org.apache.hadoop.hbase.client

Interface Hbck

All Superinterfaces:

Abortable, AutoCloseable, Closeable

All Known Implementing Classes:

HBaseHbck

@InterfaceAudience.LimitedPrivate(value="HBCK")
public interface Hbck
extends Abortable, Closeable

Hbck fixup tool APIs. Obtain an instance from Connection.getHbck() and call Closeable.close() when done.

WARNING: the below methods can damage the cluster. It may leave the cluster in an indeterminate state, e.g. region not assigned, or some hdfs files left behind. After running any of the below, operators may have to do some clean up on hdfs or schedule some assign procedures to get regions back online. DO AT YOUR OWN RISK. For experienced users only.

Since:

2.0.2, 2.1.1

See Also:

ConnectionFactory, ClusterConnection

Method Summary

Modifier and Type	Method and Description
default List<Long>	assigns(List<String> encodedRegionNames)
List<Long>	assigns(List<String> encodedRegionNames, boolean override) Like Admin.assign(byte[]) but 'raw' in that it can do more than one Region at a time -- good if many Regions to online -- and it will schedule the assigns even in the case where Master is initializing (as long as the ProcedureExecutor is up).
List<Boolean>	bypassProcedure(List<Long> pids, long waitTime, boolean override, boolean recursive) Bypass specified procedure and move it to completion.
void	fixMeta() Fix Meta.
boolean	runHbckChore() Request HBCK chore to run at master side.
List<Long>	scheduleSCPsForUnknownServers()
default List<Long>	scheduleServerCrashProcedure(List<org.apache.hadoop.hbase.shaded.protobuf.generated.HBaseProtos.ServerName> serverNames) Deprecated. since 2.2.1. Will removed in 3.0.0.
List<Long>	scheduleServerCrashProcedures(List<ServerName> serverNames)
Map<String,RegionState.State>	setRegionStateInMeta(Map<String,RegionState.State> nameOrEncodedName2State) Update region state in Meta only.
TableState	setTableStateInMeta(TableState state) Update table state in Meta only.
default List<Long>	unassigns(List<String> encodedRegionNames)
List<Long>	unassigns(List<String> encodedRegionNames, boolean override) Like Admin.unassign(byte[], boolean) but 'raw' in that it can do more than one Region at a time -- good if many Regions to offline -- and it will schedule the assigns even in the case where Master is initializing (as long as the ProcedureExecutor is up).

Methods inherited from interface org.apache.hadoop.hbase.Abortable

abort, abort, isAborted

Methods inherited from interface java.io.Closeable

close

Method Detail

setTableStateInMeta

TableState setTableStateInMeta(TableState state)
                        throws IOException

Update table state in Meta only. No procedures are submitted to open/assign or close/unassign regions of the table.

Parameters:

state - table state

Returns:

previous state of the table in Meta

Throws:

IOException

setRegionStateInMeta

Map<String,RegionState.State> setRegionStateInMeta(Map<String,RegionState.State> nameOrEncodedName2State)
                                            throws IOException

Update region state in Meta only. No procedures are submitted to manipulate the given region or any other region from same table.

Parameters:

nameOrEncodedName2State - list of all region states to be updated in meta

Returns:

previous state of the region in Meta

Throws:

IOException

assigns

List<Long> assigns(List<String> encodedRegionNames,
                   boolean override)
            throws IOException

Like Admin.assign(byte[]) but 'raw' in that it can do more than one Region at a time -- good if many Regions to online -- and it will schedule the assigns even in the case where Master is initializing (as long as the ProcedureExecutor is up). Does NOT call Coprocessor hooks.

Parameters:

override - You need to add the override for case where a region has previously been bypassed. When a Procedure has been bypassed, a Procedure will have completed but no other Procedure will be able to make progress on the target entity (intentionally). This override flag will override this fencing mechanism.

encodedRegionNames - Region encoded names; e.g. 1588230740 is the hard-coded encoding for hbase:meta region and de00010733901a05f5a2a3a382e27dd4 is an example of what a random user-space encoded Region name looks like.

Throws:

IOException

assigns

default List<Long> assigns(List<String> encodedRegionNames)
                    throws IOException

Throws:

IOException

unassigns

List<Long> unassigns(List<String> encodedRegionNames,
                     boolean override)
              throws IOException

Like Admin.unassign(byte[], boolean) but 'raw' in that it can do more than one Region at a time -- good if many Regions to offline -- and it will schedule the assigns even in the case where Master is initializing (as long as the ProcedureExecutor is up). Does NOT call Coprocessor hooks.

Parameters:

override - You need to add the override for case where a region has previously been bypassed. When a Procedure has been bypassed, a Procedure will have completed but no other Procedure will be able to make progress on the target entity (intentionally). This override flag will override this fencing mechanism.

encodedRegionNames - Region encoded names; e.g. 1588230740 is the hard-coded encoding for hbase:meta region and de00010733901a05f5a2a3a382e27dd4 is an example of what a random user-space encoded Region name looks like.

Throws:

IOException

unassigns

default List<Long> unassigns(List<String> encodedRegionNames)
                      throws IOException

Throws:

IOException

bypassProcedure

List<Boolean> bypassProcedure(List<Long> pids,
                              long waitTime,
                              boolean override,
                              boolean recursive)
                       throws IOException

Bypass specified procedure and move it to completion. Procedure is marked completed but no actual work is done from the current state/step onwards. Parents of the procedure are also marked for bypass.

Parameters:

pids - of procedures to complete.

waitTime - wait time in ms for acquiring lock for a procedure

override - if override set to true, we will bypass the procedure even if it is executing. This is for procedures which can't break out during execution (bugs?).

recursive - If set, if a parent procedure, we will find and bypass children and then the parent procedure (Dangerous but useful in case where child procedure has been 'lost'). Does not always work. Experimental.

Returns:

true if procedure is marked for bypass successfully, false otherwise

Throws:

IOException

scheduleServerCrashProcedure

@Deprecated
default List<Long> scheduleServerCrashProcedure(List<org.apache.hadoop.hbase.shaded.protobuf.generated.HBaseProtos.ServerName> serverNames)
                                                     throws IOException

Deprecated. since 2.2.1. Will removed in 3.0.0.

Use scheduleServerCrashProcedures(List) instead.

Throws:

IOException

scheduleServerCrashProcedures

List<Long> scheduleServerCrashProcedures(List<ServerName> serverNames)
                                  throws IOException

Throws:

IOException

scheduleSCPsForUnknownServers

List<Long> scheduleSCPsForUnknownServers()
                                  throws IOException

Throws:

IOException

runHbckChore

boolean runHbckChore()
              throws IOException

Request HBCK chore to run at master side.

Returns:

true if HBCK chore ran, false if HBCK chore already running

Throws:

IOException - if a remote or network exception occurs

fixMeta

void fixMeta()
      throws IOException

Fix Meta.

Throws:

IOException