Cold standby for oak-segment-tar

[Francesco Mari <fr...@apache.org>]

Hi all,

The cold standby functionality originally implemented for oak-segment
is currently unusable with oak-segment-tar. The cold standby requires
access to the SegmentStore API and its lower level functionalities,
but this API has been changed in oak-segment-tar in an incompatible
way.

I would like to restore the possibility of running the cold standby on
top of oak-segment-tar, but I can only see two ways of achieving this.

1. Assume that the cold standby functionality is intimately coupled
with oak-segment, duplicate its code in oak-segment-tar and make the
appropriate adjustments to make it work. No change will be done in
oak-segment or oak-tarmk-standby.

2. Define an API (or, more appropriately, an SPI) sitting in between
oak-tarmk-standby and the segment stores. This API will have two
implementations in oak-segment and oak-segment-tar and will isolate
oak-tarmk-standby from the differences of the two modules.

I would be in favour for the second option for the obvious benefits of
code reuse. What do you guys think about this problem? Is there any
other option I'm missing?

Re: Cold standby for oak-segment-tar

[Alex Parvulescu <al...@gmail.com>]

Hi,

The second option sounds best to me.

alex


On Mon, Aug 15, 2016 at 3:00 PM, Francesco Mari <fr...@apache.org> wrote:

> Hi all,
>
> The cold standby functionality originally implemented for oak-segment
> is currently unusable with oak-segment-tar. The cold standby requires
> access to the SegmentStore API and its lower level functionalities,
> but this API has been changed in oak-segment-tar in an incompatible
> way.
>
> I would like to restore the possibility of running the cold standby on
> top of oak-segment-tar, but I can only see two ways of achieving this.
>
> 1. Assume that the cold standby functionality is intimately coupled
> with oak-segment, duplicate its code in oak-segment-tar and make the
> appropriate adjustments to make it work. No change will be done in
> oak-segment or oak-tarmk-standby.
>
> 2. Define an API (or, more appropriately, an SPI) sitting in between
> oak-tarmk-standby and the segment stores. This API will have two
> implementations in oak-segment and oak-segment-tar and will isolate
> oak-tarmk-standby from the differences of the two modules.
>
> I would be in favour for the second option for the obvious benefits of
> code reuse. What do you guys think about this problem? Is there any
> other option I'm missing?
>

Re: Cold standby for oak-segment-tar

[Michael Dürig <md...@apache.org>]

I would go with 2. if possible.

Michael

On 15.8.16 3:00 , Francesco Mari wrote:
> Hi all,
>
> The cold standby functionality originally implemented for oak-segment
> is currently unusable with oak-segment-tar. The cold standby requires
> access to the SegmentStore API and its lower level functionalities,
> but this API has been changed in oak-segment-tar in an incompatible
> way.
>
> I would like to restore the possibility of running the cold standby on
> top of oak-segment-tar, but I can only see two ways of achieving this.
>
> 1. Assume that the cold standby functionality is intimately coupled
> with oak-segment, duplicate its code in oak-segment-tar and make the
> appropriate adjustments to make it work. No change will be done in
> oak-segment or oak-tarmk-standby.
>
> 2. Define an API (or, more appropriately, an SPI) sitting in between
> oak-tarmk-standby and the segment stores. This API will have two
> implementations in oak-segment and oak-segment-tar and will isolate
> oak-tarmk-standby from the differences of the two modules.
>
> I would be in favour for the second option for the obvious benefits of
> code reuse. What do you guys think about this problem? Is there any
> other option I'm missing?
>