hbase-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Jean-Marc Spaggiari <jean-m...@spaggiari.org>
Subject Re: Review of "HBASE-10201 Port 'Make flush decisions per column family' to trunk" please
Date Wed, 05 Nov 2014 14:12:51 GMT
Ok. DEFAULT_MEMSTORE_COLUMNFAMILY_FLUSH_SIZE_LOWER_BOUND seems to prevent
this. Sorry about that. I will read the entire JIRA before making any other
comment ;)

2014-11-05 8:51 GMT-05:00 Jean-Marc Spaggiari <jean-marc@spaggiari.org>:

> Ok. Small question on my side.
>
> Let's say I have a VERY slow growing CF and a rapidly growing CF.
>
> initially fist is let say 1MB and 2nd too.
>
> 2nd grows quickly to 127MB and get flushed, fine.
>
> Now, 1st grows slowly to 10mb, then 20, then 60, then 100... But since the
> 2nd is always more active, the Memstore might reachs its limit when the 2nd
> is written, and we might end up writting small files again and again
> because the 1st slowly growing CF has not been flushed...
>
> So I don't think flushing a single CF is always good. This might just move
> the issue into another location. We need to have some logic to decide when
> we flush all vs when we flush only one.
>
> My 2ยข ;)
>
> I will go in HBASE-10201 and read...
>
> JM
>
> 2014-10-30 17:07 GMT-04:00 Ted Yu <yuzhihong@gmail.com>:
>
> This is a feature users have been asking for.
>>
>> +1 to turning it on in master branch.
>>
>> Cheers
>>
>> On Thu, Oct 30, 2014 at 1:49 PM, Stack <stack@duboce.net> wrote:
>>
>> > This is a very nice feature.  Its been reviewed once.  Needs another
>> > review. As is, it is on by default. IMO this is how it should be as I
>> think
>> > what we have currently -- region-wide flush --  is 'broke' (smile).
>> Would
>> > be good to get into branch-1.  Anyone else up for a +1?
>> >
>> > Thanks,
>> > St.Ack
>> >
>>
>
>

Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message