org.apache.hadoop.hbase.mob

Class DefaultMobStoreCompactor

java.lang.Object

org.apache.hadoop.hbase.regionserver.compactions.Compactor<StoreFileWriter>

org.apache.hadoop.hbase.regionserver.compactions.DefaultCompactor

org.apache.hadoop.hbase.mob.DefaultMobStoreCompactor

@InterfaceAudience.Private
public class DefaultMobStoreCompactor
extends org.apache.hadoop.hbase.regionserver.compactions.DefaultCompactor

Compact passed set of files in the mob-enabled column family.

Nested Class Summary

Nested classes/interfaces inherited from class org.apache.hadoop.hbase.regionserver.compactions.Compactor

org.apache.hadoop.hbase.regionserver.compactions.Compactor.CellSinkFactory<S>, org.apache.hadoop.hbase.regionserver.compactions.Compactor.FileDetails, org.apache.hadoop.hbase.regionserver.compactions.Compactor.InternalScannerFactory

Field Summary

Fields inherited from class org.apache.hadoop.hbase.regionserver.compactions.Compactor

COMPACTION_PROGRESS_LOG_INTERVAL, compactionCompression, compactionKVMax, conf, defaultScannerFactory, keepSeqIdPeriod, MAJOR_COMPACTION_DROP_CACHE, MINOR_COMPACTION_DROP_CACHE, progress, store

Constructor Summary

Constructors

Constructor and Description
DefaultMobStoreCompactor(org.apache.hadoop.conf.Configuration conf, HStore store)

Method Summary

Modifier and Type	Method and Description
List<org.apache.hadoop.fs.Path>	compact(org.apache.hadoop.hbase.regionserver.compactions.CompactionRequestImpl request, ThroughputController throughputController, User user) Do a minor/major compaction on an explicit set of storefiles from a Store.
protected boolean	performCompaction(org.apache.hadoop.hbase.regionserver.compactions.Compactor.FileDetails fd, InternalScanner scanner, CellSink writer, long smallestReadPoint, boolean cleanSeqId, ThroughputController throughputController, boolean major, int numofFilesToCompact) Performs compaction on a column family with the mob flag enabled.

Methods inherited from class org.apache.hadoop.hbase.regionserver.compactions.DefaultCompactor

abortWriter, commitWriter, compactForTesting

Methods inherited from class org.apache.hadoop.hbase.regionserver.compactions.Compactor

compact, createScanner, createScanner, createTmpWriter, getProgress

Methods inherited from class java.lang.Object

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

Constructor Detail

DefaultMobStoreCompactor

public DefaultMobStoreCompactor(org.apache.hadoop.conf.Configuration conf,
                                HStore store)

Method Detail

compact

public List<org.apache.hadoop.fs.Path> compact(org.apache.hadoop.hbase.regionserver.compactions.CompactionRequestImpl request,
                                               ThroughputController throughputController,
                                               User user)
                                        throws IOException

Description copied from class: org.apache.hadoop.hbase.regionserver.compactions.DefaultCompactor

Do a minor/major compaction on an explicit set of storefiles from a Store.

Overrides:

compact in class org.apache.hadoop.hbase.regionserver.compactions.DefaultCompactor

Throws:

IOException

performCompaction

protected boolean performCompaction(org.apache.hadoop.hbase.regionserver.compactions.Compactor.FileDetails fd,
                                    InternalScanner scanner,
                                    CellSink writer,
                                    long smallestReadPoint,
                                    boolean cleanSeqId,
                                    ThroughputController throughputController,
                                    boolean major,
                                    int numofFilesToCompact)
                             throws IOException

Performs compaction on a column family with the mob flag enabled. This is for when the mob threshold size has changed or if the mob column family mode has been toggled via an alter table statement. Compacts the files by the following rules. 1. If the Put cell has a mob reference tag, the cell's value is the path of the mob file.

1. If the value size of a cell is larger than the threshold, this cell is regarded as a mob, directly copy the (with mob tag) cell into the new store file.
2. Otherwise, retrieve the mob cell from the mob file, and writes a copy of the cell into the new store file.

2. If the Put cell doesn't have a reference tag.

1. If the value size of a cell is larger than the threshold, this cell is regarded as a mob, write this cell to a mob file, and write the path of this mob file to the store file.
2. Otherwise, directly write this cell into the store file.

3. Decide how to write a Delete cell.

1. If a Delete cell does not have a mob reference tag which means this delete marker have not been written to the mob del file, write this cell to the mob del file, and write this cell with a ref tag to a store file.
2. Otherwise, directly write it to a store file.

After the major compaction on the normal hfiles, we have a guarantee that we have purged all deleted or old version mob refs, and the delete markers are written to a del file with the suffix _del. Because of this, it is safe to use the del file in the mob compaction. The mob compaction doesn't take place in the normal hfiles, it occurs directly in the mob files. When the small mob files are merged into bigger ones, the del file is added into the scanner to filter the deleted cells.

Overrides:

performCompaction in class org.apache.hadoop.hbase.regionserver.compactions.Compactor<StoreFileWriter>

Parameters:

fd - File details

scanner - Where to read from.

writer - Where to write to.

smallestReadPoint - Smallest read point.

cleanSeqId - When true, remove seqId(used to be mvcc) value which is <= smallestReadPoint

throughputController - The compaction throughput controller.

major - Is a major compaction.

numofFilesToCompact - the number of files to compact

Returns:

Whether compaction ended; false if it was interrupted for any reason.

Throws:

IOException