bq. revive some notion of tiered compaction

Did you have a chance to try out Stripe compaction ?

Thanks

On Mon, Aug 3, 2015 at 11:14 AM, Dave Latham <[email protected]> wrote:

> Jean-Marc,
>
> "Recent" is often last 24 hours or so, though if this is worked out I may
> use it for other ranges as well.  Yes, currently there are weekly major
> compactions, so recently compacted regions would not be able to exclude the
> old store files. That's why I'm also hoping to revive some notion of tiered
> compaction to keep older data in separate store files from recent data.
>
> Dave
>
> On Sun, Aug 2, 2015 at 6:22 AM, Jean-Marc Spaggiari <
> [email protected]
> > wrote:
>
> > Just thinking at loud :
> > "Cutting out the old store files could well also reduce disk IO for
> > that family by 100x."
> >
> > What is "recent"  for your data? More than 7 days?  Or less? Don't you
> have
> > weekly major compactions?  If so and if you are scanning for  more than 7
> > days,  then you will read the older files anyway, no?
> >
> > JM
> > Le 2015-08-02 05:57, "Ted Yu" <[email protected]> a écrit :
> >
> > > Dave:
> > > I wonder if Filter response can be enhanced in the following manner:
> > >
> > > http://pastebin.com/sb6apTPm
> > >
> > > My approach is based on using essential column family (column family A
> in
> > > your case) to guide whether the remaining column families should be
> > loaded.
> > > To be specific, if outside the TimeRange you specify (last day), your
> > > filter returns ReturnCode.INCLUDE_AND_SEEK_NEXT_ROW.
> > >
> > > What do you think ?
> > >
> > > Cheers
> > >
> > > On Sat, Aug 1, 2015 at 8:06 PM, Dave Latham <[email protected]>
> wrote:
> > >
> > > > Thanks for brainstorming, Ted.  That sounds like option 2 I listed
> > using
> > > a
> > > > separate scanner for A vs B which "adds complexity to the job and
> gives
> > > up
> > > > the atomicity/consistency guarantees as new writes hit both column
> > > > families".
> > > >
> > > > On Sat, Aug 1, 2015 at 9:07 AM, Ted Yu <[email protected]> wrote:
> > > >
> > > > > Can you achieve your goal with two scans ?
> > > > > The first scan specifies TimeRange corresponding to last day. This
> > scan
> > > > > returns both column families.
> > > > > The other scan specifies TimeRange excluding last day. This scan
> > > returns
> > > > > column family A.
> > > > >
> > > > > Cheers
> > > > >
> > > > > On Sat, Aug 1, 2015 at 8:35 AM, Dave Latham <[email protected]>
> > > wrote:
> > > > >
> > > > > > Hi Ted,
> > > > > >
> > > > > > Thanks for the suggestion, but I'm not sure that it helps my case
> > > much.
> > > > > I
> > > > > > wasn't very familiar with the feature, and it doesn't seem very
> > well
> > > > > > documented - I had to go to the source and the originating JIRA
> to
> > > > > > understand how it works.  It sounds like it allows you to mark
> > which
> > > > > column
> > > > > > families the filter operates on ("essential" seems an odd name).
> > If
> > > > any
> > > > > > data from those column families passes the filter, then the scan
> > > loads
> > > > > and
> > > > > > includes data from the remaining families without filtering it.
> In
> > > my
> > > > > > case, it's not clear from a row's family A whether or not family
> B
> > > for
> > > > > that
> > > > > > row is required (though that could probably be added).  Moreover,
> > > even
> > > > > if a
> > > > > > row has recent data, we don't want to load all the old data from
> > that
> > > > > row.
> > > > > > We'd prefer to be able to entirely skip reading the data off disk
> > for
> > > > the
> > > > > > old store files.
> > > > > >
> > > > > > Dave
> > > > > >
> > > > > > On Sat, Aug 1, 2015 at 7:53 AM, Ted Yu <[email protected]>
> > wrote:
> > > > > >
> > > > > > > Have you considered using essential column family feature
> > (through
> > > > > > Filter)
> > > > > > > ?
> > > > > > > In your case A would be the essential column family.
> > > > > > > Within TimeRange for recent data, the filter would return both
> > > column
> > > > > > > families.
> > > > > > > Outside the TimeRange, only family A is returned.
> > > > > > >
> > > > > > > Cheers
> > > > > > >
> > > > > > > On Sat, Aug 1, 2015 at 7:17 AM, Dave Latham <
> [email protected]
> > >
> > > > > wrote:
> > > > > > >
> > > > > > > > I have a table with 2 column families, call them A and B,
> with
> > > new
> > > > > data
> > > > > > > > regularly being added. They are very different sizes: B is
> 100x
> > > the
> > > > > > size
> > > > > > > of
> > > > > > > > A.  Among other uses for this data, I have a MapReduce job
> that
> > > > needs
> > > > > > to
> > > > > > > > read all of A, but only recent data from B (e.g. last day).
> > Here
> > > > are
> > > > > > > some
> > > > > > > > methods I've considered:
> > > > > > > >
> > > > > > > >    1. Use a Filter to get throw out older data from B (this
> is
> > > > what I
> > > > > > > >    currently do).  However, all the data from B still needs
> to
> > be
> > > > > read
> > > > > > > from
> > > > > > > >    disk, causing a disk IO bottleneck.
> > > > > > > >    2. Configure the table input format to read from B only,
> > > using a
> > > > > > > >    TimeRange for recent data, and have each map task open a
> > > > separate
> > > > > > > > scanner
> > > > > > > >    for A (without a TimeRange) then merge the data in the map
> > > task.
> > > > > > > > However,
> > > > > > > >    this adds complexity to the job and gives up the
> > > > > > atomicity/consistency
> > > > > > > >    guarantees as new writes hit both column families.
> > > > > > > >    3. Add a new column family C to the table with an
> additional
> > > > copy
> > > > > of
> > > > > > > the
> > > > > > > >    data in B, but set a TTL on it.  All writes duplicate the
> > data
> > > > > > written
> > > > > > > > to B
> > > > > > > >    and C.  Change the scan to include C instead of B.
> However,
> > > > this
> > > > > > adds
> > > > > > > > all
> > > > > > > >    the overhead of another column family, more writes, and
> > having
> > > > to
> > > > > > set
> > > > > > > > the
> > > > > > > >    TTL to the maximum of any time window I want to scan
> > > > efficiently.
> > > > > > > >    4. Implement an enhancement to HBase's Scan to allow
> giving
> > > each
> > > > > > > column
> > > > > > > >    family its own TimeRange.  The job would then be able to
> > skip
> > > > most
> > > > > > old
> > > > > > > >    large store files (hopefully all of them with tiered
> > > compaction
> > > > at
> > > > > > > some
> > > > > > > >    point).
> > > > > > > >
> > > > > > > > Does anyone have other suggestions?  Would HBase be willing
> to
> > > > accept
> > > > > > > > updating Scan to have different TimeRange's for each column
> > > > families?
> > > > > > > >
> > > > > > > >
> > > > > > > > Dave
> > > > > > > >
> > > > > > >
> > > > > >
> > > > >
> > > >
> > >
> >
>

Reply via email to