Package org.apache.hadoop.hbase.util

Class HbckTableInfo.HDFSIntegrityFixer

java.lang.Object

org.apache.hadoop.hbase.util.hbck.TableIntegrityErrorHandlerImpl

org.apache.hadoop.hbase.util.HbckTableInfo.IntegrityFixSuggester

org.apache.hadoop.hbase.util.HbckTableInfo.HDFSIntegrityFixer

All Implemented Interfaces:

TableIntegrityErrorHandler

Enclosing class:

HbckTableInfo

class HbckTableInfo.HDFSIntegrityFixer
extends HbckTableInfo.IntegrityFixSuggester

This handler fixes integrity errors from hdfs information. There are basically three classes of integrity problems 1) holes, 2) overlaps, and 3) invalid regions. This class overrides methods that fix holes and the overlap group case. Individual cases of particular overlaps are handled by the general overlap group merge repair case. If hbase is online, this forces regions offline before doing merge operations.

Field Summary

Fields

Modifier and Type	Field	Description
(package private) org.apache.hadoop.conf.Configuration	conf
(package private) boolean	fixOverlaps

Fields inherited from class org.apache.hadoop.hbase.util.HbckTableInfo.IntegrityFixSuggester

errors

Constructor Summary

Constructors

Constructor	Description
HDFSIntegrityFixer(HbckTableInfo ti, HbckErrorReporter errors, org.apache.hadoop.conf.Configuration conf, boolean fixHoles, boolean fixOverlaps)

Method Summary

Modifier and Type	Method	Description
void	handleHoleInRegionChain(byte[] holeStartKey, byte[] holeStopKey)	There is a hole in the hdfs regions that violates the table integrity rules.
void	handleOverlapGroup(Collection<HbckRegionInfo> overlap)	This takes set of overlapping regions and merges them into a single region.
void	handleRegionEndKeyNotEmpty(byte[] curEndKey)	Callback for handling case where a Table has a last region that does not have an empty end key.
void	handleRegionStartKeyNotEmpty(HbckRegionInfo next)	This is a special case hole -- when the first region of a table is missing from META, HBase doesn't acknowledge the existance of the table.
(package private) void	mergeOverlaps(Collection<HbckRegionInfo> overlap)
(package private) void	removeParentsAndFixSplits(Collection<HbckRegionInfo> overlap)
(package private) void	sidelineBigOverlaps(Collection<HbckRegionInfo> bigOverlap)	Sideline some regions in a big overlap group so that it will have fewer regions, and it is easier to merge them later on.

Methods inherited from class org.apache.hadoop.hbase.util.HbckTableInfo.IntegrityFixSuggester

handleDegenerateRegion, handleDuplicateStartKeys, handleOverlapInRegionChain, handleSplit

Methods inherited from class org.apache.hadoop.hbase.util.hbck.TableIntegrityErrorHandlerImpl

getTableInfo, setTableInfo

Methods inherited from class java.lang.Object

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

Field Details

conf

org.apache.hadoop.conf.Configuration conf

fixOverlaps

boolean fixOverlaps

Constructor Details

HDFSIntegrityFixer

HDFSIntegrityFixer(HbckTableInfo ti,
 HbckErrorReporter errors,
 org.apache.hadoop.conf.Configuration conf,
 boolean fixHoles,
 boolean fixOverlaps)

Method Details

handleRegionStartKeyNotEmpty

public void handleRegionStartKeyNotEmpty(HbckRegionInfo next)
                                  throws IOException

This is a special case hole -- when the first region of a table is missing from META, HBase doesn't acknowledge the existance of the table.

Specified by:

handleRegionStartKeyNotEmpty in interface TableIntegrityErrorHandler

Overrides:

handleRegionStartKeyNotEmpty in class HbckTableInfo.IntegrityFixSuggester

Parameters:

next - An HbckRegionInfo of the second region in a table. This should have a non-empty startkey, and can be used to fabricate a first region that has an empty start key.

Throws:

IOException

handleRegionEndKeyNotEmpty

public void handleRegionEndKeyNotEmpty(byte[] curEndKey)
                                throws IOException

Description copied from class: TableIntegrityErrorHandlerImpl

Callback for handling case where a Table has a last region that does not have an empty end key.

Specified by:

handleRegionEndKeyNotEmpty in interface TableIntegrityErrorHandler

Overrides:

handleRegionEndKeyNotEmpty in class HbckTableInfo.IntegrityFixSuggester

Parameters:

curEndKey - The end key of the current last region. There should be a new region with start key as this and an empty end key.

Throws:

IOException

handleHoleInRegionChain

public void handleHoleInRegionChain(byte[] holeStartKey,
 byte[] holeStopKey)
                             throws IOException

There is a hole in the hdfs regions that violates the table integrity rules. Create a new empty region that patches the hole.

Specified by:

handleHoleInRegionChain in interface TableIntegrityErrorHandler

Overrides:

handleHoleInRegionChain in class HbckTableInfo.IntegrityFixSuggester

Parameters:

holeStartKey - key at the beginning of the region hole

holeStopKey - key at the end of the region hole

Throws:

IOException

handleOverlapGroup

public void handleOverlapGroup(Collection<HbckRegionInfo> overlap)
                        throws IOException

This takes set of overlapping regions and merges them into a single region. This covers cases like degenerate regions, shared start key, general overlaps, duplicate ranges, and partial overlapping regions. Cases: - Clean regions that overlap - Only .oldlogs regions (can't find start/stop range, or figure out) This is basically threadsafe, except for the fixer increment in mergeOverlaps.

Specified by:

handleOverlapGroup in interface TableIntegrityErrorHandler

Overrides:

handleOverlapGroup in class TableIntegrityErrorHandlerImpl

Parameters:

overlap - Collection of overlapping regions.

Throws:

IOException

removeParentsAndFixSplits

void removeParentsAndFixSplits(Collection<HbckRegionInfo> overlap)
                        throws IOException

Throws:

IOException

mergeOverlaps

void mergeOverlaps(Collection<HbckRegionInfo> overlap)
            throws IOException

Throws:

IOException

sidelineBigOverlaps

void sidelineBigOverlaps(Collection<HbckRegionInfo> bigOverlap)
                  throws IOException

Sideline some regions in a big overlap group so that it will have fewer regions, and it is easier to merge them later on.

Parameters:

bigOverlap - the overlapped group with regions more than maxMerge

Throws:

IOException