[HBase-dev] [jira] Created: (HBASE-1574) Client and server APIs to do batch deletes.

[ https://issues.apache.org/jira/browse/HBASE-1574?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12723657#action_12723657 ]

Erik Holstad commented on HBASE-1574:
-------------------------------------

How are you thinking this Ryan, that they should be buffered too, or just batched and sent over?

[ https://issues.apache.org/jira/browse/HBASE-1574?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

stack updated HBASE-1574:
-------------------------

Fix Version/s: (was: 0.20.0)
0.21.0

Moving to 0.21

[ https://issues.apache.org/jira/browse/HBASE-1574?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

stack updated HBASE-1574:
-------------------------

Fix Version/s: (was: 0.21.0)
0.20.1

Pulling into 0.20.1. We need it (I won't up the RPC version when I add the client/server methods since its a pure addition).

[ https://issues.apache.org/jira/browse/HBASE-1574?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

stack updated HBASE-1574:
-------------------------

Attachment: batchdelete.patch

Any chance of a review J-D?

This is not a pretty addition. In particular, the code in HConnectionManager is effectively duplicated. I tried to make a class to hold common code but it just made stuff worse; state is held in too many variables across the full breadth of the method. I factored out what I could. It would help if Put and Delete had an ancestor or even if it was just an interface such as Row with a getRow method in it or ComparableRow.

[ https://issues.apache.org/jira/browse/HBASE-1574?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12755255#action_12755255 ]

stack commented on HBASE-1574:
------------------------------

Hang on.. had an idea.... back in a minute.

[ https://issues.apache.org/jira/browse/HBASE-1574?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12755264#action_12755264 ]

Erik Holstad commented on HBASE-1574:
-------------------------------------

I'm not against having a common base class or interface, like update or so. The reason that we didn't do that in the first place was the different way that we were using them, for example with buffering and so on, but this might change now when we are doing this multi delete stuff.

[ https://issues.apache.org/jira/browse/HBASE-1574?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

stack updated HBASE-1574:
-------------------------

Attachment: batchdelete-v2.patch

This is a bit better. The "common" interface is Row.. things that have a row can implement Row. Row has a getRow method. Thats it. I had RowComparable as the interface but thought it better to leave the two concepts distinct in case we want Row without Comparable.

In this patch, the duplication of code is gone. We instantiate a Batch private utility class to run the batching.

J-D (or Holstad), review please.

I'm not upping the RPC version number because I want this to go into 0.20.1 (upping RPC version will make 0.20.0 incompatible with 0.20.1). I figure its ok because this is straight addtion of new method.

[ https://issues.apache.org/jira/browse/HBASE-1574?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

stack updated HBASE-1574:
-------------------------

Status: Patch Available (was: Open)

[ https://issues.apache.org/jira/browse/HBASE-1574?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12755524#action_12755524 ]

Jean-Daniel Cryans commented on HBASE-1574:
-------------------------------------------

In HRS, why do you copied the code from the delete method instead of calling it in a for()?

In HTable, I think there's something missing in the comment. "If exception, list will have "

I like the refactoring of HCM and that it handles more than Put, that's the way to go. I think processBatchOfRows should have a new name, my goal when I wrote it was that it be the single method to handle any batches but now it's different.

[ https://issues.apache.org/jira/browse/HBASE-1574?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12755528#action_12755528 ]

stack commented on HBASE-1574:
------------------------------

The batch delete is different from delete, no? Would be hard to refactor so they shared commonage.

Yeah, will fix the HTable comment on commit?

Yeah, processBatchOfRows is not a good name but I don't want to change it for this commit to 0.20 branch. I could open new issue for 0.21 where we look at these batch operations -- including batch get -- and figure out better method namings deprecating the old?

One other thing, I was going to change Row to be package private (Any reaction to this new interface? Do you think Row a good name? Should it be HasRow or something?).

Thanks J-D.

St.Ack

[ https://issues.apache.org/jira/browse/HBASE-1574?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12755534#action_12755534 ]

Jean-Daniel Cryans commented on HBASE-1574:
-------------------------------------------

I think Row is good, the rest is ok. +1 that you fix on commit.

[ https://issues.apache.org/jira/browse/HBASE-1574?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12755602#action_12755602 ]

Erik Holstad commented on HBASE-1574:
-------------------------------------

Patch looks good.

I agree that we might have to rethink the structure of that code as it looks today if we want to support all these batch actions, to make it in a proper way.
I would be good to know how people are using the batch methods to see what code we share between the different calls.

[ https://issues.apache.org/jira/browse/HBASE-1574?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

stack updated HBASE-1574:
-------------------------

Resolution: Fixed
Fix Version/s: 0.21.0
Status: Resolved (was: Patch Available)

Commited branch and trunk (Thanks for review J-D and Holstad). I made Row package private and fixed javadoc on commit.