com.thinkaurelius.titan.diskstorage.cassandra.thrift.CassandraThriftStoreManager Maven / Gradle / Ivy
The newest version!
package com.thinkaurelius.titan.diskstorage.cassandra.thrift;
import static com.thinkaurelius.titan.diskstorage.cassandra.CassandraTransaction.getTx;
import static org.apache.cassandra.db.Table.SYSTEM_KS;
import java.nio.ByteBuffer;
import java.util.ArrayList;
import java.util.HashMap;
import java.util.LinkedList;
import java.util.List;
import java.util.Map;
import org.apache.cassandra.dht.ByteOrderedPartitioner;
import org.apache.cassandra.dht.IPartitioner;
import org.apache.cassandra.dht.Token;
import org.apache.cassandra.thrift.Cassandra;
import org.apache.cassandra.thrift.CfDef;
import org.apache.cassandra.thrift.Column;
import org.apache.cassandra.thrift.ColumnOrSuperColumn;
import org.apache.cassandra.thrift.ConsistencyLevel;
import org.apache.cassandra.thrift.Deletion;
import org.apache.cassandra.thrift.InvalidRequestException;
import org.apache.cassandra.thrift.KsDef;
import org.apache.cassandra.thrift.NotFoundException;
import org.apache.cassandra.thrift.SchemaDisagreementException;
import org.apache.cassandra.thrift.SlicePredicate;
import org.apache.cassandra.utils.FBUtilities;
import org.apache.commons.configuration.Configuration;
import org.apache.commons.pool.impl.GenericKeyedObjectPool;
import org.apache.thrift.TException;
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;
import com.google.common.base.Preconditions;
import com.google.common.collect.ImmutableMap;
import com.thinkaurelius.titan.diskstorage.Backend;
import com.thinkaurelius.titan.diskstorage.PermanentStorageException;
import com.thinkaurelius.titan.diskstorage.StaticBuffer;
import com.thinkaurelius.titan.diskstorage.StorageException;
import com.thinkaurelius.titan.diskstorage.TemporaryStorageException;
import com.thinkaurelius.titan.diskstorage.cassandra.AbstractCassandraStoreManager;
import com.thinkaurelius.titan.diskstorage.cassandra.thrift.thriftpool.CTConnection;
import com.thinkaurelius.titan.diskstorage.cassandra.thrift.thriftpool.CTConnectionFactory;
import com.thinkaurelius.titan.diskstorage.cassandra.thrift.thriftpool.CTConnectionPool;
import com.thinkaurelius.titan.diskstorage.keycolumnvalue.Entry;
import com.thinkaurelius.titan.diskstorage.keycolumnvalue.KCVMutation;
import com.thinkaurelius.titan.diskstorage.keycolumnvalue.StoreTransaction;
import com.thinkaurelius.titan.diskstorage.util.ByteBufferUtil;
import com.thinkaurelius.titan.diskstorage.util.Hex;
import com.thinkaurelius.titan.graphdb.configuration.GraphDatabaseConfiguration;
/**
* This class creates {@see CassandraThriftKeyColumnValueStore}s and
* handles Cassandra-backed allocation of vertex IDs for Titan (when so
* configured).
*
* @author Dan LaRocque
*/
public class CassandraThriftStoreManager extends AbstractCassandraStoreManager {
private static final Logger log = LoggerFactory.getLogger(CassandraThriftStoreManager.class);
private final Map openStores;
private final CTConnectionPool pool;
public CassandraThriftStoreManager(Configuration config) throws StorageException {
super(config);
int thriftTimeoutMS = config.getInt(
GraphDatabaseConfiguration.CONNECTION_TIMEOUT_KEY,
GraphDatabaseConfiguration.CONNECTION_TIMEOUT_DEFAULT);
int maxTotalConnections = config.getInt(
GraphDatabaseConfiguration.CONNECTION_POOL_SIZE_KEY,
GraphDatabaseConfiguration.CONNECTION_POOL_SIZE_DEFAULT);
CTConnectionFactory factory = new CTConnectionFactory(hostnames, port, username, password, thriftTimeoutMS, thriftFrameSize);
CTConnectionPool p = new CTConnectionPool(factory);
p.setTestOnBorrow(true);
p.setTestOnReturn(true);
p.setTestWhileIdle(false);
p.setWhenExhaustedAction(GenericKeyedObjectPool.WHEN_EXHAUSTED_BLOCK);
p.setMaxActive(-1); // "A negative value indicates no limit"
p.setMaxTotal(maxTotalConnections); // maxTotal limits active + idle
p.setMinIdle(0); // prevent evictor from eagerly creating unused connections
p.setMinEvictableIdleTimeMillis(60 * 1000L);
p.setTimeBetweenEvictionRunsMillis(30 * 1000L);
this.pool = p;
this.openStores = new HashMap();
}
@Override
public Deployment getDeployment() {
return Deployment.REMOTE;
}
@Override
@SuppressWarnings("unchecked")
public IPartitioner> getCassandraPartitioner() throws StorageException {
CTConnection conn = null;
try {
conn = pool.borrowObject(SYSTEM_KS);
return FBUtilities.newPartitioner(conn.getClient().describe_partitioner());
} catch (Exception e) {
throw new TemporaryStorageException(e);
} finally {
pool.returnObjectUnsafe(SYSTEM_KS, conn);
}
}
@Override
public String toString() {
return "thriftCassandra" + super.toString();
}
@Override
public void close() throws StorageException {
openStores.clear();
closePool();
}
@Override
public void mutateMany(Map> mutations, StoreTransaction txh) throws StorageException {
Preconditions.checkNotNull(mutations);
final Timestamp timestamp = getTimestamp(txh);
ConsistencyLevel consistency = getTx(txh).getWriteConsistencyLevel().getThriftConsistency();
// Generate Thrift-compatible batch_mutate() datastructure
// key -> cf -> cassmutation
int size = 0;
for (Map mutation : mutations.values()) size += mutation.size();
Map>> batch =
new HashMap>>(size);
for (Map.Entry> keyMutation : mutations.entrySet()) {
String columnFamily = keyMutation.getKey();
for (Map.Entry mutEntry : keyMutation.getValue().entrySet()) {
StaticBuffer key = mutEntry.getKey();
ByteBuffer keyBB = key.asByteBuffer();
// Get or create the single Cassandra Mutation object responsible for this key
Map> cfmutation = batch.get(keyBB);
if (cfmutation == null) {
cfmutation = new HashMap>(3); // TODO where did the magic number 3 come from?
batch.put(keyBB, cfmutation);
}
KCVMutation mutation = mutEntry.getValue();
List thriftMutation =
new ArrayList(mutations.size());
if (mutation.hasDeletions()) {
for (StaticBuffer buf : mutation.getDeletions()) {
Deletion d = new Deletion();
SlicePredicate sp = new SlicePredicate();
sp.addToColumn_names(buf.asByteBuffer());
d.setPredicate(sp);
d.setTimestamp(timestamp.deletionTime);
org.apache.cassandra.thrift.Mutation m = new org.apache.cassandra.thrift.Mutation();
m.setDeletion(d);
thriftMutation.add(m);
}
}
if (mutation.hasAdditions()) {
for (Entry ent : mutation.getAdditions()) {
ColumnOrSuperColumn cosc = new ColumnOrSuperColumn();
Column column = new Column(ent.getColumn().asByteBuffer());
column.setValue(ent.getValue().asByteBuffer());
column.setTimestamp(timestamp.additionTime);
cosc.setColumn(column);
org.apache.cassandra.thrift.Mutation m = new org.apache.cassandra.thrift.Mutation();
m.setColumn_or_supercolumn(cosc);
thriftMutation.add(m);
}
}
cfmutation.put(columnFamily, thriftMutation);
}
}
CTConnection conn = null;
try {
conn = pool.borrowObject(keySpaceName);
Cassandra.Client client = conn.getClient();
client.batch_mutate(batch, consistency);
} catch (Exception ex) {
throw CassandraThriftKeyColumnValueStore.convertException(ex);
} finally {
pool.returnObjectUnsafe(keySpaceName, conn);
}
}
@Override // TODO: *BIG FAT WARNING* 'synchronized is always *bad*, change openStores to use ConcurrentLinkedHashMap
public synchronized CassandraThriftKeyColumnValueStore openDatabase(final String name) throws StorageException {
if (openStores.containsKey(name))
return openStores.get(name);
ensureColumnFamilyExists(keySpaceName, name);
CassandraThriftKeyColumnValueStore store = new CassandraThriftKeyColumnValueStore(keySpaceName, name, this, pool);
openStores.put(name, store);
return store;
}
/**
* Connect to Cassandra via Thrift on the specified host and port and attempt to truncate the named keyspace.
*
* This is a utility method intended mainly for testing. It is
* equivalent to issuing 'truncate ' for each of the column families in keyspace using
* the cassandra-cli tool.
*
* Using truncate is better for a number of reasons, most significantly because it doesn't
* involve any schema modifications which can take time to propagate across the cluster such
* leaves nodes in the inconsistent state and could result in read/write failures.
* Any schema modifications are discouraged until there is no traffic to Keyspace or ColumnFamilies.
*
* @throws StorageException if any checked Thrift or UnknownHostException is thrown in the body of this method
*/
public void clearStorage() throws StorageException {
openStores.clear();
final String lp = "ClearStorage: "; // "log prefix"
/*
* log4j is capable of automatically writing the name of a method that
* generated a log message, but the docs warn that "generating caller
* location information is extremely slow and should be avoided unless
* execution speed is not an issue."
*/
CTConnection conn = null;
try {
conn = pool.borrowObject(SYSTEM_KS);
Cassandra.Client client = conn.getClient();
KsDef ksDef;
try {
client.set_keyspace(keySpaceName);
ksDef = client.describe_keyspace(keySpaceName);
} catch (NotFoundException e) {
log.debug(lp + "Keyspace {} does not exist, not attempting to truncate.", keySpaceName);
return;
} catch (InvalidRequestException e) {
log.debug(lp + "InvalidRequestException when attempting to describe keyspace {}, not attempting to truncate.", keySpaceName);
return;
}
if (null == ksDef) {
log.debug(lp + "Received null KsDef for keyspace {}; not truncating its CFs", keySpaceName);
return;
}
List cfDefs = ksDef.getCf_defs();
if (null == cfDefs) {
log.debug(lp + "Received empty CfDef list for keyspace {}; not truncating CFs", keySpaceName);
return;
}
for (CfDef cfDef : ksDef.getCf_defs()) {
client.truncate(cfDef.name);
log.info(lp + "Truncated CF {} in keyspace {}", cfDef.name, keySpaceName);
}
/*
* Clearing the CTConnectionPool is unnecessary. This method
* removes no keyspaces. All open Cassandra connections will
* remain valid.
*/
} catch (Exception e) {
throw new TemporaryStorageException(e);
} finally {
if (conn != null && conn.getClient() != null) {
try {
conn.getClient().set_keyspace(SYSTEM_KS);
} catch (InvalidRequestException e) {
log.warn("Failed to reset keyspace", e);
} catch (TException e) {
log.warn("Failed to reset keyspace", e);
}
}
pool.returnObjectUnsafe(SYSTEM_KS, conn);
}
}
private KsDef ensureKeyspaceExists(String keyspaceName)
throws NotFoundException, InvalidRequestException, TException,
SchemaDisagreementException, StorageException {
CTConnection connection = null;
try {
connection = pool.borrowObject(SYSTEM_KS);
Cassandra.Client client = connection.getClient();
try {
// Side effect: throws Exception if keyspaceName doesn't exist
client.set_keyspace(keyspaceName); // Don't remove
client.set_keyspace(SYSTEM_KS);
log.debug("Found existing keyspace {}", keyspaceName);
} catch (InvalidRequestException e) {
// Keyspace didn't exist; create it
log.debug("Creating keyspace {}...", keyspaceName);
KsDef ksdef = new KsDef().setName(keyspaceName)
.setCf_defs(new LinkedList()) // cannot be null but can be empty
.setStrategy_class("org.apache.cassandra.locator.SimpleStrategy")
.setStrategy_options(ImmutableMap.of("replication_factor", String.valueOf(replicationFactor)));
client.set_keyspace(SYSTEM_KS);
try {
client.system_add_keyspace(ksdef);
log.debug("Created keyspace {}", keyspaceName);
} catch (InvalidRequestException ire) {
log.error("system_add_keyspace failed for keyspace=" + keyspaceName, ire);
throw ire;
}
}
return client.describe_keyspace(keyspaceName);
} catch (Exception e) {
throw new TemporaryStorageException(e);
} finally {
pool.returnObjectUnsafe(SYSTEM_KS, connection);
}
}
private void ensureColumnFamilyExists(String ksName, String cfName) throws StorageException {
ensureColumnFamilyExists(ksName, cfName, "org.apache.cassandra.db.marshal.BytesType");
}
private void ensureColumnFamilyExists(String ksName, String cfName, String comparator) throws StorageException {
CTConnection conn = null;
try {
KsDef keyspaceDef = ensureKeyspaceExists(ksName);
conn = pool.borrowObject(ksName);
Cassandra.Client client = conn.getClient();
log.debug("Looking up metadata on keyspace {}...", ksName);
boolean foundColumnFamily = false;
for (CfDef cfDef : keyspaceDef.getCf_defs()) {
String curCfName = cfDef.getName();
if (curCfName.equals(cfName))
foundColumnFamily = true;
}
if (!foundColumnFamily) {
createColumnFamily(client, ksName, cfName, comparator);
} else {
log.debug("Keyspace {} and ColumnFamily {} were found.", ksName, cfName);
}
} catch (SchemaDisagreementException e) {
throw new TemporaryStorageException(e);
} catch (Exception e) {
throw new PermanentStorageException(e);
} finally {
pool.returnObjectUnsafe(ksName, conn);
}
}
private void createColumnFamily(Cassandra.Client client,
String ksName,
String cfName,
String comparator) throws StorageException {
CfDef createColumnFamily = new CfDef();
createColumnFamily.setName(cfName);
createColumnFamily.setKeyspace(ksName);
createColumnFamily.setComparator_type(comparator);
ImmutableMap.Builder compressionOptions = new ImmutableMap.Builder();
if (compressionEnabled) {
compressionOptions.put("sstable_compression", compressionClass)
.put("chunk_length_kb", Integer.toString(compressionChunkSizeKB));
}
createColumnFamily.setCompression_options(compressionOptions.build());
// Hard-coded caching settings
if (cfName.startsWith(Backend.EDGESTORE_NAME)) {
createColumnFamily.setCaching("keys_only");
} else if (cfName.startsWith(Backend.VERTEXINDEX_STORE_NAME)) {
createColumnFamily.setCaching("rows_only");
}
log.debug("Adding column family {} to keyspace {}...", cfName, ksName);
try {
client.system_add_column_family(createColumnFamily);
} catch (SchemaDisagreementException e) {
throw new TemporaryStorageException("Error in setting up column family", e);
} catch (Exception e) {
throw new PermanentStorageException(e);
}
log.debug("Added column family {} to keyspace {}.", cfName, ksName);
}
@Override
public Map getCompressionOptions(String cf) throws StorageException {
CTConnection conn = null;
Map result = null;
try {
conn = pool.borrowObject(keySpaceName);
Cassandra.Client client = conn.getClient();
KsDef ksDef = client.describe_keyspace(keySpaceName);
for (CfDef cfDef : ksDef.getCf_defs()) {
if (null != cfDef && cfDef.getName().equals(cf)) {
result = cfDef.getCompression_options();
break;
}
}
return result;
} catch (InvalidRequestException e) {
log.debug("Keyspace {} does not exist", keySpaceName);
return null;
} catch (Exception e) {
throw new TemporaryStorageException(e);
} finally {
pool.returnObjectUnsafe(keySpaceName, conn);
}
}
private void closePool() {
/*
* pool.close() does not affect borrowed connections.
*
* Connections currently borrowed by some thread which are
* talking to the old host will eventually be destroyed by
* CTConnectionFactory#validateObject() returning false when
* those connections are returned to the pool.
*/
try {
pool.close();
log.info("Closed Thrift connection pooler.");
} catch (Exception e) {
log.warn("Failed to close connection pooler. "
+ "We might be leaking Cassandra connections.", e);
// There's still hope: CTConnectionFactory#validateObject()
// will be called on borrow() and might tear down the
// connections that close() failed to tear down
}
}
}
© 2015 - 2025 Weber Informatics LLC | Privacy Policy