The namenode piggybacks invalidations on block reporting, and if I recall
correctly might send only a percentage of the pending work at each
heartbeat.


On Tue, Jul 15, 2014 at 11:54 AM, Vladimir Rodionov <[email protected]>
wrote:

>  OK, seems they are harmless ... but  I do not do any
> deletes/writes/updates - only reads. There is a substantial delay between
> major compaction finish and last message in  a log file (many minutes).
>
>
> On Tue, Jul 15, 2014 at 11:18 AM, Andrew Purtell <[email protected]>
> wrote:
>
> > The blocks of a deleted file are invalidated by the block manager so the
> > datanode will delete them.
> >
> >
> > On Tue, Jul 15, 2014 at 10:57 AM, Vladimir Rodionov <
> > [email protected]>
> > wrote:
> >
> > > I do some performance testing of parallel scanners over the same store
> > file
> > >
> > > I insert data, than major compact it , make sure that I have only one
> > > region with only one store file. Than I run tests. Sometimes, during
> > tests
> > > I see the following messages in a log file (lot of them):
> > >
> > > 2014-07-15 10:54:00,378 [INFO|BlockStateChange|BlockManager] BLOCK*
> > > addToInvalidates: blk_1073742034_1210 127.0.0.1:49599
> > >
> > > Can someone tell me what do they mean?
> > >
> > > best,
> > >
> > > -Vladimir Rodionov
> > >
> > > PS
> > >
> > > HBase 0.98.3
> > >
> ​
> ​
> ​
> ​
> ​
>
>


-- 
Best regards,

   - Andy

Problems worthy of attack prove their worth by hitting back. - Piet Hein
(via Tom White)

Reply via email to