org.apache.cassandra.db.compaction

Class CompactionController

java.lang.Object

org.apache.cassandra.db.AbstractCompactionController

org.apache.cassandra.db.compaction.CompactionController

All Implemented Interfaces:

java.lang.AutoCloseable

Direct Known Subclasses:

SSTableSplitter.SplitController, TimeWindowCompactionController

public class CompactionController
extends AbstractCompactionController

Manage compaction options.

Field Summary

Fields inherited from class org.apache.cassandra.db.AbstractCompactionController

cfs, gcBefore, tombstoneOption

Constructor Summary

Constructors

Modifier	Constructor and Description
protected	CompactionController(ColumnFamilyStore cfs, int maxValue)
	CompactionController(ColumnFamilyStore cfs, java.util.Set<SSTableReader> compacting, int gcBefore)
	CompactionController(ColumnFamilyStore cfs, java.util.Set<SSTableReader> compacting, int gcBefore, com.google.common.util.concurrent.RateLimiter limiter, CompactionParams.TombstoneOption tombstoneOption)

Method Summary

Modifier and Type	Method and Description
void	close()
boolean	compactingRepaired()
java.util.Set<SSTableReader>	getFullyExpiredSSTables()
static java.util.Set<SSTableReader>	getFullyExpiredSSTables(ColumnFamilyStore cfStore, java.lang.Iterable<SSTableReader> compacting, java.lang.Iterable<SSTableReader> overlapping, int gcBefore)
static java.util.Set<SSTableReader>	getFullyExpiredSSTables(ColumnFamilyStore cfStore, java.lang.Iterable<SSTableReader> compacting, java.lang.Iterable<SSTableReader> overlapping, int gcBefore, boolean ignoreOverlaps) Finds expired sstables works something like this; 1.
java.util.function.LongPredicate	getPurgeEvaluator(DecoratedKey key)
protected boolean	ignoreOverlaps() Is overlapped sstables ignored Control whether or not we are taking into account overlapping sstables when looking for fully expired sstables.
void	maybeRefreshOverlaps()
java.lang.Iterable<UnfilteredRowIterator>	shadowSources(DecoratedKey key, boolean tombstoneOnly)

Methods inherited from class org.apache.cassandra.db.AbstractCompactionController

getColumnFamily, getKeyspace

Methods inherited from class java.lang.Object

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

Constructor Detail

CompactionController

protected CompactionController(ColumnFamilyStore cfs,
                               int maxValue)

CompactionController

public CompactionController(ColumnFamilyStore cfs,
                            java.util.Set<SSTableReader> compacting,
                            int gcBefore)

CompactionController

public CompactionController(ColumnFamilyStore cfs,
                            java.util.Set<SSTableReader> compacting,
                            int gcBefore,
                            com.google.common.util.concurrent.RateLimiter limiter,
                            CompactionParams.TombstoneOption tombstoneOption)

Method Detail

maybeRefreshOverlaps

public void maybeRefreshOverlaps()

getFullyExpiredSSTables

public java.util.Set<SSTableReader> getFullyExpiredSSTables()

getFullyExpiredSSTables

public static java.util.Set<SSTableReader> getFullyExpiredSSTables(ColumnFamilyStore cfStore,
                                                                   java.lang.Iterable<SSTableReader> compacting,
                                                                   java.lang.Iterable<SSTableReader> overlapping,
                                                                   int gcBefore,
                                                                   boolean ignoreOverlaps)

Finds expired sstables works something like this; 1. find "global" minTimestamp of overlapping sstables, compacting sstables and memtables containing any non-expired data 2. build a list of fully expired candidates 3. check if the candidates to be dropped actually can be dropped (maxTimestamp < global minTimestamp) - if not droppable, remove from candidates 4. return candidates.

Parameters:

cfStore -

compacting - we take the drop-candidates from this set, it is usually the sstables included in the compaction

overlapping - the sstables that overlap the ones in compacting.

gcBefore -

ignoreOverlaps - don't check if data shadows/overlaps any data in other sstables

Returns:

getFullyExpiredSSTables

public static java.util.Set<SSTableReader> getFullyExpiredSSTables(ColumnFamilyStore cfStore,
                                                                   java.lang.Iterable<SSTableReader> compacting,
                                                                   java.lang.Iterable<SSTableReader> overlapping,
                                                                   int gcBefore)

getPurgeEvaluator

public java.util.function.LongPredicate getPurgeEvaluator(DecoratedKey key)

Specified by:

getPurgeEvaluator in class AbstractCompactionController

Parameters:

key -

Returns:

a predicate for whether tombstones marked for deletion at the given time for the given partition are purgeable; we calculate this by checking whether the deletion time is less than the min timestamp of all SSTables containing his partition and not participating in the compaction. This means there isn't any data in those sstables that might still need to be suppressed by a tombstone at this timestamp.

close

public void close()

compactingRepaired

public boolean compactingRepaired()

Specified by:

compactingRepaired in class AbstractCompactionController

shadowSources

public java.lang.Iterable<UnfilteredRowIterator> shadowSources(DecoratedKey key,
                                                               boolean tombstoneOnly)

Overrides:

shadowSources in class AbstractCompactionController

ignoreOverlaps

protected boolean ignoreOverlaps()

Is overlapped sstables ignored Control whether or not we are taking into account overlapping sstables when looking for fully expired sstables. In order to reduce the amount of work needed, we look for sstables that can be dropped instead of compacted. As a safeguard mechanism, for each time range of data in a sstable, we are checking globally to see if all data of this time range is fully expired before considering to drop the sstable. This strategy can retain for a long time a lot of sstables on disk (see CASSANDRA-13418) so this option control whether or not this check should be ignored.

Returns:

false by default