com.bigdata.relation.AbstractResource 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 Jul 10, 2008
*/
package com.bigdata.relation;
import java.util.Iterator;
import java.util.Map;
import java.util.Properties;
import java.util.UUID;
import java.util.concurrent.BlockingQueue;
import java.util.concurrent.ExecutorService;
import java.util.concurrent.SynchronousQueue;
import org.apache.log4j.Logger;
import com.bigdata.bop.BOp;
import com.bigdata.bop.engine.QueryEngine;
import com.bigdata.bop.fed.QueryEngineFactory;
import com.bigdata.config.Configuration;
import com.bigdata.config.IValidator;
import com.bigdata.config.IntegerValidator;
import com.bigdata.config.LongValidator;
import com.bigdata.journal.IIndexManager;
import com.bigdata.journal.IJournal;
import com.bigdata.journal.IResourceLock;
import com.bigdata.journal.IResourceLockService;
import com.bigdata.journal.NoSuchIndexException;
import com.bigdata.journal.TimestampUtility;
import com.bigdata.rdf.rules.FastClosure;
import com.bigdata.rdf.rules.FullClosure;
import com.bigdata.rdf.rules.RuleFastClosure5;
import com.bigdata.rdf.rules.RuleFastClosure6;
import com.bigdata.rdf.sparql.ast.cache.CacheConnectionFactory;
import com.bigdata.rdf.sparql.ast.cache.ICacheConnection;
import com.bigdata.rdf.store.AbstractTripleStore;
import com.bigdata.relation.accesspath.AccessPath;
import com.bigdata.relation.accesspath.BlockingBuffer;
import com.bigdata.relation.accesspath.IAccessPath;
import com.bigdata.relation.accesspath.IBuffer;
import com.bigdata.relation.locator.DefaultResourceLocator;
import com.bigdata.relation.locator.ILocatableResource;
import com.bigdata.relation.rule.eval.IJoinNexus;
import com.bigdata.relation.rule.eval.IJoinNexusFactory;
import com.bigdata.relation.rule.eval.ISolution;
import com.bigdata.relation.rule.eval.ProgramTask;
import com.bigdata.service.IBigdataFederation;
import com.bigdata.sparse.GlobalRowStoreUtil;
/**
* Base class for locatable resources.
*
* @author Bryan Thompson
* @version $Id$
*
* @param
*/
abstract public class AbstractResource implements IMutableResource {
protected final transient static Logger log = Logger.getLogger(AbstractResource.class);
protected final IIndexManager indexManager;
final private String namespace;
final private String containerNamespace;
final private long timestamp;
final private Long commitTime;
final private Properties properties;
final private int chunkOfChunksCapacity;
final private int chunkCapacity;
final private long chunkTimeout;
private final int fullyBufferedReadThreshold;
private final boolean forceSerialExecution;
private final int maxParallelSubqueries;
// private final boolean nestedSubquery;
/**
* The capacity of the buffers accumulating chunks from concurrent producers.
*
* @see Options#CHUNK_OF_CHUNKS_CAPACITY
*
* @deprecated by {@link BOp} annotations.
*/
final public int getChunkOfChunksCapacity() {
return chunkOfChunksCapacity;
}
/**
* The target chunk size.
*
* @see Options#CHUNK_CAPACITY
*
* @deprecated by {@link BOp} annotations.
*/
final public int getChunkCapacity() {
return chunkCapacity;
}
/**
* The timeout in milliseconds that the {@link BlockingBuffer} will wait for
* another chunk to combine with the current chunk before returning the
* current chunk. This may be ZERO (0) to disable the chunk combiner.
*
* @see Options#CHUNK_TIMEOUT
*
* @deprecated by {@link BOp} annotations.
*/
public final long getChunkTimeout() {
return chunkTimeout;
}
/**
* If the estimated range count for an
* {@link IAccessPath#iterator(long,long, int)} is LTE this threshold then
* do a fully buffered (synchronous) read. Otherwise we will do an
* asynchronous read.
*
* @see Options#FULLY_BUFFERED_READ_THRESHOLD
*
* @deprecated by {@link BOp} annotations.
*/
public int getFullyBufferedReadThreshold() {
return fullyBufferedReadThreshold;
}
/**
* When true, rule sets will be forced to execute
* sequentially even when they are not flagged as a sequential program.
*
* @see Options#FORCE_SERIAL_EXECUTION
*
* @deprecated by {@link BOp} annotations.
*/
public boolean isForceSerialExecution() {
return forceSerialExecution;
}
/**
* The maximum #of subqueries for the first join dimension that will be
* issued in parallel. Use ZERO(0) to avoid submitting tasks to the
* {@link ExecutorService} entirely and ONE (1) to submit a single task at a
* time to the {@link ExecutorService}.
*
* @see Options#MAX_PARALLEL_SUBQUERIES
*
* @deprecated by {@link BOp} annotations.
*/
public int getMaxParallelSubqueries() {
return maxParallelSubqueries;
}
/**
* Options for locatable resources.
*
* @todo most of these options effect asynchronous iterators, access path
* behavior, and join behavior. these are general features for bigdata
* resources, but some of the code to support this stuff is still
* local to the RDF module. That can be fixed using an abstract base
* class for {@link IJoinNexus} and {@link IJoinNexusFactory}.
*
* @todo some of these defaults need to be re-examined. see notes in the
* javadoc below.
*
* @author Bryan Thompson
* @version $Id$
*/
public static interface Options {
/**
*
* Set the maximum #of chunks from concurrent producers that can be
* buffered before an {@link IBuffer} containing chunks of
* {@link ISolution}s would block (default
* {@link #DEFAULT_CHUNK_OF_CHUNKS_CAPACITY}). This is used to
* provision a {@link BlockingQueue} for {@link BlockingBuffer}. A
* value of ZERO(0) indicates that a {@link SynchronousQueue} should be
* used instead. The best value may be more than the #of concurrent
* producers if the producers are generating small chunks, e.g., because
* there are few solutions for a join subquery.
*
* @deprecated by {@link BOp} annotations.
*/
String CHUNK_OF_CHUNKS_CAPACITY = BlockingBuffer.class.getName()
+ ".chunkOfChunksCapacity";
/**
* Default for {@link #CHUNK_OF_CHUNKS_CAPACITY}
* @deprecated by {@link BOp} annotations.
*/
String DEFAULT_CHUNK_OF_CHUNKS_CAPACITY = "10"; // was 1000
/**
*
* Sets the capacity of the {@link IBuffer}s used to accumulate a chunk
* when evaluating rules, etc (default {@value #CHUNK_CAPACITY}). Note
* that many processes use a {@link BlockingBuffer} to accumulate
* "chunks of chunks".
*
*
* @see #CHUNK_OF_CHUNKS_CAPACITY
* @deprecated by {@link BOp} annotations.
*/
String CHUNK_CAPACITY = IBuffer.class.getName() + ".chunkCapacity";
/**
* Default for {@link #CHUNK_CAPACITY}
*
* Note: This used to be 20k, but chunks of chunks works better than
* just a large chunk.
*
* @deprecated by {@link BOp} annotations.
*/
String DEFAULT_CHUNK_CAPACITY = "100"; // was 100
/**
* The timeout in milliseconds that the {@link BlockingBuffer} will wait
* for another chunk to combine with the current chunk before returning
* the current chunk (default {@link #DEFAULT_CHUNK_TIMEOUT}). This may
* be ZERO (0) to disable the chunk combiner.
*
* @deprecated by {@link BOp} annotations.
*/
String CHUNK_TIMEOUT = BlockingBuffer.class.getName() + ".chunkTimeout";
/**
* The default for {@link #CHUNK_TIMEOUT}.
*
* @todo this is probably much larger than we want. Try 10ms.
* @deprecated by {@link BOp} annotations.
*/
String DEFAULT_CHUNK_TIMEOUT = "10"; // was 1000
/**
* If the estimated rangeCount for an
* {@link AccessPath#iterator()} is LTE this threshold then use
* a fully buffered (synchronous) iterator. Otherwise use an
* asynchronous iterator whose capacity is governed by
* {@link #CHUNK_OF_CHUNKS_CAPACITY}.
*
* @deprecated by {@link BOp} annotations.
*/
String FULLY_BUFFERED_READ_THRESHOLD = AccessPath.class
.getName()
+ ".fullyBufferedReadThreadshold";
/**
* Default for {@link #FULLY_BUFFERED_READ_THRESHOLD}
*
* @todo figure out how good this value is.
* @deprecated by {@link BOp} annotations.
*/
String DEFAULT_FULLY_BUFFERED_READ_THRESHOLD = "200";//""+20*Bytes.kilobyte32;
/**
* When true ({@value #DEFAULT_FORCE_SERIAL_EXECUTION}),
* rule sets will be forced to execute sequentially even when they are
* not flagged as a sequential program.
*
* @todo The following discussion applies to the
* {@link AbstractTripleStore}. and should be relocated.
*
* The {@link #CLOSURE_CLASS} option defaults to
* {@link FastClosure}, which has very little possible parallelism
* (it is mostly a sequential program by nature). For that reason,
* {@link #FORCE_SERIAL_EXECUTION} defaults to false
* since the overhead of parallel execution is more likely to
* lower the observed performance with such limited possible
* parallelism. However, when using {@link FullClosure} the
* benefits of parallelism MAY justify its overhead.
*
* The following data are for LUBM datasets.
*
*
* U1 Fast Serial : closure = 2250ms; 2765, 2499, 2530
* U1 Fast Parallel : closure = 2579ms; 2514, 2594
* U1 Full Serial : closure = 10437ms.
* U1 Full Parallel : closure = 10843ms.
*
* U10 Fast Serial : closure = 41203ms, 39171ms (38594, 35360 when running in caller's thread rather than on the executorService).
* U10 Fast Parallel : closure = 30722ms.
* U10 Full Serial : closure = 108110ms.
* U10 Full Parallel : closure = 248550ms.
*
*
* Note that the only rules in the fast closure program that have
* potential parallelism are {@link RuleFastClosure5} and
* {@link RuleFastClosure6} and these rules are not being
* triggered by these datasets, so there is in fact NO potential
* parallelism (in the data) for these datasets.
*
* It is possible that a machine with more cores would perform
* better under the "full" closure program with parallel rule
* execution (these data were collected on a laptop with 2 cores)
* since performance tends to be CPU bound for small data sets.
* However, the benefit of the "fast" closure program is so large
* that there is little reason to consider parallel rule execution
* for the "full" closure program.
*
* @todo collect new timings for this option. The LUBM performance has
* basically doubled since these data were collected. Look further
* into ways in which overhead might be reduced for rule
* parallelism and also for when rule parallelism is not enabled.
*
* @todo rename as parallel_rule_execution.
* @deprecated by {@link BOp} annotations.
*/
String FORCE_SERIAL_EXECUTION = ProgramTask.class.getName()
+ ".forceSerialExecution";
/**
* @deprecated by {@link BOp} annotations.
*/
String DEFAULT_FORCE_SERIAL_EXECUTION = "true";
/**
* The maximum #of subqueries for the first join dimension that will be
* issued in parallel. Use ZERO(0) to avoid submitting tasks to the
* {@link ExecutorService} entirely and ONE (1) to submit a single task
* at a time to the {@link ExecutorService}.
*
* @deprecated by {@link BOp} annotations.
*/
String MAX_PARALLEL_SUBQUERIES = ProgramTask.class.getName()
+ ".maxParallelSubqueries";
/**
* @deprecated by {@link BOp} annotations.
*/
String DEFAULT_MAX_PARALLEL_SUBQUERIES = "5";
}
protected AbstractResource(final IIndexManager indexManager,
final String namespace, final Long timestamp,
final Properties properties) {
this(null/* container */, indexManager, namespace, timestamp,
properties);
}
/**
* Alternative version used when a resource exists within some container.
* The additional container argument provides access to the container
* before the container has been written to the global row store.
*/
protected AbstractResource(final ILocatableResource container,
final IIndexManager indexManager, final String namespace,
final Long timestamp, final Properties properties) {
// Note: [container] MAY be null.
if (indexManager == null)
throw new IllegalArgumentException();
if (namespace == null)
throw new IllegalArgumentException();
if (timestamp == null)
throw new IllegalArgumentException();
if (properties == null)
throw new IllegalArgumentException();
// Note: Non-null if this resource exists in some container
this.container = container;
// Note: Bound before we lookup property values!
this.indexManager = indexManager;
// Note: Bound before we lookup property values!
this.namespace = namespace;
this.containerNamespace = properties.getProperty(RelationSchema.CONTAINER);
this.timestamp = timestamp;
this.properties = properties;
properties.setProperty(RelationSchema.NAMESPACE, namespace);
properties.setProperty(RelationSchema.CLASS, getClass().getName());
if (log.isInfoEnabled()) {
log.info("namespace=" + namespace + ", timestamp=" + timestamp
+ ", container=" + containerNamespace + ", indexManager="
+ indexManager);
}
/*
* Resolve the commit time from which this view was materialized (if
* known and otherwise null).
*/
commitTime = (Long)properties.get(RelationSchema.COMMIT_TIME);
forceSerialExecution = Boolean.parseBoolean(getProperty(
Options.FORCE_SERIAL_EXECUTION,
Options.DEFAULT_FORCE_SERIAL_EXECUTION));
maxParallelSubqueries = getProperty(Options.MAX_PARALLEL_SUBQUERIES,
Options.DEFAULT_MAX_PARALLEL_SUBQUERIES,
IntegerValidator.GTE_ZERO);
chunkOfChunksCapacity = getProperty(Options.CHUNK_OF_CHUNKS_CAPACITY,
Options.DEFAULT_CHUNK_OF_CHUNKS_CAPACITY,
IntegerValidator.GT_ZERO);
chunkCapacity = getProperty(Options.CHUNK_CAPACITY,
Options.DEFAULT_CHUNK_CAPACITY, IntegerValidator.GT_ZERO);
chunkTimeout = getProperty(Options.CHUNK_TIMEOUT,
Options.DEFAULT_CHUNK_TIMEOUT, LongValidator.GTE_ZERO);
fullyBufferedReadThreshold = getProperty(
Options.FULLY_BUFFERED_READ_THRESHOLD,
Options.DEFAULT_FULLY_BUFFERED_READ_THRESHOLD,
IntegerValidator.GT_ZERO);
}
public final String getNamespace() {
return namespace;
}
public final String getContainerNamespace() {
return containerNamespace;
}
/**
* Return the container.
*
* @return The container -or- null if there is no container.
*/
public ILocatableResource getContainer() {
if (container == null) {
synchronized (this) {
if (container == null) {
if (getContainerNamespace() != null) {
if (log.isInfoEnabled()) {
log.info("resolving container: "
+ getContainerNamespace());
}
container = getIndexManager()
.getResourceLocator()
.locate(getContainerNamespace(), getTimestamp());
}
}
}
}
return container;
}
private volatile ILocatableResource container;
public final long getTimestamp() {
return timestamp;
}
/**
* The commit time from which a read-only view was materialized (if known)
* and otherwise null.
*
* @see https://sourceforge.net/apps/trac/bigdata/ticket/266
*/
protected Long getCommitTime() {
return commitTime;
}
/**
* Wrap and return the properties specified to the ctor. Wrapping the
* {@link Properties} object prevents inadvertent side-effects.
*/
public final Properties getProperties() {
return new Properties(properties);
}
/**
* Return the {@link Properties} object without wrapping it. This method can
* be used in those cases where you need to access non-String property
* values. The caller is responsible for avoiding mutation to the returned
* Properties object.
*
* Note: This explicitly does NOT wrap the properties. Doing so makes it
* impossible to access the default properties using Hashtable#get(), which
* in turn means that we can not access non-String objects which have been
* materialized from the GRS in the {@link Properties}. This does introduce
* some potential for side-effects between read-only instances of the same
* resource view which share the same properties object.
*/
protected final Properties getBareProperties() {
return properties;
}
/**
* Return the object used to locate indices, relations, and relation
* containers and to execute operations on those resources.
*
* Note: For scale-out, this is always the federation's index
* manager NOT the data service local index manager. This is an artifact of
* how we resolve the metadata for the relation against the global row
* store.
*
* @return The {@link IIndexManager}.
*
* @todo If we support the notion of a "relation shard" then this could
* become the shard's data service local index manager in that
* instance but, regardless, we would need a means to resolve the
* metadata for the relation against the federation's index manager
*/
public IIndexManager getIndexManager() {
return indexManager;
}
final public ExecutorService getExecutorService() {
return indexManager.getExecutorService();
}
/**
* The class name, timestamp and namespace for the relation view.
*/
public String toString() {
return getClass().getSimpleName() + "{timestamp=" + timestamp
+ ", namespace=" + namespace + ", container=" + containerNamespace
+ ", indexManager=" + indexManager + "}";
}
/**
* The default implementation only logs the event.
*/
public AbstractResource init() {
if (log.isInfoEnabled())
log.info(toString());
return this;
}
/**
*
* @todo Lock service supporting shared locks, leases and lease renewal,
* escalation of shared locks to exclusive locks, deadlock detection,
* and possibly a resource hierarchy. Leases should be Callable
* objects that are submitted by the client to its executor service so
* that they will renew automatically until cancelled (and will cancel
* automatically if not renewed).
*
* There is existing code that could be adapted for this purpose. It
* might have to be adapted to support lock escalation (shared to
* exclusive), a resource hierarchy, and a delay queue to cancel
* leases that are not renewed. It would have to be wrapped up as a
* low-latency service and made available via the
* {@link IBigdataFederation}. It also needs to use a weak reference
* cache for the collection of resource queues so that they are GC'd
* rather than growing as new resources are locked and never
* shrinking.
*
* If we require pre-declaration of locks, then we do not need the
* dependency graph since deadlocks can only arise with 2PL.
*
* Since the service is remote it should use {@link UUID}s to
* identify the lock owner(s).
*
* The lock service would be used to bracket operations such as
* relation {@link #create()} and {@link #destroy()} and would be used
* to prevent those operations while a lease is held by concurrent
* processes with a shared lock.
*
* Add ctor flag to create iff not found?
*
* There needs to be a lock protocol for subclasses so that they can
* ensure that they are the only task running create (across the
* federation) and so that they can release the lock when they are
* done. The lock can be per the notes above, but the protocol with
* the subclass will require some coordinating methods.
*
* Full transactions are another way to solve this problem.
*/
@Override
public void create() {
if (log.isInfoEnabled())
log.info(toString());
/*
* Convert the Properties to a Map.
*/
final Map map = GlobalRowStoreUtil.convert(properties);
// Write the map on the row store.
final Map afterMap = indexManager.getGlobalRowStore()
.write(RelationSchema.INSTANCE, map);
if(log.isDebugEnabled()) {
log.debug("Properties after write: " + afterMap);
}
/*
* Add this instance to the locator cache.
*
* Note: Normally, the instances are created by the locator cache
* itself. In general the only time the application creates an instance
* directly is when it is going to attempt to create the relation. This
* takes advantage of that pattern to notify the locator that it should
* cache this instance.
*/
((DefaultResourceLocator) getIndexManager().getResourceLocator())
.putInstance(this);
}
@Override
public void destroy() {
if (log.isInfoEnabled())
log.info(toString());
/*
* Destroy all indices spanned by the namespace for this relation.
*/
{
final Iterator itr = indexManager.indexNameScan(namespace
+ ".", timestamp);
while (itr.hasNext()) {
final String name = itr.next();
try {
indexManager.dropIndex(name);
} catch (NoSuchIndexException ex) {
// If the index does not exist, keep on going.
log.warn("Ignored: " + ex);
}
}
}
/*
* Destroy any caches associated with this relation.
*/
{
if (indexManager instanceof IJournal
|| indexManager instanceof IBigdataFederation) {
/*
* Note: No cache associated with a TemporaryStore (since no
* QueryEngine is associated with a TemporaryStore).
*
* Note: If the cache is remote, then we need to create the
* QueryEngine instance on demand so we can access the cache.
* Hence getQueryController() and not getExistingQueryEngine().
*
* TODO This could wind up using the wrong index manager if the
* destroy() is executed by an AbstractTask (neither the Journal
* nor the JiniFederation).
*/
final QueryEngine queryEngine = QueryEngineFactory.getInstance()
.getQueryController(indexManager);
/*
* Connect to the cache provider.
*
* Note: This will create the cache if it does not exist. At all
* other places in the code we use getExistingSparqlCache() to
* access the cache IFF it exists. Here is where we create it.
*/
final ICacheConnection cacheConn = CacheConnectionFactory
.getCacheConnection(queryEngine);
if (cacheConn != null) {
cacheConn.destroyCaches(namespace, timestamp);
}
}
}
// Delete the entry for this relation from the row store.
indexManager.getGlobalRowStore().delete(RelationSchema.INSTANCE,
namespace);
// Clear the entry from the resource locator cache.
indexManager.getResourceLocator().discard(this, true/* destroyed */);
}
/**
* Acquires an exclusive lock for the {@link #getNamespace()}.
*
* @return the lock.
*
* @throws RuntimeException
* if anything goes wrong.
*
* @see IResourceLockService
*/
protected IResourceLock acquireExclusiveLock() {
return indexManager.getResourceLockService().acquireLock(getNamespace());
}
/**
* Release the lock.
*
* @param resourceLock
* The lock.
*/
protected void unlock(final IResourceLock resourceLock) {
resourceLock.unlock();
}
/**
* Resolve the property value using the {@link IIndexManager}, the
* namespace of the resource, and the {@link Properties} instance to be
* tested as hidden parameters.
*
* @param globalName
* The global property name.
* @param defaultValue
* The default.
*
* @return The resolved property value.
*
* @see Configuration
*/
public String getProperty(final String localName,
final String defaultValue) {
return Configuration.getProperty(indexManager, properties, namespace,
localName, defaultValue);
}
/**
* Resolves, parses, and validates the property value.
*
* @param name
* The property name.
* @param defaultValue
* The default value.
* @return
*/
public T getProperty(final String name, final String defaultValue,
final IValidator validator) {
return Configuration.getProperty(indexManager, properties, namespace,
name, defaultValue, validator);
}
// /**
// * Sets the property on the underlying properties object but DOES NOT set
// * the property on the global row store (GRS). This method may be used when
// * a resource is newly created in order to cache objects which are persisted
// * on the GRS.
// *
// * @param name
// * The property name.
// * @param value
// * The property value.
// */
// protected void setProperty(final String name, final Object value) {
//
// properties.put(name, value);
//
// }
public boolean isReadOnly() {
return TimestampUtility.isReadOnly(getTimestamp());
}
final protected void assertWritable() {
if(isReadOnly()) {
throw new IllegalStateException("READ_ONLY");
}
}
}