspath.IAsynchronousIterator;
import com.bigdata.relation.accesspath.IBuffer;
import com.bigdata.relation.accesspath.UnsynchronizedArrayBuffer;
import com.bigdata.relation.rule.IAccessPathExpander;
import com.bigdata.relation.rule.IProgram;
import com.bigdata.relation.rule.IRule;
import com.bigdata.relation.rule.IStep;
import com.bigdata.relation.rule.eval.AbstractJoinNexus;
import com.bigdata.relation.rule.eval.AbstractSolutionBuffer;
import com.bigdata.relation.rule.eval.ActionEnum;
import com.bigdata.relation.rule.eval.IJoinNexus;
import com.bigdata.relation.rule.eval.IRuleState;
import com.bigdata.relation.rule.eval.IRuleStatisticsFactory;
import com.bigdata.relation.rule.eval.ISolution;
import com.bigdata.relation.rule.eval.RuleStats;
import com.bigdata.striterator.ChunkedArrayIterator;
import com.bigdata.striterator.IChunkedIterator;
import com.bigdata.striterator.IChunkedOrderedIterator;
import com.bigdata.striterator.IKeyOrder;
/**
* {@link IProgram} execution support for the RDF DB.
*
* The rules have potential parallelism when performing closure. Each join has
* potential parallelism as well for subqueries. We could even define a PARALLEL
* iterator flag and have parallelism across index partitions for a
* read-historical iterator since the data service locators are immutable for
* historical reads.
*
* Rule-level parallelism (for fix point closure of a rule set) and join
* subquery-level parallelism could be distributed to available workers in a
* cluster. In a similar way, high-level queries could be distributed to workers
* in a cluster to evaluation. Such distribution would increase the practical
* parallelism beyond what a single machine could support as long as the total
* parallelism does not overload the cluster.
*
* There is a pragmatic limit on the #of concurrent threads for a single host.
* When those threads target a blocking queue, then thread contention becomes
* very high and throughput drops dramatically. We can reduce this problem by
* allocating a distinct {@link UnsynchronizedArrayBuffer} to each task. The
* task collects a 'chunk' in the {@link UnsynchronizedArrayBuffer}. When full,
* the buffer propagates onto a thread-safe buffer of chunks which flushes
* either on an {@link IMutableRelation} (mutation) or feeding an
* {@link IAsynchronousIterator} (high-level query). It is chunks themselves
* that accumulate in this thread-safe buffer, so each add() on that buffer may
* cause the thread to yield, but the return for yielding is an entire chunk in
* the buffer, not just a single element.
*
* There is one high-level buffer factory corresponding to each of the kinds of
* {@link ActionEnum}: {@link #newQueryBuffer()};
* {@link #newInsertBuffer(IMutableRelation)}; and
* {@link #newDeleteBuffer(IMutableRelation)}. In addition there is one for
* {@link UnsynchronizedArrayBuffer}s -- this is a buffer that is NOT
* thread-safe and that is designed to store a single chunk of elements, e.g.,
* in an array E[N]).
*
* @author Bryan Thompson
* @version $Id$
*/
public class RDFJoinNexus extends AbstractJoinNexus implements IJoinNexus {
protected final static transient Logger log = Logger.getLogger(RDFJoinNexus.class);
private final RDFJoinNexusFactory joinNexusFactory;
private final boolean justify;
/**
* when true the backchainer will be enabled for access path
* reads.
*/
private final boolean backchain;
private final IRuleStatisticsFactory ruleStatisticsFactory = new IRuleStatisticsFactory() {
public RuleStats newInstance(IStep step) {
return new RDFRuleStats(step);
}
public RuleStats newInstance(IRuleState ruleState) {
return new RDFRuleStats(null, getReadTimestamp(), ruleState);
}
// /**
// * Factory will resolve term identifiers in {@link IPredicate}s in the
// * tail of the {@link IRule} to {@link BigdataValue}s unless the
// * {@link IIndexManager} is an {@link IBigdataFederation}.
// *
// * @todo translation of term identifiers is disabled. someone is
// * interrupting the thread logging the {@link RuleStats}. until i
// * can figure out who that is, you will see term identifiers
// * rather than {@link BigdataValue}s.
// */
// public RuleStats newInstancex(IRuleState ruleState) {
//
// return new RDFRuleStats(
// (indexManager instanceof IBigdataFederation ? null
// : indexManager), //
// getReadTimestamp(), //
// ruleState//
// );
//
// }
};
/**
* Extends {@link RuleStats}s to translate the tail predicates back into
* RDF by resolving the term identifiers to {@link BigdataValue}s.
*/
private static class RDFRuleStats extends RuleStats {
private final IIndexManager indexManager;
private final long timestamp;
public RDFRuleStats(IStep step) {
super(step);
indexManager = null;
timestamp = 0L; // ignored.
}
/**
*
* @param indexManager
* When non-null, this is used to resolve
* the term identifiers in the {@link IPredicate}s in the
* tail of the rule to {@link BigdataValue}s.
*
* @param ruleState
*/
public RDFRuleStats(final IIndexManager indexManager,
final long timestamp, final IRuleState ruleState) {
super(ruleState);
this.indexManager = indexManager;
this.timestamp = timestamp;
}
@Override
@SuppressWarnings("unchecked")
protected String toString(final IPredicate pred) {
if (indexManager == null) {
return pred.toString().replace(", ", " ");
}
final SPORelation spoRelation = (SPORelation) indexManager
.getResourceLocator().locate(pred.getRelationName(0),
timestamp);
final AbstractTripleStore db = spoRelation.getContainer();
final Object s, p, o;
try {
{
final IVariableOrConstant t = pred.get(0);
if (t.isVar())
s = t.getName();
else
s = db.toString(t.get());
}
{
final IVariableOrConstant t = pred.get(1);
if (t.isVar())
p = t.getName();
else
p = db.toString(t.get());
}
{
final IVariableOrConstant t = pred.get(2);
if (t.isVar())
o = t.getName();
else
o = db.toString(t.get());
}
} catch (Throwable t) {
/*
* @todo It appears that someone is interrupting the thread in
* which the logging data is being generated. You can see this
* if you enable translation of term identifiers above in the
* factory that produces instances of this class.
*/
throw new RuntimeException("pred=" + pred + ", timestamp="
+ timestamp + ", indexManager=" + indexManager
+ ", db=" + db, t);
}
return "(" + s + " " + p + " " + o + ")";
}
}
/**
* @param joinNexusFactory
* The object used to create this instance and which can be used
* to create other instances as necessary for distributed rule
* execution.
* @param indexManager
* The object used to resolve indices, relations, etc.
*/
public RDFJoinNexus(final RDFJoinNexusFactory joinNexusFactory,
final IIndexManager indexManager) {
super(joinNexusFactory, indexManager);
this.joinNexusFactory = joinNexusFactory;
this.justify = joinNexusFactory.justify;
this.backchain = joinNexusFactory.backchain;
}
@Override
public IRuleStatisticsFactory getRuleStatisticsFactory() {
return ruleStatisticsFactory;
}
/**
* When {@link #backchain} is true and the tail predicate is
* reading on the {@link SPORelation}, then the {@link IAccessPath} is
* wrapped so that the iterator will visit the backchained inferences as
* well. On the other hand, if {@link IPredicate#getPartitionId()} is
* defined (not -1) then the returned access path will be for
* the specified shard using the data service local index manager (
* {@link #indexManager} MUST be the data service local index manager for
* this case) and expanders WILL NOT be applied (they require a view of the
* total relation, not just a shard).
*
* @see InferenceEngine
* @see BackchainAccessPath
*
* @todo consider encapsulating the {@link IRangeCountFactory} in the
* returned access path for non-exact range count requests. this will
* make it slightly harder to write the unit tests for the
* {@link IEvaluationPlanFactory}
*/
@Override
@SuppressWarnings("unchecked")
public IAccessPath getTailAccessPath(final IRelation relation,
final IPredicate predicate) {
// if (predicate.getPartitionId() != -1) {
//
// /*
// * Note: This handles a read against a local index partition. For
// * scale-out, the [indexManager] will be the data service's local
// * index manager.
// *
// * Note: Expanders ARE NOT applied in this code path. Expanders
// * require a total view of the relation, which is not available
// * during scale-out pipeline joins. Likewise, the [backchain]
// * property will be ignored since it is handled by an expander.
// *
// * @todo If getAccessPathForIndexPartition() is raised into the
// * IRelation interface, then we can get rid of the cast to the
// * SPORelation implementation.
// */
//
//// return ((SPORelation) relation).getAccessPathForIndexPartition(
//// indexManager, predicate);
// return relation.getAccessPath(indexManager, relation
// .getKeyOrder(predicate), predicate);
//
// }
//
// // Find the best access path for the predicate for that relation.
// IAccessPath accessPath = relation.getAccessPath(predicate);
////
//// if (predicate.getPartitionId() != -1) {
////
//// /*
//// * Note: The expander can not run against a shard since it assumes
//// * access to the full key range of the index. Expanders are
//// * convenient and work well for stand alone indices, but they should
//// * be replaced by rule rewrites for scale-out.
//// */
////
//// return accessPath;
////
//// }
//
final IKeyOrder keyOrder = relation.getKeyOrder(predicate);
IAccessPath accessPath = relation.getAccessPath(
indexManager/* localIndexManager */, keyOrder, predicate);
final IAccessPathExpander expander = predicate.getAccessPathExpander();
//
// if (expander != null) {
//
// // allow the predicate to wrap the access path
// accessPath = expander.getAccessPath(accessPath);
//
// }
// @todo raise into SPORelation#getAccessPath/3?
// @see https://sourceforge.net/apps/trac/bigdata/ticket/231
if(backchain && relation instanceof SPORelation) {
if (expander == null || expander.backchain()) {
final SPORelation spoRelation = (SPORelation)relation;
accessPath = new BackchainAccessPath(
spoRelation.getContainer(), accessPath,
joinNexusFactory.isOwlSameAsUsed ? Boolean.TRUE
: Boolean.FALSE);
}
}
// return that access path.
return accessPath;
}
// @SuppressWarnings("unchecked")
// public boolean bind(final IRule rule, final int index, final Object e,
// final IBindingSet bindings) {
//
// // propagate bindings from the visited object into the binding set.
// copyValues((IElement) e, rule.getTail(index), bindings);
//
// // verify constraints.
// return rule.isConsistent(bindings);
//
// }
//
// public boolean bind(final IPredicate pred, final IConstraint constraint,
// final Object e, final IBindingSet bindings) {
//
// // propagate bindings from the visited object into the binding set.
// copyValues((IElement) e, pred, bindings);
//
// if (constraint != null) {
//
// // verify constraint.
// return constraint.accept(bindings);
//
// }
//
// // no constraint.
// return true;
//
// }
// @SuppressWarnings("unchecked")
// private void copyValues(final IElement e, final IPredicate pred,
// final IBindingSet bindingSet) {
//
// for (int i = 0; i < pred.arity(); i++) {
//
// final IVariableOrConstant t = pred.get(i);
//
// if (t.isVar()) {
//
// final IVariable var = (IVariable) t;
//
// final Constant newval = new Constant(e.get(i));
//
// bindingSet.set(var, newval);
//
// }
//
// }
//
// }
public IConstant fakeBinding(IPredicate pred, Var var) {
return fakeTermId;
}
final private static transient IConstant fakeTermId =
new Constant(TermId.mockIV(VTE.URI));
/**
* FIXME unit tests for DISTINCT with a head and ELEMENT, with bindings and
* a head, with bindings but no head, and with a head but no bindings
* (error). See {@link #runQuery(IStep)}
*
* FIXME unit tests for SORT with and without DISTINCT and with the various
* combinations used in the unit tests for DISTINCT. Note that SORT, unlike
* DISTINCT, requires that all solutions are materialized before any
* solutions can be returned to the caller. A lot of optimization can be
* done for SORT implementations, including merge sort of large blocks (ala
* map/reduce), using compressed sort keys or word sort keys with 2nd stage
* disambiguation, etc.
*
* FIXME Add property for sort {ascending,descending,none} to {@link IRule}.
* The sort order can also be specified in terms of a sequence of variables.
* The choice of the variable order should be applied here.
*
* FIXME The properties that govern the Unicode collator for the generated
* sort keys should be configured by the {@link RDFJoinNexusFactory}. In
* particular, Unicode should be handled however it is handled for the
* {@link LexiconRelation}.
*/
public ISortKeyBuilder newBindingSetSortKeyBuilder(final IRule rule) {
final IKeyBuilder keyBuilder = KeyBuilder.newUnicodeInstance();
final int nvars = rule.getVariableCount();
final IVariable[] vars = new IVariable[nvars];
{
final Iterator itr = rule.getVariables();
int i = 0;
while (itr.hasNext()) {
vars[i++] = itr.next();
}
}
// @todo this class has RDF specific stuff in it.
return new BindingSetSortKeyBuilder(keyBuilder, vars);
}
@Override
protected ISortKeyBuilder newSortKeyBuilder(final IPredicate head) {
return new SPOSortKeyBuilder(head.arity());
}
/**
* Buffer writes on {@link IMutableRelation#insert(IChunkedIterator)} when it is
* {@link #flush() flushed}.
*
* @author Bryan Thompson
* @version $Id$
* @param
*/
public static class InsertSPOAndJustificationBuffer extends AbstractSolutionBuffer {
/**
* @param capacity
* @param relation
*/
public InsertSPOAndJustificationBuffer(final int capacity,
final IMutableRelation relation) {
super(capacity, relation);
}
@Override
protected long flush(final IChunkedOrderedIterator> itr) {
try {
/*
* The mutation count is the #of SPOs written (there is one
* justification written per solution generated, but the
* mutation count does not reflect duplicate justifications -
* only duplicate statements).
*
* Note: the optional filter for the ctor was already applied.
* If an element/solution was rejected, then it is not in the
* buffer and we will never see it during flush().
*/
long mutationCount = 0;
while (itr.hasNext()) {
final ISolution[] chunk = itr.nextChunk();
mutationCount += writeChunk(chunk);
}
return mutationCount;
} finally {
itr.close();
}
}
private long writeChunk(final ISolution[] chunk) {
final int n = chunk.length;
if(log.isDebugEnabled())
log.debug("chunkSize="+n);
final long begin = System.currentTimeMillis();
final SPO[] a = new SPO[ n ];
final Justification[] b = new Justification[ n ];
for (int i = 0; i < chunk.length; i++) {
if(log.isDebugEnabled()) {
log.debug("chunk["+i+"] = "+chunk[i]);
}
final ISolution solution = (ISolution) chunk[i];
a[i] = solution.get();
b[i] = new Justification(solution);
}
final SPORelation r = (SPORelation) (IMutableRelation) getRelation();
/*
* Use a thread pool to write out the statement and the
* justifications concurrently. This drammatically reduces the
* latency when also writing justifications.
*/
final List> tasks = new ArrayList>(2);
/*
* Note: we reject using the filter before stmts or justifications
* make it into the buffer so we do not need to apply the filter
* again here.
*/
tasks.add(new Callable(){
public Long call() {
return r.insert(a,a.length,null/*filter*/);
}
});
tasks.add(new Callable(){
public Long call() {
return r
.addJustifications(new ChunkedArrayIterator(
b.length, b, null/* keyOrder */));
}
});
final List> futures;
/*
* @todo The timings for the tasks that we run here are not being
* reported up to this point.
*/
final long mutationCount;
try {
futures = r.getExecutorService().invokeAll(tasks);
mutationCount = futures.get(0).get();
futures.get(1).get();
} catch (InterruptedException ex) {
throw new RuntimeException(ex);
} catch (ExecutionException ex) {
throw new RuntimeException(ex);
}
final long elapsed = System.currentTimeMillis() - begin;
if (log.isInfoEnabled())
log.info("Wrote " + mutationCount
+ " statements and justifications in "
+ elapsed + "ms");
return mutationCount;
}
}
/**
* Overridden to handle justifications when using truth maintenance.
*
* {@inheritDoc}
*/
@Override
@SuppressWarnings("unchecked")
public IBuffer newInsertBuffer(final IMutableRelation relation) {
if (getAction() != ActionEnum.Insert)
throw new IllegalStateException();
if (log.isDebugEnabled()) {
log.debug("relation=" + relation);
}
if(justify) {
/*
* Buffer knows how to write the computed elements on the statement
* indices and the computed binding sets on the justifications
* indices.
*/
return new InsertSPOAndJustificationBuffer(chunkOfChunksCapacity,
relation);
}
/*
* Buffer resolves the computed elements and writes them on the
* statement indices.
*/
return new AbstractSolutionBuffer.InsertSolutionBuffer(
chunkOfChunksCapacity, relation);
}
}