org.apache.cassandra.db.partitions

Class PartitionUpdate

java.lang.Object

org.apache.cassandra.db.partitions.AbstractBTreePartition

org.apache.cassandra.db.partitions.PartitionUpdate

All Implemented Interfaces:

java.lang.Iterable<Row>, Partition

public class PartitionUpdate
extends AbstractBTreePartition

Stores updates made on a partition.

A PartitionUpdate object requires that all writes/additions are performed before we try to read the updates (attempts to write to the PartitionUpdate after a read method has been called will result in an exception being thrown). In other words, a Partition is mutable while it's written but becomes immutable as soon as it is read.

A typical usage is to create a new update (new PartitionUpdate(metadata, key, columns, capacity)) and then add rows and range tombstones through the add() methods (the partition level deletion time can also be set with addPartitionDeletion()). However, there is also a few static helper constructor methods for special cases (emptyUpdate(), fullPartitionDelete and singleRowUpdate).

Nested Class Summary

Nested Classes

Modifier and Type	Class and Description
static class	PartitionUpdate.CounterMark A counter mark is basically a pointer to a counter update inside this partition update.
static class	PartitionUpdate.PartitionUpdateSerializer

Nested classes/interfaces inherited from class org.apache.cassandra.db.partitions.AbstractBTreePartition

AbstractBTreePartition.Holder, AbstractBTreePartition.SliceableIterator, AbstractBTreePartition.SlicesIterator

Field Summary

Fields

Modifier and Type	Field and Description
protected static org.slf4j.Logger	logger
static PartitionUpdate.PartitionUpdateSerializer	serializer

Fields inherited from class org.apache.cassandra.db.partitions.AbstractBTreePartition

EMPTY, metadata, partitionKey

Constructor Summary

Constructors

Constructor and Description
PartitionUpdate(CFMetaData metadata, java.nio.ByteBuffer key, PartitionColumns columns, int initialRowCapacity)
PartitionUpdate(CFMetaData metadata, DecoratedKey key, PartitionColumns columns, int initialRowCapacity)

Method Summary

Modifier and Type	Method and Description
void	add(RangeTombstone range)
void	add(Row row) Adds a row to this update.
void	addPartitionDeletion(DeletionTime deletionTime)
void	allowNewUpdates() If a partition update has been read (and is thus unmodifiable), a call to this method makes the update modifiable again.
protected boolean	canHaveShadowedData()
java.util.List<PartitionUpdate.CounterMark>	collectCounterMarks() For an update on a counter table, returns a list containing a CounterMark for every counter contained in the update.
PartitionColumns	columns()
int	dataSize() The size of the data contained in this update.
static PartitionUpdate	emptyUpdate(CFMetaData metadata, DecoratedKey key) Creates a empty immutable partition update.
static PartitionUpdate	fromBytes(java.nio.ByteBuffer bytes, int version, DecoratedKey key) Deserialize a partition update from a provided byte buffer.
static PartitionUpdate	fromIterator(RowIterator iterator)
static PartitionUpdate	fromIterator(UnfilteredRowIterator iterator) Turns the given iterator into an update.
static PartitionUpdate	fullPartitionDelete(CFMetaData metadata, java.nio.ByteBuffer key, long timestamp, int nowInSec) Creates a partition update that entirely deletes a given partition.
static PartitionUpdate	fullPartitionDelete(CFMetaData metadata, DecoratedKey key, long timestamp, int nowInSec) Creates an immutable partition update that entirely deletes a given partition.
protected AbstractBTreePartition.Holder	holder()
java.util.Iterator<Row>	iterator() Returns an iterator that iterates over the rows of this update in clustering order.
long	maxTimestamp() The maximum timestamp used in this update.
static PartitionUpdate	merge(java.util.List<PartitionUpdate> updates) Merges the provided updates, yielding a new update that incorporates all those updates.
int	operationCount() The number of "operations" contained in the update.
static PartitionUpdate	singleRowUpdate(CFMetaData metadata, java.nio.ByteBuffer key, Row row) Creates an immutable partition update that contains a single row update.
static PartitionUpdate	singleRowUpdate(CFMetaData metadata, DecoratedKey key, Row row) Creates an immutable partition update that contains a single row update.
SliceableUnfilteredRowIterator	sliceableUnfilteredIterator(ColumnFilter columns, boolean reversed)
EncodingStats	stats()
static java.nio.ByteBuffer	toBytes(PartitionUpdate update, int version) Serialize a partition update as a byte buffer.
java.lang.String	toString()
void	updateAllTimestamp(long newTimestamp) Modify this update to set every timestamp for live data to newTimestamp and every deletion timestamp to newTimestamp - 1.
void	validate() Validates the data contained in this update.

Methods inherited from class org.apache.cassandra.db.partitions.AbstractBTreePartition

build, build, deletionInfo, getRow, hasRows, isEmpty, lastRow, metadata, partitionKey, partitionLevelDeletion, rowCount, searchIterator, sliceableUnfilteredIterator, staticRow, unfilteredIterator, unfilteredIterator, unfilteredIterator

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, wait, wait, wait

Methods inherited from interface java.lang.Iterable

forEach, spliterator

Field Detail

logger

protected static final org.slf4j.Logger logger

serializer

public static final PartitionUpdate.PartitionUpdateSerializer serializer

Constructor Detail

PartitionUpdate

public PartitionUpdate(CFMetaData metadata,
                       DecoratedKey key,
                       PartitionColumns columns,
                       int initialRowCapacity)

PartitionUpdate

public PartitionUpdate(CFMetaData metadata,
                       java.nio.ByteBuffer key,
                       PartitionColumns columns,
                       int initialRowCapacity)

Method Detail

emptyUpdate

public static PartitionUpdate emptyUpdate(CFMetaData metadata,
                                          DecoratedKey key)

Creates a empty immutable partition update.

Parameters:

metadata - the metadata for the created update.

key - the partition key for the created update.

Returns:

the newly created empty (and immutable) update.

fullPartitionDelete

public static PartitionUpdate fullPartitionDelete(CFMetaData metadata,
                                                  DecoratedKey key,
                                                  long timestamp,
                                                  int nowInSec)

Creates an immutable partition update that entirely deletes a given partition.

Parameters:

metadata - the metadata for the created update.

key - the partition key for the partition that the created update should delete.

timestamp - the timestamp for the deletion.

nowInSec - the current time in seconds to use as local deletion time for the partition deletion.

Returns:

the newly created partition deletion update.

singleRowUpdate

public static PartitionUpdate singleRowUpdate(CFMetaData metadata,
                                              DecoratedKey key,
                                              Row row)

Creates an immutable partition update that contains a single row update.

Parameters:

metadata - the metadata for the created update.

key - the partition key for the partition to update.

row - the row for the update.

Returns:

the newly created partition update containing only row.

singleRowUpdate

public static PartitionUpdate singleRowUpdate(CFMetaData metadata,
                                              java.nio.ByteBuffer key,
                                              Row row)

Creates an immutable partition update that contains a single row update.

Parameters:

metadata - the metadata for the created update.

key - the partition key for the partition to update.

row - the row for the update.

Returns:

the newly created partition update containing only row.

fromIterator

public static PartitionUpdate fromIterator(UnfilteredRowIterator iterator)

Turns the given iterator into an update. Warning: this method does not close the provided iterator, it is up to the caller to close it.

fromIterator

public static PartitionUpdate fromIterator(RowIterator iterator)

canHaveShadowedData

protected boolean canHaveShadowedData()

Specified by:

canHaveShadowedData in class AbstractBTreePartition

fromBytes

public static PartitionUpdate fromBytes(java.nio.ByteBuffer bytes,
                                        int version,
                                        DecoratedKey key)

Deserialize a partition update from a provided byte buffer.

Parameters:

bytes - the byte buffer that contains the serialized update.

version - the version with which the update is serialized.

key - the partition key for the update. This is only used if version &lt 3.0 and can be null otherwise.

Returns:

the deserialized update or null if bytes == null.

toBytes

public static java.nio.ByteBuffer toBytes(PartitionUpdate update,
                                          int version)

Serialize a partition update as a byte buffer.

Parameters:

update - the partition update to serialize.

version - the version to serialize the update into.

Returns:

a newly allocated byte buffer containing the serialized update.

fullPartitionDelete

public static PartitionUpdate fullPartitionDelete(CFMetaData metadata,
                                                  java.nio.ByteBuffer key,
                                                  long timestamp,
                                                  int nowInSec)

Creates a partition update that entirely deletes a given partition.

Parameters:

metadata - the metadata for the created update.

key - the partition key for the partition that the created update should delete.

timestamp - the timestamp for the deletion.

nowInSec - the current time in seconds to use as local deletion time for the partition deletion.

Returns:

the newly created partition deletion update.

merge

public static PartitionUpdate merge(java.util.List<PartitionUpdate> updates)

Merges the provided updates, yielding a new update that incorporates all those updates.

Parameters:

updates - the collection of updates to merge. This shouldn't be empty.

Returns:

a partition update that include (merge) all the updates from updates.

updateAllTimestamp

public void updateAllTimestamp(long newTimestamp)

Modify this update to set every timestamp for live data to newTimestamp and every deletion timestamp to newTimestamp - 1. There is no reason to use that expect on the Paxos code path, where we need ensure that anything inserted use the ballot timestamp (to respect the order of update decided by the Paxos algorithm). We use newTimestamp - 1 for deletions because tombstones always win on timestamp equality and we don't want to delete our own insertions (typically, when we overwrite a collection, we first set a complex deletion to delete the previous collection before adding new elements. If we were to set that complex deletion to the same timestamp that the new elements, it would delete those elements). And since tombstones always wins on timestamp equality, using -1 guarantees our deletion will still delete anything from a previous update.

operationCount

public int operationCount()

The number of "operations" contained in the update.

This is used by Memtable to approximate how much work this update does. In practice, this count how many rows are updated and how many ranges are deleted by the partition update.

Returns:

the number of "operations" performed by the update.

dataSize

public int dataSize()

The size of the data contained in this update.

Returns:

the size of the data contained in this update.

columns

public PartitionColumns columns()

Specified by:

columns in interface Partition

Overrides:

columns in class AbstractBTreePartition

holder

protected AbstractBTreePartition.Holder holder()

Specified by:

holder in class AbstractBTreePartition

stats

public EncodingStats stats()

Specified by:

stats in interface Partition

Overrides:

stats in class AbstractBTreePartition

allowNewUpdates

public void allowNewUpdates()

If a partition update has been read (and is thus unmodifiable), a call to this method makes the update modifiable again.

Please note that calling this method won't result in optimal behavior in the sense that even if very little is added to the update after this call, the whole update will be sorted again on read. This should thus be used sparingly (and if it turns that we end up using this often, we should consider optimizing the behavior).

iterator

public java.util.Iterator<Row> iterator()

Returns an iterator that iterates over the rows of this update in clustering order.

Note that this might trigger a sorting of the update, and as such the update will not be modifiable anymore after this call.

Specified by:

iterator in interface java.lang.Iterable<Row>

Overrides:

iterator in class AbstractBTreePartition

Returns:

an iterator over the rows of this update.

sliceableUnfilteredIterator

public SliceableUnfilteredRowIterator sliceableUnfilteredIterator(ColumnFilter columns,
                                                                  boolean reversed)

Overrides:

sliceableUnfilteredIterator in class AbstractBTreePartition

validate

public void validate()

Validates the data contained in this update.

Throws:

MarshalException - if some of the data contained in this update is corrupted.

maxTimestamp

public long maxTimestamp()

The maximum timestamp used in this update.

Returns:

the maximum timestamp used in this update.

collectCounterMarks

public java.util.List<PartitionUpdate.CounterMark> collectCounterMarks()

For an update on a counter table, returns a list containing a CounterMark for every counter contained in the update.

Returns:

a list with counter marks for every counter in this update.

addPartitionDeletion

public void addPartitionDeletion(DeletionTime deletionTime)

add

public void add(RangeTombstone range)

add

public void add(Row row)

Adds a row to this update. There is no particular assumption made on the order of row added to a partition update. It is further allowed to add the same row (more precisely, multiple row objects for the same clustering). Note however that the columns contained in the added row must be a subset of the columns used when creating this update.

Parameters:

row - the row to add.

toString

public java.lang.String toString()

Overrides:

toString in class AbstractBTreePartition