hbase-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From stack <st...@duboce.net>
Subject Re: HBase looses regions.
Date Tue, 26 May 2009 17:18:44 GMT
On Tue, May 26, 2009 at 8:41 AM, llpind <sonny_heer@hotmail.com> wrote:
>
> Haven't tried the RC yet, but regions do get lost when i do intensive write
> operations (e.g. they are no longer listed under online regions).


Have you tried writing less virulently?  Maybe we just can't take your
write load on your setup.

Above you say all writes are going to one region only.   How many rows
have gone in in your estimation?  This is the case when hbase starts
up;  all writes go to a single region until sufficient to split.
Maybe enough data has not yet gone in?  Perhaps your write rate is
such that hbase is unable to split?  Try taking load off.  Try
manually splitting regions (See the hbase shell.  Type 'toos' to see
list of admin methods).

I would suggest you not change default flush and region sizes.
We're better able to help if sizes are default.

>
> Will this RC work with Haoop version 0.19.1?


Yes.  An hbase will run on a hadoop of same major and minor version
(They can different in the point version number).

>
> When I mentioned i tried different configurations, I was tweaking different
> properties mentioned in that post.  We are still at a lost for what to do.
>
>
List is short:

+ Up your file descriptors.  1024 is not enough.
+ Up your hadoop xceivers.  256 is too little
+ Set the timeout on dfsclient to 0.

For more detail on how to do above configurations, their exact names
and whether client or server-side config, see the FAQ and
troubleshooting.

Let us know the particular issues you are running into.  We'd like to help out.

St.Ack





>
> stack-3 wrote:
>>
>> The RC has fixes to help w/ regionserver/master disagreement as to who
>> has what regions; i.e. "region loss".  You might give it a go?
>> St.Ack
>>
>> On Sun, May 24, 2009 at 10:33 AM, llpind <sonny_heer@hotmail.com> wrote:
>>>
>>> Hey Stack, I'm using 0.19.1.  Also, would like to know if I should check
>>> out
>>> the latest and try that or try the RC you mentioned above.
>>>
>>>
>>> stack-3 wrote:
>>>>
>>>> Are you using TRUNK (If you have answered this question already,
>>>> please excuse my not remembering)?
>>>>
>>>> St.Ack
>>>>
>>>> On Sat, May 23, 2009 at 2:17 PM, llpind <sonny_heer@hotmail.com> wrote:
>>>>>
>>>>> I see similar behavior in my small cluster.  (1 master, 3 datanodes)
>>>>>
>>>>> I am also planning on trying this RC version.  I've tried various
>>>>> configurations, and I continue to lose Regions with intensive writes.
>>>>>  I
>>>>> really hope something like this will work, because we are starting to
>>>>> consider other options now.
>>>>>
>>>>> --
>>>>> View this message in context:
>>>>> http://www.nabble.com/HBase-looses-regions.-tp23657983p23688361.html
>>>>> Sent from the HBase User mailing list archive at Nabble.com.
>>>>>
>>>>>
>>>>
>>>>
>>>
>>> --
>>> View this message in context:
>>> http://www.nabble.com/HBase-looses-regions.-tp23657983p23696025.html
>>> Sent from the HBase User mailing list archive at Nabble.com.
>>>
>>>
>>
>>
>
> --
> View this message in context: http://www.nabble.com/HBase-looses-regions.-tp23657983p23725775.html
> Sent from the HBase User mailing list archive at Nabble.com.
>
>

Mime
View raw message