[
https://issues.apache.org/jira/browse/HBASE-4358?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13101551#comment-13101551
]
[email protected] commented on HBASE-4358:
------------------------------------------------------
bq. On 2011-09-09 19:48:45, Ted Yu wrote:
bq. >
/src/main/java/org/apache/hadoop/hbase/master/handler/TableAddFamilyHandler.java,
line 64
bq. > <https://reviews.apache.org/r/1768/diff/1/?file=38941#file38941line64>
bq. >
bq. > After HBASE-451, changes to table descriptor have to be persisted to
HDFS.
bq. >
bq. > I browsed handleTableOperation() methods in this patch and didn't
find that.
I'm not familiar enough with how exactly table descriptors are persisted to be
able to tell you for certain that this approach correctly ensures persistence.
But I can confidently tell you that this diff does everything that the current
trunk does with regards to updating table descriptors. If you look at the
actual implementation of MasterFileSystem.{add,modify,Delete}Column(), it gets
a table descriptor from master services, modifies the table descriptor
appropriately, then adds it back to master services' table descriptors. Between
handleTableOperation() and updateTableDescriptor(), this patch follows the
exact same procedure, and has the same instance of MasterServices. This
separation is for the purpose of enabling batching to happen in a way that
doesn't leave the system in an intermediate state in the case of a thrown
exception. From basic testing, I see that the table descriptor changes are
actually being written to my fs. If this procedure is not enough to actually
ensure that this happens, we should file a separate JIRA to look into it.
- Riley
-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/1768/#review1843
-----------------------------------------------------------
On 2011-09-09 18:39:05, Riley Patterson wrote:
bq.
bq. -----------------------------------------------------------
bq. This is an automatically generated e-mail. To reply, visit:
bq. https://reviews.apache.org/r/1768/
bq. -----------------------------------------------------------
bq.
bq. (Updated 2011-09-09 18:39:05)
bq.
bq.
bq. Review request for hbase.
bq.
bq.
bq. Summary
bq. -------
bq.
bq. Currently, the RPC provides no way of asking for several table alterations
at once, and the master has no way of batch handling alter requests. Thus, when
the user requests several changes at the same time (i.e. add these I columns,
delete these J columns, and modify these K columns), each region is brought
down (I+J+K) times so that it can reflect the new schema. Additionally,
multiple writes are made to META, and multiple RPC calls must be made.
bq.
bq. This patch provides batching for these operations, both at the RPC level
and within the Master's TableEventHandlers. This involves a bit of
reorganization in the TableEventHandler class hierarchy, and a new
TableEventHandler, TableMultiFamilyHandler. The net effect ends up being the
difference seen here:
bq.
bq. Before patch:
bq. hbase(main):001:0> alter 'peeps', {NAME => 'rawr'}, {METHOD => 'delete',
NAME => 'name'}
bq. Updating all regions with the new schema...
bq. 1/1 regions updated.
bq. Done.
bq. Updating all regions with the new schema...
bq. 1/1 regions updated.
bq. Done.
bq. 0 row(s) in 2.6450 seconds
bq.
bq. After patch:
bq. hbase(main):002:0> alter 'peeps', {NAME => 'rawr'}, {METHOD => 'delete',
NAME => 'name'}
bq. Updating all regions with the new schema...
bq. 1/1 regions updated.
bq. Done.
bq. 0 row(s) in 1.1930 seconds
bq.
bq. Regions are only brought down once, and the duration is cut 1/N.
bq.
bq.
bq. This addresses bug HBASE-4358.
bq. https://issues.apache.org/jira/browse/HBASE-4358
bq.
bq.
bq. Diffs
bq. -----
bq.
bq. /src/main/java/org/apache/hadoop/hbase/client/HBaseAdmin.java 1166933
bq.
/src/main/java/org/apache/hadoop/hbase/coprocessor/BaseMasterObserver.java
1166933
bq. /src/main/java/org/apache/hadoop/hbase/coprocessor/MasterObserver.java
1166933
bq. /src/main/java/org/apache/hadoop/hbase/executor/EventHandler.java
1166933
bq. /src/main/java/org/apache/hadoop/hbase/ipc/HMasterInterface.java 1166933
bq. /src/main/java/org/apache/hadoop/hbase/master/HMaster.java 1166933
bq. /src/main/java/org/apache/hadoop/hbase/master/MasterCoprocessorHost.java
1166933
bq.
/src/main/java/org/apache/hadoop/hbase/master/handler/TableAddFamilyHandler.java
1166933
bq.
/src/main/java/org/apache/hadoop/hbase/master/handler/TableDeleteFamilyHandler.java
1166933
bq.
/src/main/java/org/apache/hadoop/hbase/master/handler/TableFamilyHandler.java
PRE-CREATION
bq.
/src/main/java/org/apache/hadoop/hbase/master/handler/TableModifyFamilyHandler.java
1166933
bq.
/src/main/java/org/apache/hadoop/hbase/master/handler/TableMultiFamilyHandler.java
PRE-CREATION
bq. /src/main/ruby/hbase/admin.rb 1166933
bq.
/src/test/java/org/apache/hadoop/hbase/coprocessor/TestMasterObserver.java
1166933
bq.
bq. Diff: https://reviews.apache.org/r/1768/diff
bq.
bq.
bq. Testing
bq. -------
bq.
bq. Sanity checked functionality in psuedo-distributed mode (tried several
permutations of different alterations, all completed successfully and with only
one round of region restarts). Ran all unit tests successfully.
bq.
bq.
bq. Thanks,
bq.
bq. Riley
bq.
bq.
> Batch Table Alter Operations
> ----------------------------
>
> Key: HBASE-4358
> URL: https://issues.apache.org/jira/browse/HBASE-4358
> Project: HBase
> Issue Type: Improvement
> Components: ipc, master, shell
> Affects Versions: 0.92.0
> Reporter: Riley Patterson
> Assignee: Riley Patterson
> Priority: Minor
> Attachments: HBASE-4358.patch
>
>
> Currently, the RPC provides no way of asking for several table alterations at
> once, and the master has no way of batch handling alter requests. Thus, when
> the user requests several changes at the same time (i.e. add these I columns,
> delete these J columns, and modify these K columns), each region is brought
> down (I+J+K) times so that it can reflect the new schema. Additionally,
> multiple writes are made to META, and multiple RPC calls must be made.
> This patch provides batching for these operations, both at the RPC level and
> within the Master's TableEventHandlers. This involves a bit of reorganization
> in the TableEventHandler class hierarchy, and a new TableEventHandler,
> TableMultiFamilyHandler. The net effect ends up being the difference seen
> here:
> Before patch:
> hbase(main):001:0> alter 'peeps', {NAME => 'rawr'}, {METHOD => 'delete', NAME
> => 'name'}
> Updating all regions with the new schema...
> 1/1 regions updated.
> Done.
> Updating all regions with the new schema...
> 1/1 regions updated.
> Done.
> 0 row(s) in 2.6450 seconds
> After patch:
> hbase(main):002:0> alter 'peeps', {NAME => 'rawr'}, {METHOD => 'delete', NAME
> => 'name'}
> Updating all regions with the new schema...
> 1/1 regions updated.
> Done.
> 0 row(s) in 1.1930 seconds
> Regions are only brought down once, and the duration is cut 1/N.
--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira
