+1 makes sense :) On 7 January 2015 at 09:51, Kostas Tzoumas <ktzou...@apache.org> wrote:
> Perhaps we should have both code paths (Spargel and Spargel inside Graph > API) in the code for a while to enable current Spargel users to migrate > their applications before the breaking change. > > This means that you have Spargel inside Graph API as you plan to, but you > do not remove the current Spargel classes in your PR. > > What do you think? > > Kostas > > > > On Tue, Jan 6, 2015 at 10:32 PM, Vasiliki Kalavri < > vasilikikala...@gmail.com > > wrote: > > > Spargel will be included in the graph API. > > We are already using spargel, but we will move its classes to the graph > > API, change them to work with Vertex and Edge types and implement some > > extensions. This is the last big change we need to make and then we can > > include it in flink-addons :-) > > > > -V. > > > > On 6 January 2015 at 22:09, Henry Saputra <henry.sapu...@gmail.com> > wrote: > > > > > Just curious, is the new Graph APIs suppose to replace flink-spargel? > > > > > > - Henry > > > > > > On Tue, Jan 6, 2015 at 3:51 AM, Vasiliki Kalavri > > > <vasilikikala...@gmail.com> wrote: > > > > Hello Aljoscha and everyone, > > > > > > > > Andra and Daniel might also want to give input :) > > > > > > > > - As Kostas already said, we've had some issues with the > TypeExtractor, > > > but > > > > these were mostly due to our lack of understanding of how it works. > It > > > > would be great if we could document the functionality and provide > some > > > tips > > > > on usage of different methods. For example, when to use > createTypeInfo > > > and > > > > when getReturnTypes? When is it necessary to implement > > > ResultTypeQueryable? > > > > etc. > > > > > > > > - An ugly/bothering issue is not being able to return a scalar value > > from > > > > an operation and having to work with a single-value DataSet. I > > understand > > > > this is not trivial to support and we are currently working around > this > > > > using broadcast variables. > > > > > > > > - The lack of a count operator (which is work in progress, no?) > > > > > > > > - Not being able to easily access the group key inside a > reduce/coGroup > > > > UDF. If I'm not mistaken, there was an issue/discussion on this as > > well. > > > > > > > > If anything else comes to mind, I'll let you know! > > > > Overall, I think we've had very few problems and nothing for which we > > > > (Stephan) couldn't find a workaround :-) > > > > > > > > Cheers, > > > > -V. > > > > > > > > PS: I believe we'll be able to make the Graph API (code name Gelly) > PR > > > very > > > > very soon ^^ > > > > > > > > On 6 January 2015 at 11:24, Kostas Tzoumas <ktzou...@apache.org> > > wrote: > > > > > > > >> From what I have been "overhearing" from the Graph API development, > > some > > > >> hurdles were type extractor errors that could not be deciphered > > easily, > > > and > > > >> the inability to bring back data easily back to the client. > > > >> > > > >> But it would be great if Vasia could elaborate! > > > >> > > > >> On Tue, Jan 6, 2015 at 10:39 AM, Aljoscha Krettek < > > aljos...@apache.org> > > > >> wrote: > > > >> > > > >> > Hi, > > > >> > I guess this is mostly directed at Vasia, since you are the main > > > >> > developer behind the graph API right now. (Correct me if I'm > wrong). > > > >> > > > > >> > Could you share some of the difficulties that you encountered > while > > > >> > developing the API and some of the stuff that bothers you. I want > to > > > >> > work on making our core API better for developing specialised > APIs. > > > >> > > > > >> > Cheers, > > > >> > Aljoscha > > > >> > > > > >> > > > > > >
