hbase-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Matteo Bertozzi (JIRA)" <j...@apache.org>
Subject [jira] [Updated] (HBASE-16519) Procedure v2 - Avoid sync wait on DDLs operation
Date Wed, 31 Aug 2016 14:50:22 GMT

     [ https://issues.apache.org/jira/browse/HBASE-16519?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]

Matteo Bertozzi updated HBASE-16519:
------------------------------------
    Attachment: HBASE-16519-addendum.patch

Thanks [~Apache9], this is actually an AM problem and a bad behavior on the old client. basically
the region cannot be open because the coproc does not exist, the AM will kill the RS. The
client before had an async behavior on a supposedly sync call, so you were exiting early from
modifyTable() but the regions were not available. 

I added an addendum, that expects a failure on modify. at the moment you'll get a timeout
failure on modify because the regions are not up yet. but this will be fixed with the new
AM. the new AM will give info about failed open for missing coproc.

> Procedure v2 - Avoid sync wait on DDLs operation
> ------------------------------------------------
>
>                 Key: HBASE-16519
>                 URL: https://issues.apache.org/jira/browse/HBASE-16519
>             Project: HBase
>          Issue Type: Sub-task
>          Components: master, proc-v2
>    Affects Versions: 2.0.0
>            Reporter: Matteo Bertozzi
>            Assignee: Matteo Bertozzi
>             Fix For: 2.0.0
>
>         Attachments: HBASE-16519-addendum.patch, HBASE-16519-v0.patch
>
>
> Some operation ModifyColumnFamily, AddDeleteColumnFamily, DeleteColumnFamily, ModifyTable,
TruncateTable are still synchronous on the master side. with a wait until the operation completes
before returning. 
> this was done to keep the sync behavior for old client. but instead of using the procLatch
which recognize the client version and decide if the operation should be sync or not it just
always wait. making the client side proc fault tolerance ineffective.
> also the add/delete/modifyColumnFamily operation does not seems to follow the Async()
naming in master. and the comment claim to be async but everyone uses them as sync. (this
is something from HBASE-13538)



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Mime
View raw message