08:01:54 RRSAgent has joined #social 08:01:54 logging to http://www.w3.org/2016/09/23-social-irc 08:01:56 RRSAgent, make logs public 08:01:58 Zakim, this will be SOCL 08:01:58 ok, trackbot 08:01:59 Meeting: Social Web Working Group Teleconference 08:01:59 Date: 23 September 2016 08:02:58 present+ 08:02:59 present+ 08:02:59 present+ 08:03:01 present+ 08:03:03 Arnaud has joined #social 08:03:30 AnnBass has joined #social 08:03:41 present+ AnnBass 08:03:46 present+ Arnaud 08:04:10 scribenick: AnnBass 08:04:25 Tantek: begin with agenda scheduling 08:04:58 cwebber2 shows mediagoblin video 08:06:51 applicable to social web concepts 08:07:21 shepazu has joined #social 08:09:04 http://mediagoblin.org/pages/campaign.html 08:09:23 decentralized ... federated ... 08:10:41 08:13:46 tantek: 08:14:11 ... assuming we get through the agenda as listed, afternoon is open 08:14:30 https://www.w3.org/wiki/Socialwg/2016-09-22 08:14:31 Social Web WG Face to Face Meeting in Lisbon (F2F7) 08:14:51 ..... any other suggestions / edits / ? 08:15:09 sandro: 08:15:53 ... we had scheduled for 1:00pm Lisbon, 8:00am ? Evan's time ... may want an hour later 08:16:19 ... hoping Evan will let us know before we go to lunch 08:16:36 tantek: left off with ActivityPub open issues 08:16:56 cwebber2: "source" field 107 08:17:27 ... sugg putting a note in there 08:17:44 ... your source has warning that if you edit this, you'll lose your previous source 08:18:49 rhiaro: yesterday it felt like blocking the user, but this seems like a warning 08:19:10 cwebber2: adds source, provides info for user to know what to do 08:19:52 tantek: what about server choosing what's canonical? 08:20:00 aaronpk: seems OK 08:20:19 cwebber2: I'd like to use this period to try this out, see if we can figure it out 08:20:37 tantek: you have implementation experience? 08:20:43 cwebber2: partially ... 08:21:14 s/partially/we've seen other pumpio implementations 08:21:27 s/seems OK/I wouldn't do it that way, but i'm not going to try to convince you to change to the other model. however this doesn't feel like it's been fully thought out all the way through 08:21:45 ... tsyesika sugg we include history, but now we think that makes it too complicated 08:22:14 sandro: I'm hesitant about the Note only being informative .. esp if user is going to lose source 08:22:38 cwebber2: I'm hesitant to specify a UI when it might be improved 08:22:59 sandro: is there an idea about ActivityPub UI? 08:23:46 cwebber2: partially, "conformant client" ... could do this as "should"; don't think it should be "must" 08:23:56 rhiaro: what about doing something on server? 08:24:04 sandro: I"m cool with "should" 08:24:37 cwebber2: 08:25:02 scribenick: rhiaro 08:25:37 RESOLUTION: Add source field to AP as at risk per proposal in activitypub/issues/107 08:25:42 tantek: There were no objections 08:26:04 arnaud: You sai dyesterday from a process point of view you can do the resolution liek this which is fine, I still think it's good to have the proposal listed in IRC before you do that 08:26:11 tantek: I called for objections, didn't get minuted.. 08:26:14 sandro: for people who are remote 08:26:37 TODO: add proposal to minute above ^^^^^^^^ 08:26:51 cwebber2: 106 is something I could do on my own but i wanted to get some clarity about testing requirements 08:26:57 KevinMarks has joined #social 08:27:02 ... Testing servers is pretty clear to me. I can write a client that hits a bunch of URLs and does things 08:27:04 ... Testing clients is a lot trickier 08:27:09 sandro: is that a validator? 08:27:16 cwebber2: I'm trying to figure it out.. 08:27:21 tantek: you don't have to figure it out right now 08:27:58 cwebber2: I think the general plan is since there are a billion different systesm we can't touch UIs. We don't know the platform, we can't touch all those toolkits. What we can do is provide some sort of lightweight server that a client connects to, you get some prompts about the actiosn you were supposed to do and it asks you whether or not the expected behaviour happens 08:28:35 timbl has joined #social 08:28:37 aaronpk: if it's like micropub where most of the spec is around CUD posts, my plan was to create a reference server that behaves the way that I expect a server to behave and have a bunch of tests that, eg. tell the client to create a post with these properties, to run it however they want to run it no under your control 08:28:43 ... on the serverside you can check did these properties come through 08:28:56 ... you tell the user your client needs to blah, and the server can check 08:29:01 tantek: can you write that down for micropub 08:29:07 ... it's in CR, your test suite plan is up but not that level of detail 08:29:21 csarven: We're doing this as well, do you throw an error as to where they failed or what they should do? 08:29:38 aaronpk: I did a lot of error rpeoritng with webmention in very fine detail in a way I would not do in a live implementation 08:29:47 ... REsponses come back with here is exactly where you failed 08:29:48 ... I'd do the same with micropub 08:30:09 tantek: You said you would have a couple of tests for mp by october 4, canyou include this elvel of detail in your plan for how to do the tests? 08:30:18 ... Sounds like you have thought about it, I didn't hear chris object to the methodology 08:30:22 cwebber2: I think it's a good idea 08:30:32 ... I'm going to record it here 08:30:45 s/You said/aaronpk said 08:30:47 aaronpk: sure 08:31:06 sandro: so this is to clients 08:31:08 cwebber2: yes 08:31:17 tantek: when do you think you can have your tes tplan written up? 08:31:26 cwebber2: It's not the full test plan... 08:31:29 sandro: what do you mean by a test plan? 08:31:37 tantek: what is your approach? 08:31:52 cwebber2: I think I can have a general here's what we're planning on doing? 08:31:58 tantek: it can be in or outside the spec, I want a date 08:32:06 cwebber2: by next meeting or the one after 08:32:09 tantek: so the 4th 08:32:17 ... for a general plan to the level of micropub.rocks 08:32:20 https://github.com/aaronpk/micropub.rocks 08:32:28 https://micropub.rocks/ 08:32:47 ... if you can add more detail like you were just discussing, that's even better 08:32:48 ... that's not a must have 08:32:56 sandro: would be nice if the prompts to the extent possible were the same.. 08:33:04 tantek: general coodrination sounds like a good thing between cwebber2 and aaronpk 08:33:22 cwebber2: and can coordinate with rhiaro and csarven about LDN 08:33:33 ... That wraps up this whole thing, we can do things closely 08:33:55 sandro: the server tests have to have ?? 08:34:07 sandro: you need another server the server can federate to 08:34:20 aaronpk: yes... webmention does that 08:34:37 cwebber2: checking another server is behaving correctly was easy to figure it, it was just when you can't touch the client I was struggling with 08:35:04 https://github.com/aaronpk/micropub.rocks/issues/40 08:35:22 sandro: It was nice to imagine micropub and activitypub clients can test both.. 08:35:36 cwebber2: That's the test *plan*, not the tests? 08:35:39 tantek: Correct 08:35:47 ... So we can have everything in place to go to CR 08:36:00 sandro: I always thought you should have at least one test working... we don' thave to do that now. THe bar is technically test plan 08:36:07 tantek: He can work on that while we're dealing with transition call stuff 08:36:12 ... I'd like for us to be able to propose to the group asap 08:36:26 cwebber2: I know how intesne these few months are gonna be, I'm going to be starting tests as soon as I can 08:36:36 ... One reason I wrote pubstrate this way was to be the basis for something like this 08:36:58 sandro: CR is you're telling the world please try to implement this. If you tell me it's ready to implement I'm gonna say I'll wait til I have the test suite before I implement. I like test driven development. 08:37:12 tantek: That's your feedback as a propsective developer 08:37:22 ... With mp we entered CR with how many implmentations? 08:37:27 aaronpk: A dozen clients, handful of servers 08:37:37 ... Emprirical testing with each other 08:37:57 cwebber2: It would be great if we end up doing it 08:38:04 sandro: micropub is trivial compared 08:38:22 cwebber2: There's one issue left, needs process help 08:38:30 ... 30 JSON-LD context 08:38:37 ... I need to test it more to make sure it expands correctly 08:38:47 tantek: is this beyond the context for as2? 08:38:48 s/compared to the first W3C spec I implemented (OWL Full) so my perspective may be a bit different/ 08:38:54 cwebber2: We have a a few additional things 08:39:04 s/compared/compared to the first W3C spec I implemented (OWL Full) so my perspective may be a bit different/ 08:39:11 ... We have an inbox endoint, outbox, followers, following, they're not defined in AS2 because they're applicable to APIs which talk to each other 08:39:19 ... So ours imports the AS2 context, so you only need to put one on there 08:39:47 ... Maybe if the namespaces thing turns out we can put it in AS2 08:42:08 ... I would be totally happy to have it in AS2. I'ts likely to hit rec before AP does 08:42:19 ... Putting it in the AS2 spec would require another CR 08:42:27 tantek: putting in the namespace is different because AS2 is not at rec 08:42:31 rhiaro: I don't understand how that's different 08:42:40 cwebber2: If we can put this in AS2 that would simplify a lot o fthings 08:42:58 ... I feel I need to talk to Evan and jasnell to find out what their feeling on it is 08:43:06 tantek: and you need to talk to sandro about the implications 08:43:18 ... because there's not a consensus in the community about what adding to namespaces means 08:44:07 cwebber2: And it impacts one thing, currently we say the ld+json with profile, but having the as2 profile means it should be AP. It would be great if we didn't have that question, it would make things a lot easier if we could do things in AS2 namespace 08:44:12 AnnBass has left #social 08:44:17 ... I've always felt this way. I just thought it wasn't possible 08:44:37 sandro: the whole community issue is about letting a WG defend its territory. Nobody would object if we said the AS2 ns is world writeable 08:44:47 ... So that we can add to the ns in other specs, but also that other groups can 08:45:02 ... I almost object to AS2 last december when we discussed CR, saying how to define the extension mechanism 08:45:07 tantek: there's no mechanism? 08:45:17 sandro: Right. The default is in the spec is use another namespace 08:45:31 ... But that is totally messy because you have to use a bunch of namespaces, when you could just put them in if you agreed to do that 08:45:51 ... So something with a community process that says this is okkay put it in the namespace. We couldd o that 08:46:00 tantek: we as a WG might not be arround, so we need to define a community process 08:46:06 cwebber2: I hope we find out soon what that process is 08:46:14 tantek: we need a resolution to this before we enter CR 08:46:34 cwebber2: If it's possible to get a resolution before CR great, but the reason I used our own namespace is because I thought that was unlikely 08:46:56 ... I don't want to end up having.. it would be worse to me to be caught up trying to negotiate vocab sutff which seems to be really complex in w3c right now, than to have a separate context 08:47:31 ... I did write up a separate context file. AP in theory works if we can put this context file somewhere. I would like us ot put it somewhere in the meanwhile, and is it possible to put it out there and then remove it? 08:47:38 sandro: let's try to figure this out more through the day 08:47:47 cwebber2: I really don't want to be blocked on CR by this 08:48:07 sandro: from your perspective you can say you'll use your own context.. 08:48:17 tantek: anyone have a objection to doing it either way? 08:48:22 ... not preferences, objection 08:48:34 KjetilK has joined #social 08:49:12 rhiaro: If we are able to define our namespaces as writeable going forward that's awesome 08:49:21 arnaud: I agree with the sentiment, I just don't know how you make that happen practically 08:49:28 present+ 08:49:29 sandro: Something like the microformats process or the schema.org process 08:49:36 q? 08:49:44 ... Basically you need some benevolent dictator to manage a community that reviews and objects or doesn't object 08:49:54 ... Since this is more under w3c... over lunch we'll figure out some.. 08:50:02 tantek: is there an existing example of a community at w3c managing a ns that way 08:50:06 sandro: I don't think so 08:50:17 tantek: I'm hearing from arnaud a soft objection because we don't know if that could work 08:50:21 arnaud: I'm not objecting 08:50:36 ... Look at the example we saw earlier with LDP. Ther'es nobody really around to say yeah sure 08:50:42 rhiaro: I thought we had a bunch of LDP people in the room.. 08:51:02 cwebber2: Figuring out the extensibility of AS2 in the future is an orthoganal issue 08:51:03 tantek: right 08:51:13 ... The isssue we are considering now is shall we try to add AP terms to AS2 right now 08:51:21 ... vs shall we have a separate context for AP 08:51:36 sandro: Do we add it as a one time special case only we can do this, or do we add it as a first example of how to do an extension 08:51:39 ... I'd prefer the latter 08:51:47 tantek: but that's not required. THere are no objections 08:51:57 cwebber2: I have a preference for AS2 if it's possible 08:52:01 ... But an overriding preference to ship 08:52:41 ... We're going to try to figure this out before we enter CR. 08:52:47 tantek: We have to pick a date by which we decide which we're going with 08:52:53 ... If by then we haven't, you have a way forward 08:53:20 cwebber2: So the answer seems to be if we can find out we're going to put things in AS2 namespace by some date (CR date) 08:53:25 ... Shooting for Oct 11 for CR 08:53:36 tantek: no objection for the same date 08:53:53 cwebber2: So oct 11 if we can find out by then that we can put these terms in AS2 ns, we'll do that 08:54:02 ... If we find out then we can't, I'll put AP ns 08:54:31 PROPOSAL: We'll wait until October 11 to find out if we can put these in AS2 vocab. If we can do that, we will. If not, I'll use AP's own namespace. 08:54:39 s/wait/give 08:55:06 RESOLVED: We'll give until October 11 to find out if we can put these in AS2 vocab. If we can do that, we will. If not, I'll use AP's own namespace. 08:56:36 AnnBass has joined #social 08:56:42 rhiaro: Preference to add to AS2. Preference to make our namespaces extensible according to some sensible process (CG, W3C ns extensibility policy) in the future. 08:56:57 boris_anthony has joined #social 08:57:28 cwebber2: Preference is to add to AS2 08:57:30 sandro: same 08:57:36 tantek: anyone else want to indicate a preference on the record? 08:57:38 (as the first extesion!) 08:57:54 sandro: My preference is to add it to AS2 as an extension 08:58:06 csarven: with an extensibility mechanism? 08:58:07 sandro: yes 08:58:18 Present+ 08:58:21 cwebber2: the upside of this is that it puts pressure to resolve this within a certain period of time 08:58:32 ... Otherwise there's a consequence of AP getting its own namespace 08:59:43 tantek: action on sandro to define extensibiilty model that allows AP to add to AS2 08:59:51 cwebber2: All issues closed 09:00:12 tantek: We have discussion of prototypes, implementations, at risk features 09:00:28 ... Do you ahve changes to AP, including resolutions during last 2 days, that would benefit from publishing a new WD? 09:00:36 cwebber2: Someone read through it and helped me find editorial things 09:00:48 ... fixing authz vs authn 09:01:12 ... Something I'd like to do, this is pointed out to me with tremendous irony that I'm suggesting this, but I'd like to put the binary data stuff at risk, even though mediagoblin needs that 09:01:23 ... Becasuse I think it might be the one thing that I'm least comfindent that other implementations that are not mediagoblin might get in on time 09:01:27 ... I want it to happen 09:01:43 ... But if it doesn't, it should not throw the rest of AP under the bus, because it will still work without it 09:01:49 tantek: There are no objections in the room 09:01:59 arnaud: It is good practice to explain why something is at risk in the spec 09:02:17 cwebber2: I'll capture that 09:02:53 ... I notice that micropub has a much more fleshed out version of the media endpoint than we do. I took a look at ours again, and I'd like to clarify before I publish to new WD that section 09:03:02 ... I'll look at micropub and what's happening in mediagoblin currently 09:03:08 tantek: do you have an issue to track that? 09:03:11 cwebber2: I'll add that 09:03:19 aaronpk: if it' snot changing the behaviour you can change the text 09:03:27 cwebber2: I'm not sure, I need to check it's specified enough 09:03:31 tantek: if it might be normative change, file an issue 09:04:01 cwebber2: *files issues* 09:04:32 tantek: Given discussion about namespaces, and that breakout yesterday, I think it would be prudent to highlight that issue inline in the spec 09:04:47 ... where it talks about the context and the namespace 09:04:49 ... (editorial suggestion) 09:06:13 Present+ Benjamin_Young 09:06:46 newton has joined #social 09:07:07 tantek: We have next steps for you to publish a new revision 09:07:13 ... We can continue WD to CR discussion later 09:07:17 ... or in a future telecon 09:07:18 ... Move on to LDN 09:07:24 aaronpk: break? 09:07:26 ... coffee?? 09:07:32 ... *twitches* 09:07:47 various: it has started 09:08:07 tantek: coffee? 09:08:15 TOPIC: COFFEE 09:08:19 *break for 20 minutes* 09:08:23 1030 reconvene 09:10:13 csarven, I started a quickhack of LDN on the plane home last night: https://github.com/kjetilk/p5-rdf-linkeddata-notifications I had hoped to finish it on the plane, but the offline situation impeded the progress somewhat due to the lack of some documentation 09:11:18 now, I'm back to proposal writing, so I don't know when I'll more hacking time 09:14:32 dan has joined #social 09:16:27 jasnell has joined #social 09:27:36 KjetlK++ 09:27:36 kjetlk has 1 karma 09:30:38 AnnBass has joined #social 09:33:24 KjetilK: ++ Please PR to https://github.com/csarven/ldn/blob/master/implementations/readme.md 09:33:24 kjetilk has 1 karma 09:33:38 OK! 09:33:42 issue before break: https://github.com/w3c/activitystreams/issues/370 09:34:17 scribenick: aaronpk 09:35:09 TOPIC: LDN 09:35:18 csarven: KjetilK who was here yesterday built an LDN receiver on the plane yesterday 09:35:31 ... this is probably the more complicated part of the spec 09:35:46 rhiaro: the receiver has all the MUSTs, whereas the sender has a couple options 09:35:49 tantek: is it open source? 09:36:12 rhiaro: yes, here https://github.com/kjetilk/p5-rdf-linkeddata-notifications 09:36:38 ... one of my colleagues implemented LDN receiving and sending in a couple days last week in Python 09:36:43 it is just the discovery part still 09:37:32 Editor's Draft: https://linkedresearch.org/ldn/ 09:38:01 rhiaro: at the last meeting we said we'd publish a new WD and freeze this until TPAC 09:38:02 Changes since last WD: https://linkedresearch.org/ldn/#change-log-Changes-from-13-September-2016-WD-to-this-version 09:38:08 ... we had some feedback by email and Tim 09:38:19 ... the current ED has these changes which are largely editorial 09:38:34 tantek: these look great. would you like to publish this as a new WD today? 09:38:46 rhiaro: yes but let's do that at the end in case other things come up today 09:40:25 rhiaro: summary of LDN. there are three parts to LDN. one is a receiver, this is like a webmention receiver. it's an endpoint that accepts POST requests from senders. it also exposes the stuff it receives so that other things can read it 09:40:42 ... we have the capacity to add acces control so the receiver can filter what it accepts so it can do spam control for example 09:41:00 ... the activitypub use case is you'd expect your app to authenticate to read things from the endpoint 09:41:20 newton has joined #social 09:42:11 ... every existing LDP server can act as a receiver 09:42:44 sandro: can i try my soundbite version of this? 09:42:59 ... in my mind, this differs from webmention in two ways. this relies on authentication so i can just send the content rather than sending a link to the content 09:43:18 ... the second is it as the GET so you can see the content from the receiver, whereas webmention is a blind dropbox 09:43:58 rhiaro: it's not that it relies on authentication, it's that the verification process is up to the receiver. so you can have a publicly writable endpoint you can post to and do the verification by fetching, or you can do the verification by checking a signature, or whatever. so it handles any kind of notification payload 09:44:28 sandro: but if i didn't have any authentication, and if webmention had rel=mentions like ahs been brainstormed, then it seems almost equivalent to webmention 09:44:50 rhiaro: webmention requires you publish something at a URL whereas LDN does not 09:45:11 sandro: so you can send a notification from a browser, but you can't do that with webmention 09:46:15 bigbluehat: just to be clear. with webmention, you'd write to some write storage you have and then send that URL to the endpoint, and the endpoint would go fetch that from the URL. with LDN, either the sender can write to the inbox directly, or can still tell you about a URL. 09:46:27 rhiaro: yes and you can also include other information that the receiver can use to filter things out 09:47:36 cwebber2: to clarify the relationship between LDN and activitypub, AP says our delivery mechanism is the same. target->inbox is the same thing. 09:47:59 rhiaro: the mechanism is the same, but AP requires you use an activitystreams activity as the notification. LDN says you can use any payload. 09:48:25 tantek: okay. issues. 09:48:35 rhiaro: we closed a bunch of issues last week. there's a couple left open. 09:49:54 https://github.com/csarven/ldn/issues/32 09:50:24 rhiaro: we closed #4 with agreement from sandro who opened it 09:50:34 https://github.com/csarven/ldn/issues/4 09:50:38 rhiaro: issue 32 09:50:41 Now we discuss https://github.com/csarven/ldn/issues/32 09:50:50 ... yesterday we talked about a backoff strategy with webmention 09:51:02 ... webmention resolved to deal with this by adding to the user agent header 09:51:10 ... so that someone being attacked by webmention discovery can find out what's going on 09:51:24 ... but we have a bunch of JS considerations, where it's difficult or impossible to change the user agent in JS 09:51:36 ... so that's not feasible for us 09:52:00 tantek: the problem was raised about servers talking to servers, that was the real world thing happening 09:52:09 paulj has joined #social 09:52:17 ... but the scope of the actual problem was specifically servers talking to servers 09:52:28 kaorumaeda has joined #social 09:52:32 ... but don't make the error that discovery is the problem 09:52:47 sandro: i think the reason is one server can hammer another without a human having a clue. whereas if a browser is happening something the browser will start being slow 09:53:25 csarven: they're both valid, what we're addressing here is one way of getting to it from the client, but from implementation experience it doesn't seem necessary for arbitrary user agents 09:53:53 bigbluehat: in this case the webmention endpoint can control its user agent, but with this, anything can post to the inbox 09:54:07 tantek: i'm saying by doing that you're making an empirical error you're going beyond the bounds of the problem being solved 09:54:25 bigbluehat: with the annotation protocol we'd like a notification system to publish a notation in the browser and notify arbitrary servers 09:54:41 tantek: the problem being describde was one server hammering another server 09:54:59 rhiaro: the sender could be another server 09:55:12 csarven: we added text to rely on existing cache headers 09:55:51 here's a rawgit URL for the patch being discussed https://rawgit.com/csarven/ldn/da56c9afeecbd5d34b9b20ff663981aa4017fb74/index.html#retrying-discovery 09:56:01 aaronpk: yes this is the same conclusion webmention came to, which is URL-based throttling makes sense to respect cache headers, but that doesn't solve host-based throttling 09:57:38 sandro: we're all agreed about respecting cache headers, but we still haven't solved host-level throttling 09:57:56 csarven: how do we actually address the problem of getting senders to not hammer a host in the first place 09:58:07 ... what the receiver should do whether the sender is following that or not is a different issue 09:58:19 ... adding a requirement that helps the receiver solve its problem in our case is not a requirement we want to introduce 09:58:33 tantek: this entire scenario is about naive receivers 09:58:43 ... so any requirements we put in the spec does not solve this problem 09:59:09 sandro: if you are being hammered ,you look at the user agent to discover why you're being hammered, so then you want to find out what you can do to stop being hammered 09:59:36 tantek: i made a request of webmention. to add an informative paragraph that says if you are receiving webmention requests and you don't want to handle them, then blank 09:59:52 sandro: it seems like whatever the solution there we can do with LDN too 10:00:08 sandro: let's say i'm yahoo, and i'm getting all these LDN discovery requests and for some reason it's annoying me 10:00:27 where "blank" is left up to webmention editorial description 10:01:40 aaronpk: but how does the receiver know it's an LDN discovery request, that's what the user agent is for, and i haven't heard LDN say they want to recommend using the user agent 10:01:50 jungkees has joined #social 10:02:06 shepazu has joined #social 10:02:12 here's the WebMention issue comment from yesterday's meeting: https://github.com/w3c/webmention/issues/48#issuecomment-248865148 10:02:13 rhiaro: maybe the JS user agents will not be causing problems at this scale because of their nature so them being unable to set their user agent is okay 10:02:35 sandro: once the user agent tells you the request is an LDN request, then what do I do to stop that 10:02:46 rhiaro: an OPTIONS request with a retry-after 10:02:51 sandro: so are clients required to do an OPTIONS request then? 10:03:01 csarven, rhiaro, just noticed there is a broken link to pubsHubhubbub in ldn ED 10:05:02 sandro: we can make this a little fuzzy. we can say if you're being hammered, set this OPTIONS header. and then if you think you're hammering someone, say you should respect the OPTIONS header 10:05:18 csarven: what are the other specs doing in similar situations? not necessarily within this WG 10:05:27 rhiaro: does annotations do discovery? 10:05:30 bigbluehat: yes, via link headers 10:05:40 csarven: given the web architecture, is there a particular reason to mention this in the spec? 10:05:48 tantek: no, this was an empirically discovered problem 10:06:09 sandro: we could publish a WG note that talks about this problem in general and both LDN and webmention could link to it, which is about possible failure modes in discovery 10:06:29 bigbluehat: i would think this exists, because the RDF world scrapes things all the time so someone has to have dealt with this at scale before 10:06:35 tantek: i appreciate your optimism, do you want to find that? 10:06:47 csarven: i don't know if SWP would be the best place to mention this 10:06:56 sandro: i would think a specific note about backing off in discovery 10:07:08 Web Annotation Protocol's method of discovery uses Link headers/tags https://www.w3.org/TR/annotation-protocol/#discovery-of-annotation-containers 10:07:10 [Robert Sanderson] Web Annotation Protocol 10:08:08 tantek: sarven made a proposal, which is that SWP add a section about this 10:08:20 rhiaro: i wouldn't be opposed to putting it in SWP, and fine with separating it out if it's worth it 10:08:29 paulj has joined #social 10:09:08 PROPOSED: add a section to SWP describing the likely potential for server-to-server heavy load from discovery requests and approaches to mitigating 10:09:34 csarven: it's not necessarily server-to-server 10:09:36 q? 10:09:59 +1 although I think we'll want to move it to a separate document 10:10:23 tantek: the high risk situation right now is server-to-server, so i don't want to water it down with theoretical other situations 10:10:39 q? 10:10:43 ack sandro 10:11:26 cwebber2: i've been trying to figure out if there's any context in which this applies to activitypub. but activitypub is much more specific about recipients. but bigbluehat hit on a good example of where this has heppened, which is a client to server scenario where someone is trying to extract their whole history 10:11:53 bigbluehat: the reason i believe this document already exists is before this there was atom and RSS which use the same discovery mechanism so they were doing the same thing about discovering feeds 10:12:04 tantek: i agree, but there should have been evidence of this happening before 10:13:00 bigbluehat: david, as standin for all of RDF at the moment... 10:13:19 ... the case we're dealing with is doing discovery... 10:13:30 ... are there known ways to encourage a backoff strategy from the client 10:13:39 david: sure, but that's not an RDF issue, that's an HTTP issue 10:13:51 ... sure there are HTTP status codes, like "slow down your requests" 10:14:09 https://developer.mozilla.org/en-US/docs/Web/HTTP/Headers/Retry-After 10:14:18 https://http.cat/429 10:14:22 bigbluehat: i'm trying to suss out if whether this is as ancient as i think it is 10:14:27 .. .and what others have done in this situation 10:14:46 ... i don't think it's unique here and i'm looking for prior art here 10:15:11 david: i'm struggling to figure out why a 503 return code isn't appropriate 10:15:19 tantek: because the server is not too busy... 10:15:28 ... it's closer to a 429 "too many requests" 10:15:31 https://http.cat/429 10:15:32 david: fair enough 10:15:37 sandro: "for use in rate limiting schemes" 10:16:20 newton has joined #social 10:16:53 tantek: can we capture this as an action instead of continuing to discuss it here? 10:17:09 bigbluehat: my concern is we were talking about doing things like sending a user agent, rather than using 429 or referencing prior art 10:17:25 also https://http.cat/503 10:17:33 Arnaud: do we even need to say anything about this? HTTP is there, you can use those codes, we can just say this is not really our problem. it's a quality of the implementation problem 10:17:45 stack overflow about Retry-After http://stackoverflow.com/questions/3764075/retry-after-http-response-header-does-it-affect-anything/23863426#23863426 10:17:57 david: the old guys in the room would say you're right. the problem is the young guys are saying we need to give them context. 10:18:12 ... they need some indications fro mthe spec that point out the potential warnings 10:19:03 tantek: so the proposal is to start with a note or section in the SWP document and that it may be worth spinning out into a new note 10:19:34 bigbluehat: i'm happy to help with that 10:20:11 david: the bigger issue in the RDF world is the ease at which it is to write a SPARQL query that will hammer the server. but in the world of query languages there isn't a way to solve that. 10:20:18 bigbluehat: is there a client need to point out always be carefule 10:21:03 david: maybe there is. an in-javascript implementation, the idea is you'd do federated queries but all the joins are done in browser, so all the subqueries going off are much less likely to be complex. 10:21:19 tantek: okay we have a proposal on the table, i'm not hearing any objections 10:21:42 bigbluehat: this was about should we point out to clients while they're being built that they should be careful 10:21:57 I'll note that this is why when we get people telling MediaGoblin that we should abandon http and use $SOME_P2P_SYSTEM, while they're entirely wrong, they're also kind of right :) 10:22:15