Last committed window

[Chandni Singh <ch...@datatorrent.com>]

Hi,

I was working on managed state and felt that some cleanup can be easier if
the last committed window id is accessible to all operator during setup -
maybe a DAG level Attribute.

I don't think this requires a lot of code change. Also just an additional
attribute so just an attribute addition to API.

I would like to work on this change if they aren't any objections.

Thanks,
Chandni

Re: Last committed window

[Chandni Singh <ch...@datatorrent.com>]

I would NOT have to depend on the next committed callback to do some
transfer of data or cleanup.
Since I have to wait till the next committed callback, I need to hold this
data separately and handle it when I get the next committed callback.

If the committed window is provided in setup this can be contained to
setup. I can't think of any downside of providing it.

We added ACTIVATION_WINDOW_ID which is also a convenience, so the same
argument can be made for this.

Thanks,
Chandni

On Thu, Mar 10, 2016 at 2:30 PM, Pramod Immaneni <pr...@datatorrent.com>
wrote:

> Can you provide an example of where this would make your operator purge
> logic more straightforward?
>
> Thanks
>
> On Thu, Mar 10, 2016 at 1:30 PM, Chandni Singh <ch...@datatorrent.com>
> wrote:
>
> > Hi,
> >
> > I was working on managed state and felt that some cleanup can be easier
> if
> > the last committed window id is accessible to all operator during setup -
> > maybe a DAG level Attribute.
> >
> > I don't think this requires a lot of code change. Also just an additional
> > attribute so just an attribute addition to API.
> >
> > I would like to work on this change if they aren't any objections.
> >
> > Thanks,
> > Chandni
> >
>

Re: Last committed window

[Pramod Immaneni <pr...@datatorrent.com>]

Can you provide an example of where this would make your operator purge
logic more straightforward?

Thanks

On Thu, Mar 10, 2016 at 1:30 PM, Chandni Singh <ch...@datatorrent.com>
wrote:

> Hi,
>
> I was working on managed state and felt that some cleanup can be easier if
> the last committed window id is accessible to all operator during setup -
> maybe a DAG level Attribute.
>
> I don't think this requires a lot of code change. Also just an additional
> attribute so just an attribute addition to API.
>
> I would like to work on this change if they aren't any objections.
>
> Thanks,
> Chandni
>