I agree that we also should make this metadata available for traversals since we want to move users away from sending Gremlin scripts as strings and instead use Bytecode based GLVs.
Regarding Gremlin.Net: I think that the implementation would be very similar to how it can be implemented in the Java driver as we tried to stay close to the Java driver in general. The only difference is probably that we currently don't have a ResultSet in Gremlin.Net, but that's only because I didn't see much value in adding that. Metadata would of course be a good argument to also implement a ResultSet in Gremlin.Net and then the implementation should be really basically the same as in the Java driver. Am 28.02.2018 um 16:15 schrieb Stephen Mallette: > I'm fine with using the last response message as the carrier for this > metadata on a particular request. I can't really tell if there is much work > to do on Gremlin Server itself here. It seems like most of the work must > occur on the various drivers (you mention the .NET api, but all of the > drivers would need to support this feature). However, I would think that we > would want Gremlin Server itself to append in some kind metadata (maybe > query time? something easy....) so that we could write tests for the > drivers in TinkerPop itself. There is also the question of how we would > expose this metadata to GLVs which don't see response messages at all. A > traversal might need some metadata itself so that the user could retrieve > the server metadata from that. The implementation between Java and the GLVs > might be different here as the GLV traversal class is typically quite > lightweight and only used for generating bytecode. > > I'm not so sure I like the SubmitAsynWithHeaders() but I don't think too > much about how the .NET driver works. Is there a reason to not always > return metadata? could it be expensive to do so? If we just added an extra > method how would remote traversals configure this option? I think we need > another way. Generally speaking, for Java, I think I would like to see the > metadata available to the ResultSet somehow which would in turn make it > pretty easy to get it on to a Traversal instance once that facility was > made available.....but as to how to enable or disable the return of the > metadata, i'm not sure how that should work just yet - i need to think on > that some more. > > For committers who work on GLVs, please take a look at this thread and > offer your thoughts on how this might work in the GLV driver you tend to > have the most knowledge on. Let's see if we can come to one nice unified > solution. At that point, we can setup a ticket in JIRA and go from there. > > Ashwini, thanks for offering a pull request for this by the way. Once we > get consensus on how to do this, we'll see if tasks need to be divided and > how you might contribute. > > > > On Mon, Feb 26, 2018 at 6:45 PM, Ashwini Singh < > [email protected]> wrote: > >> Hi Stephen, >> >> Thanks for considering the change. >> >> We would be more inclined towards the first approach since the having a >> ping/pong websocket message can be a bit noisy and requires sophisticated >> handling on the client driver side. >> >> For handling multiple response messages. I would suggest to rely on last >> message as these are the metadata for request execution. Partial response >> is very internal to the client drivers (based on limited understanding of >> tinkerpop client drivers :) , correct me if you differ) and can be exposed >> separately (if really needed later). >> >> For implementation, Let us know if we can chip in there and submit PR. The >> high level approach to achieve this is to have corresponding >> SubmitAsynWithHeaders() for every SubmitAsync() that returns a >> encapsulated result with attributes and IReadOnlyCollectio<T>. Let me know >> if you see any concerns adding a new API. >> >> Thanks a lot, >> Ashwini Singh >> Software Engineer @ Azure Cosmos DB >> >> >> >> -----Original Message----- >> From: Stephen Mallette <[email protected]> >> Sent: Friday, February 23, 2018 5:13 PM >> To: [email protected] >> Subject: Re: [Discuss] Expose metadata from Gremlin Server to Clients. >> >> Adding those kinds of details was the reason we had the >> ResponseStatus.attributes Map. I can really only speak for the java driver >> as I only know that one really well (we might need other TinkerPop experts >> to chime in for python, .net and c#). The java driver doesn't really >> present ways to get that information easily under usage that doesn't deal >> directly with RequestMessage directly (which people typically don't do). >> Another thing to think about is that since a single request might return >> multiple ResponseMessage instances you might not want to return that kind >> of data on every response - maybe just to be returned on the first (or >> last >> message) and then we somehow preserve that information and make it >> accessible on the result somehow....we sorta have some kinda of precedent >> for that with side-effect data generated by bytecode based traversals - we >> can probably build in something similar for this sort of thing. >> >> I also toyed with the idea of using ping/pong websocket messages to carry >> general information about the server to the client. Not sure if any of the >> metadata you want to send back would fit in there, but that could be >> another option. >> >> Does any of that sound helpful? >> >> On Fri, Feb 23, 2018 at 3:41 PM, Ashwini Singh < >> [email protected]> wrote: >> >>> Hi All, >>> >>> We are working on to expose metadata as part of gremlin to response to >>> client. The metadata is simply a property bag to provide special >>> message/hints to the client. But currently client libraries strip off >>> everything and only return the data to the client. >>> >>> Specifically, We want to expose details like Request Charge, Rate >>> limiting/Retry policy details etc. In the other scenarios in Cosmos DB >>> we provide these details to the client is through response headers. We >>> did some investigation around this and one of the options is expose >>> these is through response attributes. Gremlin Server can add metadata >>> as part of gremlin response attributes (For example, set the property >>> bag on ResponseStatus.Attributes for Gremlin.Net) that can be >>> serialized by the client drivers to the clients. >>> >>> We would like to learn more if there are precedence around this and >>> if there are any recommended ways to achieve this in Gremlin protocol >>> and client drivers. >>> >>> Thanks a lot, >>> Ashwini Singh >>> Software Engineer @ Azure Cosmos DB >>> >>>
