hbase-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Todd Lipcon <t...@cloudera.com>
Subject Re: FW: HBCK reporting of possible mismatch in RS assignment
Date Thu, 22 Sep 2011 05:35:16 GMT
Hey Ramkrishna,

I agree - sounds like a bug to me.

We need to think more carefully about some of these transitions - I
think we basically have an underspecified incorrect version of
two-phase commit in 5 different places in HBase :)

-Todd

On Thu, Sep 22, 2011 at 1:37 PM, Ramkrishna S Vasudevan
<ramakrishnas@huawei.com> wrote:
> Hi All
>
> Any comments on this.
>
> Best Regards
> Ram
> -----Original Message-----
> From: Ramkrishna S Vasudevan [mailto:ramakrishnas@huawei.com]
> Sent: Wednesday, September 21, 2011 7:46 PM
> To: dev@hbase.apache.org
> Subject: HBCK reporting of possible mismatch in RS assignment
>
> Hi All,
>
>
>
> I have a query regarding the report generated by HBCK.
>
>
>
> Assume there are two RS-  RS1 and RS2.
>
> A region R1 is being opened in RS1.  If it takes more time to open
> particularly in the updateMeta part.  So now if the transition from OPENING
> to OPENED fails(due to timeout monitor reassigning) though we clear the info
> of this region in RS1 we do not clear the entry that was updated in Meta.
> So META will have an entry for RS1 hosting R1.
>
>
>
> Now the assignment of R1 to RS2 becomes successful but we have two entries
> in META for the same region.
>
>
>
> If we run the HBCK tool it reports an inconsistency saying region assignment
> in META is RS1 but actually assigned to RS2.
>
> Is this expected or we need to dig in more and make the HBCK aware of such
> scenarios ?
>
>
>
> Pls correct me if am not correct in my query?
>
>
>
> Best Regards
>
> Ram
>
>
>
>
>
>
>



-- 
Todd Lipcon
Software Engineer, Cloudera

Mime
View raw message