falcon-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Srikanth Sundarrajan <srik...@hotmail.com>
Subject RE: [DISCUSS] Apache Falcon 1.0 release
Date Wed, 06 Apr 2016 09:13:00 GMT
It is about time. We have been talking about 1.0 for nearly year and a half now and Thanks
Ajay for initiating this discussion. I am hoping that this will push us closer towards 1.0.

In general, we have been making more regular releases in the recent months and that has greatly
helped us to get better and faster feedback on features that are built and also has allowed
to be more iterative with our development. An attempt to get to 1.0 allows us to share with
our user community a stable set of APIs that they can safely assume to not break across releases,
the sooner we get there, the better it is for them.

Going through the proposals Ajay has put forth, my reading is as follows

Approach #1, auto converts entities from older version to newer and this auto convert will
be withdrawn from the system after few version in 1.0 line
Approach #2, auto converts entities, but retains the original entity definition as is for
user submissions & modification operations and this is perpetual 

I find Approach #2 to be far more suitable for the following reasons:

1> The initial implementation required for both approaches are nearly identical 
2> Users will be less confused when operating with the system
3> Approach #2 is essentially putting forth a model where there is semantic compatibility
across version with syntactic differences in entity or api definitions.

Should it help, I can do a small POC to allay any fears relating to complexity with approach

If there are more options in managing this, those should also be listed down for discussions.

Srikanth Sundarrajan

> From: ajayyadava@apache.org
> Date: Wed, 6 Apr 2016 00:13:04 +0530
> Subject: [DISCUSS] Apache Falcon 1.0 release
> To: dev@falcon.apache.org
> Hello everyone,
> For quite some time we have been discussing to make a 1.0 release and have
> had several discussions in developer sync up around it. Taking this to next
> step, I propose next release line(after 0.10) of Apache Falcon to be 1.0
> *Item 1 - Scope and Timelines*
> Some of the items that are in works and I personally feel will be idle
> candidate for 1.0 are - clean up our APIs(add a new version), introduce a
> new shell for Falcon feed sla alerts, and move to the more powerful and
> capable lifecycle framework for feeds among few.
> After lot of thought and discussions with several members, I propose to not
> aim for too many big features and a timeline of 2.5-3 months after the 0.10
> release. This will ensure that critical fixes are not delayed and there is
> only one active working line for code. We can add more features if other
> community members are able to get them committed in time and our quality
> team also feels comfortable.
> *Item 2 - Migration Strategy*
> While some of the changes like REST API clean up etc. can be done by adding
> versioning others like migration to lifecycle framework etc. are bit more
> involved. One important decision to be made is how to migrate to 1.0
> release.
> Here are some of the options to migrate to new entity definitions in 1.0
> (NOTE: REST api can be versioned and same end points can continue to work
> albeit with newer definitions)
> *Approach 1*. Take a one time hit, call the release backward incompatible
> and provide changes inside falcon to automatically migrate to newer
> definitions on start up. We can support this migration code for a couple of
> releases and then later on remove it.
>  pros:
> - Clean and easy to code -  no if else etc. for supporting features in
> multiple manners.
> - Intuitive for users - multiple options for same purpose are confusing for
> the users.
> - Easier to maintain - All bug fixes need to be done only in one code flow
> and not at many places.
> cons:
> - involves migration - it can be automated by incorporating the migration
> as part of falcon startup.
> *Approach 2*. Support both old and new entity definitions.
> pros
> - users can work with both versions and migrate at their own pace
> cons
> - Hard to maintain - Lot of code will need to be duplicated (validations
> etc.) or branched (wf builders etc.) All bug fixes will need to be fixed in
> multiple places.
> - Not scalable approach - The code will not scale easily if we decide to
> support one more version.
> - Manual migration - Users will have to migrate themselves to the new
> entity definitions.
> - Gotchas - What will happen if someone submits in newer version but calls
> update in older version?
> I invite all of you to provide your thoughts on both the items. There might
> be more approaches or points to consider, suggestions are welcome.
> Cheers
> Ajay Yadava
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message