Open Feature Project, 2 Mar 2023

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: OpenFeature - Project Meeting, March 2nd, 2023

Description

Meeting notes: https://docs.google.com/document/d/1pp6t2giTcdEdVAri_2B1Z6Mv8mHhvtZT1AmkPV9K7xQ
OpenFeature website: https://open-feature.github.io/

A

I'm just staring at myself.

B

C

I mean I'm all right.

A

Yeah um just had our first ever liquid social, um oh yeah, there's only two of us in the company, so he went for a very nice sushi.

C

A

I um open up PR to remove that field. I think I originally was using it and then in one of the conflicts, just lost it and then forgot, I didn't included it at all.

C

Yeah I kind of figured, it was something like that it was. It seemed like it's just it inert. It seemed like to me so yeah yeah.

A

This guy's refusing to join um the binary Bandits and uh I'm gonna Exile them for it. I'm gonna bring up as much as possible until the yields I don't know what it involves. Yet it.

D

Sounds scary, this.

A

Is like a reply coding challenge, but everyone wins a Macbook apparently get taxed on it. If you win it as well, so that's good. Is it an M2 uh I? Don't know it just says Macbook. So it's probably like this big um and you have to it's.

C

A

C

A

Yeah, we're weighing us guys um acceptance to join the team. It's a bit of a privilege, I, don't know why you're not accepting.

C

C

I know Mike's not going to be attending today. I, don't know what our turnout will be, but we uh Justin Abrams, I'm gonna, comment uh or put a topic up, so I guess he's gonna, be here too, so give it a few more minutes.

B

Good morning,.

D

Do we get anyone to volunteer to describe last meeting.

C

C

It doesn't seem like it. At least, we didn't make a note of it if we did.

D

If we don't have any online at mind, retroactively volunteering.

B

Okay, thank you.

C

Okay, we're three after I'm, not sure if Justin's gonna join I assume so like he said, because he he put himself on for a topic but I guess we can come back um so you're, not much of a huge Turner today. I will um really quickly uh go over some of my stuff, so um survey results, kind of interesting I'll share. My screen and I'll make these available to anybody who's interested in them. I, don't just want to make them completely public.

C

um At the moment.

C

C

um I, don't want to make something completely public at the moment, um because there is some like uh you know, company details in here, um but I could we could consider doing that too um I'll quickly go over them, I. Think a lot of this is not surprising.

C

uh A couple of the good things in here is: we did get responses from some lesser-known vendors and ones that I, don't even believe, have have ever shown up to a community meeting before so I think that's good um and I would say that most of the responses are in line with our kind of proposed SDK changes like things like events um and and supporting like a bulk in cash model as well as, like a you know, pulling down a rule definition model um yeah, so I'll go over.

C

Some of the questions um is it first of all? Is there a difference between your client, SDK and server sdks? Largely yes, here! um uh Does your future flag implementation handle flag evaluation, single user client-based applications? Is it done dynamically or bulk in cash? Do we kind of have a combination here like we expected nothing too surprising um this one I thought was interesting, so is there an explicit way of kind of doing a reevaluation of cached Flags, uh largely yes um and I?

C

Think this one this one response here is a slight misunderstanding of the question, but still um still interesting. The question was really about an API to trigger a reevaluation from, like the actual author's perspective, this. This answer kind of mentions that there's a streaming connection connection which sounds more like getting a change event from the back end, but that's also something we ask later um and then there's one answer. This has been a cache flag, so this is probably more of the dynamic evaluation model where rule sets pulled locally um and you can break.

C

If you want to access the survey, you can uh I'll gladly invite you, but you can actually see which, which respondent if they feel that their vendor name uh is responsible for each question, does API provide a means of subscribing to events? Yes, in all cases, except one which says feature is imminent um and then there is, uh you know, event subscription. So we see a lot of on change like change, ready error, um that kind of thing so again consistent with what we have proposed, which I think is good.

C

This one I found really interesting so is classified evaluation performed, synchronously or asynchronously, so I I think this was misinterpreted, because I was very surprised at this distribution.

C

um 50 50., uh then, when I looked into the respondents and looked into their particular apis, um at least in JavaScript on the web, everything was synchronous, I think every single example I looked at so I think this question was slightly misinterpreted as like. Is there some kind of ready event or something like that that you need to wait for um before you can run the flag evaluation um yeah because, like I said, I I looked into the people claiming asynchronous and and the actual apis are not there's no awaiting the actual evaluation in.

B

C

The JavaScript, uh the web sdks at least um the other thing is, maybe maybe people were again thinking of the server briefly uh I'm, not I'm, not exactly sure it might be worth following up on some of the people who who answered asynchronously, where we don't believe that to be the case. um But you can look at the data if you'd like um yeah and then so. This is context updates. So we again answers were not really too surprised about so change in identity.

C

um There's an end point to do this so grabbing a flag from a different from for a particular identity, so Anonymous versus some kind of identity, identity, um some kind of identify method. So you can see that there's like a lot of um a lot of similarity um and then this one is one that looks like it's more Dynamic evaluated. So we could support this as well.

C

um But it's one of those things where kind of the lowest common denominator is probably what we need to support. So I I think how the API actually looks is going to have to be dictated by more of these patterns, um because you can also kind of fit this Paradigm into into the others.

C

um Yeah, what happens to cash flag evaluations and then evaluation context changes. So this is a specific question that you were interested in Pete um and this is there's a lot of variants here. So it's probably worth looking at these individually um Everybody everything's a little bit different and then here do you provide a mechanism for local, disconnected, um half, no and then half yes or something something similar um this one's interesting? Yes, but only on the server side as we regard it as a bad security Vibe.

C

So I think this is like kind of referencing not pushing targeting rules to the client um uh default flag values. Again, this one's pretty highly varied.

C

um Do you like have a default initial flag value that you could set within your SDK bulk evaluation? Again pretty varied um yeah, so they're defined they're generally problematic for analytics. That's two questions, um so this one has a longer answer but yeah again you can. You can take a look at the data.

C

um What are some pain points? You've encountered When developing feature flag, um so this is interesting. It's something that we've talked about a lot. It's hard for developers to know how Network request strategy will behave. People have very different assumptions about what Will and won't trigger a network event.

C

I think this is, to a large extent, what we're trying to cope with too how to deal with the how the different paradigms do asynchronous calls and that that kind of thing um developers constantly worry about the API going down, but not about things that are way more common, such as poor network connectivity, where DNS bad blocks list Etc I also thought that was interesting.

C

um Do you allow developers to create multiple instances of your SDK um that one's kind of interesting uh and then this is a specific question about mobile apps um kind of not not nothing too surprising here, I thought this was a an interesting answer to about how to better support, client-side usage, um yeah and then this last question also like kind of talking about how mobile mobile, SDK or sorry client, sdks, mobile and web are different than server sdks um and, in most cases, they're significantly different uh once they all behave the same.

C

um Oh actually, this this one is also setting they'll, be the same. So definitely interesting, especially on a per response level. I'd recommend you look at it if you're interested, if you don't have access um just ping me. Let me know: I'll put the link in the survey.

C

And if, if you'd, like, just uh just ping me and I'll I'll, give you access um next thing, uh I opened an issue here, which I think is uh worth talking about. It's something Pete and I have been talking about a little bit um uh but I before I open in nofap I wanted to throw the set there, and so Global providers and context and kind of like just like Global State, and how to maybe deal with that.

C

um So we have some interesting back and forth um some people watching the issue haven't commented yet I'm, particularly interested in anyone. um Who's developed an SDK, especially or vendors, on what their thoughts are here um so yeah Justin has a couple takes.

C

um Pete did respond to I, did have to have time eventually to wrap my head around your response and then I responded to some of your points.

C

um And I think the very last thing that I won't take too much time on is um our experimental web SDK, um so so I think uh I've been I've published the I've published periodically an experimental version of the of the JS SDK that um basically works on the client.

C

um What I've been taking the last day or so taking some time doing, is actually refactoring the JavaScript SDK to publish to actually have like a an alpha um pre-release version of the of the of the JS SDK I'm calling the web SDK that's built from the same repo. So there's a common module. Basically it's just using an npm workspace setup, so there's three modules: a shared module, the existing server implementation and the client implementation, so I I soon, hopefully open a PR um that will kind of split those three out.

C

So basically, we'll have the web SDK, the the alpha pre-release web, SDK being built from the existing JS SDK repository um and I. Just like I said, I'm gonna make sure it's only publishing experimental versions that are tagged.

C

Experimental they'll have a pre-release npm tag, so uh it'll say like Dash experimental on the in the version name and the um in the version, but I just wanted to do that, because it's going to make us easier to actually test some of this stuff and prove it out, like even I've, had this branch open in the playground for forever. That has some demo providers in it like a split provider, a launch directly provider, um some.

C

So all the all these client providers- and it's just been like sitting out there for forever um in this one branch I've had and I've, had to keep rebasing on it. So I think that we're probably getting to the point where it's worth um like publishing that, even though it is experimental um just so, people can play with it. They can try it themselves. They can make a provider I'm going to be very loud and clear that it's it's all experimental, um but it doesn't stop people from at least experimenting with it.

C

um So that's that's my plan, I'll open up PR and unless anybody objects, that's that's what I'm going to do.

C

um I! Think! That's it for me.

C

Anybody have any questions about sorry, I kind of went fast through all those three topics, but does anybody have any.

D

Questions about any.

C

Of the stuff that I talked about or comments.

D

With this survey, what's the are we still looking to get um extra responses? Is there, like a you know, like a timeline when we're kind of closing that for new.

C

I I, what we announced was Wednesday um the 20. What whatever last Wednesday was um uh yeah. We announced uh the 22nd yeah, so it was Wednesday the 22nd and but I didn't actually close it until until two days ago. Monday. But we didn't get any more um response. We could keep it open.

C

The only reason why I even wanted to close it was because it was wide open and there's a potential for spam, but now that I've like exported the results hypothetically, we could keep it open and just like see if people add responses to it,.

B

C

I've exported results to a Google sheet, so at this point you know we're we don't have to worry about that. So.

D

It was like, what's the net, what what's the the kind of the the roadmap for getting to uh kind of like a first pass on uh on the client side? Kind of you know, because sorry I'm, like struggling to wake up this morning,.

C

D

C

Actually, I did add the other thing I created. Was this um sorry? Let me get up okay, this this link here.

C

So I did just recently create this.

C

This road map I put in this in the chat.

E

So this is like this is kind of what I think.

C

um Is is more or less the scope for um for what we need to do so we have to close the remaining ofeps. um We had some back and forth on them today. We've closed the disposal, one I think we need to close the static and dynamic con context and the events I think events is very close to being closed um and uh I guess the one that feels more controversial and kind of ties into the global State question is the static versus Dynamic context of but I mean I I think what I'm?

C

What I'm feeling is that going forward with kind of the experimental um SDK like I was mentioning is, is not a bad step um and we could see how that feels. We could, you know, update things in the playground.

C

um There's also a couple: there's a couple adopters right now of open feature um that use it for vendors uh and and use our existing uh SDK for the client. So I would like to specifically reach out to those people and see hey here's our experimental client like would you consider moving to this like? Is there anything missing here? Does it solve some of your problems?

C

So that's personally, what I'd like to do once we get it published, is I'd like to specifically reach out to those people and and point them to an artifact that they could actually use.

D

Static context, local static context over but I think I think sorry, not Global the static versus Dynamic context thing. It feels like that's the the main kind of like thing that we need to kind of commit commit to one way or the other, and then the rest of it will kind of just be following and I know. It's mostly.

C

Yeah I I agree um we'll see, but I'm gonna specifically reach out to those people once we actually have an artifact. That's another reason: why I think it's it's worth doing, um because it's it's one thing to ask somebody to take a look at a published old, starter packs, another one to ask them to like look at some branch that yeah that they have to like custom build so I think that'll be helpful.

C

uh And it looks like Justin um just had a topic, but he's still not here so um I guess we can call it short and sweet unless anybody has anything else, they want to bring up.

D

I could just do a quick update from a conversation we had um in a previous community meeting. We had a kind of a conversation about um kind of uh that, like making documentation clearer around like through people, onboarding and kind of distinguishing between um open feature like as an SDK versus uh flag, D and kind of like getting started and like for those that weren't there in that.

D

In that conversation, there's this kind of um there's been this thing that I think a few people have noticed where new people who are kind of coming to the project think that they need to be like running a kubernetes cluster to kind of just get get started with with open feature and kind of like this.

D

This kind of um com um getting conflated like the open feature as an SDK, with the supports Advantage providers versus flag D, which is like a provider um and one of the things that me and Mike would uh have kind of Taken on well Mike's, taking on and I'm helping him with is uh redoing some of the gang Stars documentation to not start with Flag Day. So Mike has a um an M VAR based provider. I actually have an even more like minimal provider. That's just like it's static in memory.

D

um You just give it like a you know, set of flags and their values, and- and it just is just there basically for for kind of demo purposes or for like also for unit testing, is useful but um yeah.

D

The idea is basically to change that tutorial over kind of getting started so that, rather than it going straight into flag, D kind of start, like super super duper simply with something: that's not like a production ready thing, but like lets, you get to the point that you're using open feature to do future flagging, and then you know it's like great, that's kind of step, one now, let's like upgrade to using like a kind of like a real, a real provider, and then it's like okay, what are the providers you could use?

D

Is you know we support all these different vendors and open source? You know external providers, one of the providers we support is, is flag, deal and we'll we'll be using flag D for kind of, like this example, so kind of trying to make it clearer to people that um that the two are um not kind of the same.

D

So that's the plan.

C

Yeah I think that makes sense, I think that'll, be clear. I mean I, think that things like targeting rules and stuff like that are kind of are kind of important, but I think that for just the Simplicity of people, understanding what the project is environment, like a very simple environment, variable provider, makes sense, and then we can like, as long as we kind of point to the more advanced use cases. I think that make that that makes it quite easy.

D

Yeah and I think it is kind of like how a lot of people it's definitely how I would encourage people to adopt feature. Flagging anyway, is like start with just something super simple, like some release: toggles that are just controlled inside of the application and then once you've got started, then you can go all fancy and kind of use, use like a full-on kind of flagging framework. But um well, you know like a flagging service, but to get started. I normally tell people to do that anyway, because it's like easier to do it incrementally so.

C

Yeah totally agree and and I mean a lot of people who are going to be doing the most basic things. I mean may already literally be using environment variables to control this sort of stuff. So just.

B

C

Yeah being able to pop in and SDK it fronts, it makes sense yep.

D

I think there might be um potentially in the future. We might have like a like some different kind of getting started kind of tutorially type things depending on like where you're coming from so like. If you're brand new to Future Flags there'll be like one path.

D

If you are already using like launch, Darkly or um flagsmith or whatever then there'd be like another path, it's more like a kind of like a migration path and then, if you're like more coming into it from like uh um you've, been using some feature flagging stuff, but you want to kind of use and open. You want to get started with flag D, then there'd be like a another tutorial. Maybe that's around that, but kind of like having having to tough that's more targeted, the the Persona of the person.

D

That's that's getting started um so yeah. Maybe we'll we'll get fancy once we've got. We've got the their first use case covered in. Oh me, witchering on his stalled enough. Hey.

C

um Hey Justin, we were, we were just uh I'm, hoping you can hear me and your audience connected, but we were just about to to call it. um I very quickly went over a few of my topics, including like the the survey, but uh I saw that you had something else we want to talk about. So if you're ready, you're up all right right.

E

On time um yeah so I wanted to talk about the data format.

E

I'm trying to gauge interest in defining a data format for Boolean logic if one doesn't already exist and then using that as a mechanism to build a client-side evaluator.

E

So we can do client-side, flag, evaluation um and just curious if there's interest in that. Does that sound like an like an open feature thing or some plug-in that someone should write on their own and not have an open feature name on it and I'm particularly interested? If there's a vendory type person on the call to get their take.

C

uh I, don't think there is today we have a pretty low turnout um today, but.

C

um I think it might be worth like posting something about it in the slack Channel I've been interested in this topic, specifically because one of the things I'll tip my hand a little bit, but one of the things I'm kind of interested in potentially defining with open Future down the road is a generic um black definition language.

C

uh So, like you know, if you imagine being able to Define flags as like a kubernetes, a standard, kubernetes resource or a terraform resource, and then having providers that would exist for vendors. So you could essentially make your flag definitions portable between vendors.

C

um So one of the things that would be required is in that is like a standard targeting rule definition, I mean you could use that, like you're, proposing on the client or as like a way to describe targeting rules across vendors to be used in like infrastructure as code.

D

I think it's also super interesting for like the idea of having Like rules that are portable from client to server too, like even if you're just kind of the same vendor, but but being able to run them in like a Java, backend and a typical front end or whatever.

E

Thank you: okay, cool yeah, so this interest interest internal to eBay around like very low latency, flag, evaluation uh and I. Think client-side evaluation is kind of lonely only option we reasonably have for that.

D

D

The the split CTO I mean Todd just had a conversation with him um and he was but I like that. That's like one of the things that they're, obviously not. Obviously that's one of the things that they are um passionate about. That split because that's the like one of their kind of differentiators.

B

D

I mean he might not be interested in chatting to you about it, but he may well be um and they've definitely got like 10 years or something of experience of like trying to make make that work. Well, so.

C

D

C

Is how they do their client client-side stuff, the the others.

B

C

B

Yeah yeah, but.

C

They're they're unique in that they actually push the targeting rules to the to the client, yeah yeah, um in terms of so so what I can say from an implementation perspective: um flag D like like the flag, B plan, at least that we have um uh for a reference implementation is to is to do client-side remotely, but um obviously there's like a rules. Engine definition that you can see all about in flag D um and you could easily enhance it or or like Port it to the client.

C

If you wanted to it, uses something called Json logic um uh which is like a basically a json-based rule definition.

C

um So I mean, if you, if you if eBay is really interested, you could also maybe um will lend your you could lend a bigger view there and see. If, if you think some kind of alternate deployment pattern would make sense- um or um you know whatever um the other thing to consider that I've heard thrown around here a couple times in a similar vein- is wassum like.

C

If you have a awesome module, you could hypothetically run that anywhere um and then it will solve your client problems just one time you know uh totally theoretical, but.

D

Am I imagining it or did um Alex at some point, look at using Q for flighty c-u-e.

E

There's been, there's been a discussion about that for sure, because.

D

That's that would be that aside, the wasm thing that would be like my mostly uninformed kind of place, where I would start looking is Q as like a I mean, because that's like right in his wheelhouse right of kind of like adding some logic to a kind of configuration-y kind of stuff.

C

Yeah the reason that I, because I I looked at the queue a little bit too, when we were looking at some of the flight D stuff. The problem that I I've had with Q is that there's not that many implementations of it I think there's a goal, implementation and I think like one or two more, but this is also a few months ago. It could be at a date.

C

um The reason I think we ended up doing Json logic was because it won it's it's easily, serialized to Json, but the other one, and um was that there was like eight different implementations uh down to like Ruby um and and a little bit more obscure languages. Q support is I, guess gonna grow rapidly, but there wasn't that much when I looked at it cool.

E

Yeah that makes sense awesome, I, think that was the only thing I I had.

C

Cool stuff, um yeah I would post it on the channel, but I'm interested. If you reach any conclusions uh there and definitely keep in touch with in terms of like the eBay client-side stuff, because I'm I'm really interested to see how that goes uh as much. However much you can share.

B

E

Yeah I'll I'll, let you know what we what we learned.

C

Okay thanks, everyone.

D

Yeah, well, we didn't get us, we didn't get a volunteer subscribe for next time. Oh.

C

Yeah I suppose we should do that. um Sorry.

C

I'll volunteer myself.

D

D

C