org.apache.cassandra.db.SerializationHeader Maven / Gradle / Ivy
Go to download
Show more of this group Show more artifacts with this name
Show all versions of cassandra-all Show documentation
Show all versions of cassandra-all Show documentation
The Apache Cassandra Project develops a highly scalable second-generation distributed database, bringing together Dynamo's fully distributed design and Bigtable's ColumnFamily-based data model.
/*
* Licensed to the Apache Software Foundation (ASF) under one
* or more contributor license agreements. See the NOTICE file
* distributed with this work for additional information
* regarding copyright ownership. The ASF licenses this file
* to you under the Apache License, Version 2.0 (the
* "License"); you may not use this file except in compliance
* with the License. You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
package org.apache.cassandra.db;
import java.io.IOException;
import java.nio.ByteBuffer;
import java.util.*;
import com.google.common.collect.ImmutableList;
import org.apache.cassandra.db.filter.ColumnFilter;
import org.apache.cassandra.db.marshal.AbstractType;
import org.apache.cassandra.db.marshal.UTF8Type;
import org.apache.cassandra.db.rows.EncodingStats;
import org.apache.cassandra.exceptions.UnknownColumnException;
import org.apache.cassandra.io.sstable.format.SSTableReader;
import org.apache.cassandra.io.sstable.format.Version;
import org.apache.cassandra.io.sstable.metadata.IMetadataComponentSerializer;
import org.apache.cassandra.io.sstable.metadata.MetadataComponent;
import org.apache.cassandra.io.sstable.metadata.MetadataType;
import org.apache.cassandra.io.util.DataInputPlus;
import org.apache.cassandra.io.util.DataOutputPlus;
import org.apache.cassandra.schema.ColumnMetadata;
import org.apache.cassandra.schema.TableMetadata;
import org.apache.cassandra.serializers.AbstractTypeSerializer;
import org.apache.cassandra.utils.ByteBufferUtil;
public class SerializationHeader
{
public static final Serializer serializer = new Serializer();
private final boolean isForSSTable;
private final AbstractType keyType;
private final List> clusteringTypes;
private final RegularAndStaticColumns columns;
private final EncodingStats stats;
private final Map> typeMap;
private SerializationHeader(boolean isForSSTable,
AbstractType keyType,
List> clusteringTypes,
RegularAndStaticColumns columns,
EncodingStats stats,
Map> typeMap)
{
this.isForSSTable = isForSSTable;
this.keyType = keyType;
this.clusteringTypes = clusteringTypes;
this.columns = columns;
this.stats = stats;
this.typeMap = typeMap;
}
public static SerializationHeader makeWithoutStats(TableMetadata metadata)
{
return new SerializationHeader(true, metadata, metadata.regularAndStaticColumns(), EncodingStats.NO_STATS);
}
public static SerializationHeader make(TableMetadata metadata, Collection sstables)
{
// The serialization header has to be computed before the start of compaction (since it's used to write)
// the result. This means that when compacting multiple sources, we won't have perfectly accurate stats
// (for EncodingStats) since compaction may delete, purge and generally merge rows in unknown ways. This is
// kind of ok because those stats are only used for optimizing the underlying storage format and so we
// just have to strive for as good as possible. Currently, we stick to a relatively naive merge of existing
// global stats because it's simple and probably good enough in most situation but we could probably
// improve our marging of inaccuracy through the use of more fine-grained stats in the future.
// Note however that to avoid seeing our accuracy degrade through successive compactions, we don't base
// our stats merging on the compacted files headers, which as we just said can be somewhat inaccurate,
// but rather on their stats stored in StatsMetadata that are fully accurate.
EncodingStats.Collector stats = new EncodingStats.Collector();
RegularAndStaticColumns.Builder columns = RegularAndStaticColumns.builder();
// We need to order the SSTables by descending generation to be sure that we use latest column metadata.
for (SSTableReader sstable : orderByDescendingGeneration(sstables))
{
stats.updateTimestamp(sstable.getMinTimestamp());
stats.updateLocalDeletionTime(sstable.getMinLocalDeletionTime());
stats.updateTTL(sstable.getMinTTL());
columns.addAll(sstable.header.columns());
}
return new SerializationHeader(true, metadata, columns.build(), stats.get());
}
private static Collection orderByDescendingGeneration(Collection sstables)
{
if (sstables.size() < 2)
return sstables;
List readers = new ArrayList<>(sstables);
readers.sort(SSTableReader.idReverseComparator);
return readers;
}
public SerializationHeader(boolean isForSSTable,
TableMetadata metadata,
RegularAndStaticColumns columns,
EncodingStats stats)
{
this(isForSSTable,
metadata.partitionKeyType,
metadata.comparator.subtypes(),
columns,
stats,
null);
}
public RegularAndStaticColumns columns()
{
return columns;
}
public boolean hasStatic()
{
return !columns.statics.isEmpty();
}
public boolean isForSSTable()
{
return isForSSTable;
}
public EncodingStats stats()
{
return stats;
}
public AbstractType keyType()
{
return keyType;
}
public List> clusteringTypes()
{
return clusteringTypes;
}
public Columns columns(boolean isStatic)
{
return isStatic ? columns.statics : columns.regulars;
}
public AbstractType getType(ColumnMetadata column)
{
return typeMap == null ? column.type : typeMap.get(column.name.bytes);
}
public void writeTimestamp(long timestamp, DataOutputPlus out) throws IOException
{
out.writeUnsignedVInt(timestamp - stats.minTimestamp);
}
public void writeLocalDeletionTime(long localDeletionTime, DataOutputPlus out) throws IOException
{
out.writeUnsignedVInt32((int) (localDeletionTime - stats.minLocalDeletionTime));
}
public void writeTTL(int ttl, DataOutputPlus out) throws IOException
{
out.writeUnsignedVInt32(ttl - stats.minTTL);
}
public void writeDeletionTime(DeletionTime dt, DataOutputPlus out) throws IOException
{
writeTimestamp(dt.markedForDeleteAt(), out);
writeLocalDeletionTime(dt.localDeletionTime(), out);
}
public long readTimestamp(DataInputPlus in) throws IOException
{
return in.readUnsignedVInt() + stats.minTimestamp;
}
public long readLocalDeletionTime(DataInputPlus in) throws IOException
{
return in.readUnsignedVInt32() + stats.minLocalDeletionTime;
}
public int readTTL(DataInputPlus in) throws IOException
{
return in.readUnsignedVInt32() + stats.minTTL;
}
public DeletionTime readDeletionTime(DataInputPlus in) throws IOException
{
long markedAt = readTimestamp(in);
long localDeletionTime = readLocalDeletionTime(in);
return DeletionTime.build(markedAt, localDeletionTime);
}
public long timestampSerializedSize(long timestamp)
{
return TypeSizes.sizeofUnsignedVInt(timestamp - stats.minTimestamp);
}
public long localDeletionTimeSerializedSize(long localDeletionTime)
{
return TypeSizes.sizeofUnsignedVInt(localDeletionTime - stats.minLocalDeletionTime);
}
public long ttlSerializedSize(int ttl)
{
return TypeSizes.sizeofUnsignedVInt(ttl - stats.minTTL);
}
public long deletionTimeSerializedSize(DeletionTime dt)
{
return timestampSerializedSize(dt.markedForDeleteAt())
+ localDeletionTimeSerializedSize(dt.localDeletionTime());
}
public void skipTimestamp(DataInputPlus in) throws IOException
{
in.readUnsignedVInt();
}
public void skipLocalDeletionTime(DataInputPlus in) throws IOException
{
in.readUnsignedVInt();
}
public void skipTTL(DataInputPlus in) throws IOException
{
in.readUnsignedVInt();
}
public void skipDeletionTime(DataInputPlus in) throws IOException
{
skipTimestamp(in);
skipLocalDeletionTime(in);
}
public Component toComponent()
{
Map> staticColumns = new LinkedHashMap<>();
Map> regularColumns = new LinkedHashMap<>();
for (ColumnMetadata column : columns.statics)
staticColumns.put(column.name.bytes, column.type);
for (ColumnMetadata column : columns.regulars)
regularColumns.put(column.name.bytes, column.type);
return new Component(keyType, clusteringTypes, staticColumns, regularColumns, stats);
}
@Override
public String toString()
{
return String.format("SerializationHeader[key=%s, cks=%s, columns=%s, stats=%s, typeMap=%s]", keyType, clusteringTypes, columns, stats, typeMap);
}
/**
* We need the TableMetadata to properly deserialize a SerializationHeader but it's clunky to pass that to
* a SSTable component, so we use this temporary object to delay the actual need for the metadata.
*/
public static class Component extends MetadataComponent
{
private final AbstractType keyType;
private final List> clusteringTypes;
private final Map> staticColumns;
private final Map> regularColumns;
private final EncodingStats stats;
private Component(AbstractType keyType,
List> clusteringTypes,
Map> staticColumns,
Map> regularColumns,
EncodingStats stats)
{
this.keyType = keyType;
this.clusteringTypes = clusteringTypes;
this.staticColumns = staticColumns;
this.regularColumns = regularColumns;
this.stats = stats;
}
public MetadataType getType()
{
return MetadataType.HEADER;
}
public SerializationHeader toHeader(TableMetadata metadata) throws UnknownColumnException
{
Map> typeMap = new HashMap<>(staticColumns.size() + regularColumns.size());
RegularAndStaticColumns.Builder builder = RegularAndStaticColumns.builder();
for (Map> map : ImmutableList.of(staticColumns, regularColumns))
{
boolean isStatic = map == staticColumns;
for (Map.Entry> e : map.entrySet())
{
ByteBuffer name = e.getKey();
AbstractType other = typeMap.put(name, e.getValue());
if (other != null && !other.equals(e.getValue()))
throw new IllegalStateException("Column " + name + " occurs as both regular and static with types " + other + "and " + e.getValue());
ColumnMetadata column = metadata.getColumn(name);
if (column == null || column.isStatic() != isStatic)
{
// TODO: this imply we don't read data for a column we don't yet know about, which imply this is theoretically
// racy with column addition. Currently, it is up to the user to not write data before the schema has propagated
// and this is far from being the only place that has such problem in practice. This doesn't mean we shouldn't
// improve this.
// If we don't find the definition, it could be we have data for a dropped column, and we shouldn't
// fail deserialization because of that. So we grab a "fake" ColumnDefinition that ensure proper
// deserialization. The column will be ignore later on anyway.
column = metadata.getDroppedColumn(name, isStatic);
if (column == null)
throw new UnknownColumnException("Unknown column " + UTF8Type.instance.getString(name) + " during deserialization");
}
builder.add(column);
}
}
return new SerializationHeader(true, keyType, clusteringTypes, builder.build(), stats, typeMap);
}
@Override
public boolean equals(Object o)
{
if(!(o instanceof Component))
return false;
Component that = (Component)o;
return Objects.equals(this.keyType, that.keyType)
&& Objects.equals(this.clusteringTypes, that.clusteringTypes)
&& Objects.equals(this.staticColumns, that.staticColumns)
&& Objects.equals(this.regularColumns, that.regularColumns)
&& Objects.equals(this.stats, that.stats);
}
@Override
public int hashCode()
{
return Objects.hash(keyType, clusteringTypes, staticColumns, regularColumns, stats);
}
@Override
public String toString()
{
return String.format("SerializationHeader.Component[key=%s, cks=%s, statics=%s, regulars=%s, stats=%s]",
keyType, clusteringTypes, staticColumns, regularColumns, stats);
}
public AbstractType getKeyType()
{
return keyType;
}
public List> getClusteringTypes()
{
return clusteringTypes;
}
public Map> getStaticColumns()
{
return staticColumns;
}
public Map> getRegularColumns()
{
return regularColumns;
}
public EncodingStats getEncodingStats()
{
return stats;
}
}
public static class Serializer implements IMetadataComponentSerializer
{
private final AbstractTypeSerializer typeSerializer = new AbstractTypeSerializer();
public void serializeForMessaging(SerializationHeader header, ColumnFilter selection, DataOutputPlus out, boolean hasStatic) throws IOException
{
EncodingStats.serializer.serialize(header.stats, out);
if (selection == null)
{
if (hasStatic)
Columns.serializer.serialize(header.columns.statics, out);
Columns.serializer.serialize(header.columns.regulars, out);
}
else
{
if (hasStatic)
Columns.serializer.serializeSubset(header.columns.statics, selection.fetchedColumns().statics, out);
Columns.serializer.serializeSubset(header.columns.regulars, selection.fetchedColumns().regulars, out);
}
}
public SerializationHeader deserializeForMessaging(DataInputPlus in, TableMetadata metadata, ColumnFilter selection, boolean hasStatic) throws IOException
{
EncodingStats stats = EncodingStats.serializer.deserialize(in);
AbstractType keyType = metadata.partitionKeyType;
List> clusteringTypes = metadata.comparator.subtypes();
Columns statics, regulars;
if (selection == null)
{
statics = hasStatic ? Columns.serializer.deserialize(in, metadata) : Columns.NONE;
regulars = Columns.serializer.deserialize(in, metadata);
}
else
{
statics = hasStatic ? Columns.serializer.deserializeSubset(selection.fetchedColumns().statics, in) : Columns.NONE;
regulars = Columns.serializer.deserializeSubset(selection.fetchedColumns().regulars, in);
}
return new SerializationHeader(false, keyType, clusteringTypes, new RegularAndStaticColumns(statics, regulars), stats, null);
}
public long serializedSizeForMessaging(SerializationHeader header, ColumnFilter selection, boolean hasStatic)
{
long size = EncodingStats.serializer.serializedSize(header.stats);
if (selection == null)
{
if (hasStatic)
size += Columns.serializer.serializedSize(header.columns.statics);
size += Columns.serializer.serializedSize(header.columns.regulars);
}
else
{
if (hasStatic)
size += Columns.serializer.serializedSubsetSize(header.columns.statics, selection.fetchedColumns().statics);
size += Columns.serializer.serializedSubsetSize(header.columns.regulars, selection.fetchedColumns().regulars);
}
return size;
}
// For SSTables
public void serialize(Version version, Component header, DataOutputPlus out) throws IOException
{
EncodingStats.serializer.serialize(header.stats, out);
typeSerializer.serialize(header.keyType, out);
typeSerializer.serializeList(header.clusteringTypes, out);
writeColumnsWithTypes(header.staticColumns, out);
writeColumnsWithTypes(header.regularColumns, out);
}
// For SSTables
public Component deserialize(Version version, DataInputPlus in) throws IOException
{
EncodingStats stats = EncodingStats.serializer.deserialize(in);
AbstractType keyType = typeSerializer.deserialize(in);
List> clusteringTypes = typeSerializer.deserializeList(in);
Map> staticColumns = readColumnsWithType(in);
Map> regularColumns = readColumnsWithType(in);
return new Component(keyType, clusteringTypes, staticColumns, regularColumns, stats);
}
// For SSTables
public int serializedSize(Version version, Component header)
{
int size = EncodingStats.serializer.serializedSize(header.stats);
size += typeSerializer.serializedSize(header.keyType);
size += typeSerializer.serializedListSize(header.clusteringTypes);
size += sizeofColumnsWithTypes(header.staticColumns);
size += sizeofColumnsWithTypes(header.regularColumns);
return size;
}
private void writeColumnsWithTypes(Map> columns, DataOutputPlus out) throws IOException
{
out.writeUnsignedVInt32(columns.size());
for (Map.Entry> entry : columns.entrySet())
{
ByteBufferUtil.writeWithVIntLength(entry.getKey(), out);
typeSerializer.serialize(entry.getValue(), out);
}
}
private long sizeofColumnsWithTypes(Map> columns)
{
long size = TypeSizes.sizeofUnsignedVInt(columns.size());
for (Map.Entry> entry : columns.entrySet())
{
size += ByteBufferUtil.serializedSizeWithVIntLength(entry.getKey());
size += typeSerializer.serializedSize(entry.getValue());
}
return size;
}
private Map> readColumnsWithType(DataInputPlus in) throws IOException
{
int length = in.readUnsignedVInt32();
Map> typeMap = new LinkedHashMap<>(length);
for (int i = 0; i < length; i++)
{
ByteBuffer name = ByteBufferUtil.readWithVIntLength(in);
typeMap.put(name, typeSerializer.deserialize(in));
}
return typeMap;
}
}
}