Mvcc - how long old versions are stored in memory

[Daniel Połaczański <dp...@gmail.com>]

Hi,
I was surprised by one HBase behaviour. I was modyfing one cell in one row
and then I was extracting this row with my custom filter. Timestamp for all
modifications in my database is currently 0.  Suprising for me was that
filter was provided with the previouos versions of a cell, praboboly
because of MVCC mechanism, I would expect that I get only one and the most
actual version for the "moment" when scan operation starts.
To Filter out the old versions of the cell I used ReturnCode.NEXT_COL
instead of SKIP.

Question, how long the old versions of a cell are maintained memory? When
they are cleaned up?
Regards

Re: Mvcc - how long old versions are stored in memory

[Robert Yokota <ra...@gmail.com>]

Related:

https://issues.apache.org/jira/browse/HBASE-17125
https://yokota.blog/2016/10/20/tips-on-writing-custom-hbase-filters/

On Thu, Nov 16, 2017 at 3:49 AM, Daniel Połaczański <dp...@gmail.com>
wrote:

> Hi,
> I was surprised by one HBase behaviour. I was modyfing one cell in one row
> and then I was extracting this row with my custom filter. Timestamp for all
> modifications in my database is currently 0.  Suprising for me was that
> filter was provided with the previouos versions of a cell, praboboly
> because of MVCC mechanism, I would expect that I get only one and the most
> actual version for the "moment" when scan operation starts.
> To Filter out the old versions of the cell I used ReturnCode.NEXT_COL
> instead of SKIP.
>
> Question, how long the old versions of a cell are maintained memory? When
> they are cleaned up?
> Regards
>