GitLab Plan Stage, 2 Jul 2019

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Chat about Akismet and reCAPTCHA on GitLab.com

Description

Related issues:
1. https://gitlab.com/gitlab-org/gitlab-ce/issues/45684
2. https://gitlab.com/gitlab-org/gitlab-ce/issues/64021
3. https://gitlab.com/gitlab-org/gitlab-ce/issues/64023

Agenda: https://docs.google.com/document/d/1Elb7JJpZ1PzsikbJ1gdm2FImX_I7USVCPjiGpjZgPqQ/edit

A

Better have a recording than deciding later that we want to have it yep.

A

So yeah in the dock, I tried first to come up with the things that we're trying to solve in that issue, because I feel like we.

A

Switch direction which we are, we changed the problem that we're trying to I.

B

Lost we need to do through some tunnels, yeah.

B

Okay, it's gonna be an awesome recording when you yeah, we got you back, no see movements.

C

B

B

When he can hear us I.

A

Can hear you okay,.

B

So we lost you from basically a start of when you were describing the current problems yeah, you.

A

Know I try to do it over, but we got this time so we trying to solve couple different problems, an issue now and I. Try to listen here. Also I try to list the strategies that we have discussed in the issue already with small additions and ordered them by effort that I think they need to be implemented. None of so there's, basically one that I think can be done within a single recycle.

A

All the others would probably spend multiple and yeah I will post the result of this discussion in the issue for future reference and for the problems. I think it started off as a licensing problem, but then became the issue for the X problem that crew capture has and through our discussion, I think we realized that we don't. We can't even measure that reCAPTCHA is a good solution for the problem that it tries to solve.

A

B

So, just to be clear, there is a little bit of a completion here between two differents spam mitigation technologies. We use and we have Akismet, which I'm not sure, if I'm pronouncing correctly but whatever, which we use to actually check the text of an issue ever for potentially spam content, and then there is recapture which we use to both verify signups. And if your contents that you submit is detected as spam by hizmet's, we allowed used to solver capture to like manually like yes, only that step like yeah I'm, not just rejecting it. So.

A

In my intention was to focus on the plant side of the recapture, so I'm not sure if our goal should be to get rid of hookups in gitlab in general, even though that was original scope of the issue, but it seems like manage, has a separate issue for coming up versus with a strategy to get rid of capture for registering. So so my intention would be to focus on the recapture path for preventing issue and comment spam. Yes,.

B

So on the issue and comment, spam Park just sort of be clear there, so we have the two components we have Akismet and we have the recapture so from an annoyance perspective, the annoyances with the recapture, but it's not caused by the recapture. It's caused by the use of a kismet for checking the issue contents for spam, and then our fallback is the recapture from a licensing perspective. Obviously that doesn't.

B

You know that doesn't matter like the problem is recapture itself and there are other issues that people have other valid issues that people have recaptured that arm to do about that. I just want to be clear that if we remove recapture we're either saying we're not doing spam checking on issues which means we're actually disabling Akismet, not recapture or we're saying that when we detects that something is spam using a kismet we're just going to reject it with no way of you know like moving forward, which I think is probably worse.

C

Do we have any kind of metrics on how many times it's actually failing, I could smell Equis yeah.

B

I know no idea how you specially announce that it really probably find a new tube or something, but we don't so if we wanted to make that the first thing but I just want to I just want to try and untangle lose to a certain amount, because I think, because recapture is the visible part and a kismet kismet.

B

Whatever is the like, hidden part that we talked about recapture when we sometimes mean Akismet and sometimes meaning the capture but I, think metrics would be a good idea so for step, one for instance, when he we could just immediately disable a kismet or get lab calm, and that would have the effects of not some checking issues etc, which means that you would never see a recapture either because that only shows up when they kiss may protect something as spam. So that's literally like zero development effort to do that, I guess.

B

The problem is that without those metrics, which is what you said, no indication of increased spam no way to react there.

B

A

I mean if we, if we disable a kiss meant in general, my intention was to keep Akismet enabled but disable direction to that. So keep track of spam behind the scenes, but have no way to override that which would be some development effort. But if we disable kismet in general, we, our reaction could be to just enable again like if he because I think I excuse me there's a happen. Setting fires are know. So if we disable it and realize that same things happen, we can just enable it which would be very low effort.

A

So we would have a way to react, even though we may not realize that we need to react. Yeah.

B

So and again just to detangle these issues, because the issue that we're talking about is about recaptured, non-being, open source and yeah being distributed, and the problem that people are mostly complaining about is that they see a CAPTCHA at all because of the spam check.

A

So do you think we should put these problems again in two separate issues, one in being finding a different capture solution that would then work with IKEA make you smile together and being.

C

A

Making a case made less annoying.

A

Because if we have realized that captures in general another good solution, you XY z--, we probably don't need to spend effort on finding an eternity of capture solution.

B

So I don't know my I.

B

Think we should look into replacing recapture anyway, because even if we remove or change the spam checks for issues like when I say we I mean as an organization, because we also use it sign ups. So that's one thing, I think for the spam detection. We could definitely add.

B

We can make it smarter right, like you know, we could say if this user hosts, like you know, submitted whatever like comments like you know that says whatever history, then it's probably not spam from them, because I think.

B

This is going to be like either somebody who is just submitting spam or somebody who is just a regular user in general, that.

A

Would be replacing Akismet with our own spam detection, all.

B

Menteng it saying like okay: well, we will go to a kismat unless the user is like, you know, they don't know, has contributed to X amount of times or we.

A

Are rate limiting what we one of the options yeah.

B

The rate limiting is another option that you mentioned: I mean.

A

Would be one of the options to augment it like if the, for example, for the task list stuff, we definitely want equipment not to consider many different checkbox clickings to be spamming.

B

Yes, so and there's another part of that which is actually a completely separate faster, which just occurred to me, is because we have like Cassie and Brett implemented this separately Koufos for checking a checkbox. We should never spam check that because the issue isn't going to get any more or less family when I change, one state changes so I think that's one.

A

Which, which is medium when just a new, which is a new possibility right because of the reimplementation of the checkbox logic? Yes,.

C

A

Now have the option not to Spencer, which we not have prior to that yeah.

B

And I feel, like that's the most irritate, there's the most irritating case for me for sure it's like you know, I'm just trying to check a box here like I'm I'm, not I, think a link I'm, not removing a leg like this clearly convey spam.

B

So actually, maybe that's, maybe that's the smallest iteration. That would give the most like immediate benefit to us a killer and then.

A

Should probably them be you separate issue? Yes, I agree, oh cool.

B

B

The capture alternatives is kind of a separate question, so I think I think we have a bunch of conflation of things and we should try and sort of step back. So the way I would like to consider.

B

Moving forward is.

B

There's an internal customer thing for security.

B

There any understand what about issues about, but it's confidential, so I won't mention the issue so I think if we remove spam detection for a task list, that's a huge potential s changes. That's a huge win like it's. It's pretty simple to do. I would imagine, and it's a big win from the user experience perspective, because this is I. Don't know about you, but, like I checked a lot more tasks than I do actually like write descriptions in issues the.

B

General, what do we do about the license of this capture? I? Think we could hunt to horse I was kind of split between Quan and manage, because it's used for two different things.

B

We could also create an issue for better spam detection in general, but I would I would basically like to.

B

How can I put this? This is a specific issue about the license of recapture I. Think is important, but this you missed, really get conflated with a lot of other things. So what I would like to do is over there to redirect conversation, so spam checks on check, corpses, you're, going to create a new issue for that spam checks not being smart enough in general.

B

We can create a new issue for that we can say, make the spam detection smarter, and we can point in that issue that actually, this isn't to recapture the stupid, okay, Smith and then for the capture license I think that's mostly a front end and question I I'm not really familiar with what the alternatives are out there, but basically the intent like a product perspective is to say, like we have this content, that was to change the spam.

B

We want some way if we use our being able to validate and say like no, this wasn't spam, I'm a legitimate user whatever and that's what we've used so far. So that's the the intent and I think if we can find a solution that does that, probably not gonna say no. Like is already my- is.

C

There a configuration on recapture at all currently or is that just it's built in we don't have any way of just turning off recapture in issues in Congress. Do we.

B

Know so, like I said earlier, what that would mean is like either we disable. So if we added that, if we added either the teacher to say, but we can't you're an issue in issues, for instance, what we will be doing is saying: well, if it detects a spam by it, gives me you can't you can't proceed, we can disable a kismet independently, which would have basically the same effect, but is not strictly what you asked.

B

C

We can still have agus MIT enabled and then just not do anything if it detects spam right. Oh that's.

B

True, so we could just have it in like monitoring mode yeah yeah. That would be a good idea, so we do have some.

B

So there is this spam logs table which I'm not really sure how it works, but we record things in it. So it's.

A

What what would be the consequence, though, if if we enable, if we keep encasement enabled, but we but disable recapture, then people being recognized as spammy, even though they have not spammed, would then need to create a support ticket to be unblocked. Yes,.

B

An alternative which I think is basically what Donnelly was saying so I might just be repeating what you just said. It's like we could have a kismet enabled, but not actually do anything, just only log the results, and then we already have these spam logs visible to admins and we could say well, it's becomes a reactive thing from an instance admin spectively. You can go in to look at the things that I've been detected to spam and see like okay. Is there anything I need to take action on here or is there anything I.

A

Don't so we would treat it as if a user had successfully soft click capture.

B

We would treat yeah kind of we would we would basically, we will basically make it reactive, so we would say like we're going to let you submit this, but we detected it as spam. So it's going with restoring that information somewhere and it's visible, to instance, administrators, and they can like, go and take a look later on and consider what they gonna block you for abuse or whatever yeah.

C

So we move the burden from users where it's currently at to admins. Essentially, yes,.

B

And we would definitely need to talk to the abuse team about this, but I think there could be another like separate issue like where we say like change,.

B

Change like spam burden from user to administrator or something like that, and we.

A

Actually make that feature flag. We could enable it as a first iteration only for good rock group right Oh, at least only on canary.

C

Can't do feature flags for just canary yeah. We could do it for four, yes, gitlab or group.

C

We could do that.

C

A

So so we would increase feature flag is to say basically disabling Nouri capture and like sort of returning true, this is supposed to go through and then in enable that feature flag. Forked lab off as a first iteration and see what happens and they.

B

A

A kismet blocks afterwards, yeah.

B

So I'm just added mark who's sitting next to me, but has also dealt with some of the spam stuff in the past, so mark I'm. You know you probably weird for you to hear this.

B

Did you just talk us through how so what we're discussing is the idea that, like instead of asking people to solve the tenth row and their content was detected as spam, we would just allow it through and then it would still go into the spam locks in the admin interface and then we'd. Let administrators still like deal with that. Normally from there could use describe how that looks from an administrators point of view and whether you think that makes sense.

D

Yeah, it's just a big list right now, so.

C

D

Will list the user and the and the contents has been reported, but I think that would be fine if you just load it through and just let's pin that exactly the same way yeah. This is the abuse team who currently manage that list.

B

D

I think also the github.com.

C

D

Team because they.

D

Get tickets from Zendesk, where someone like.

D

B

Okay, so I think those are two violent paths forward. We have them which we can do independently, but both of them give us like some kind of win. One is to disable the spam check when you're checking a checkbox because I need it, and then the other one is create a new issue for always allowing content to take to to spam through, but keeping the existing admin interface and working with the abuse team and the support team to make sure this is okay and putting that behind the feature flag.

B

So we can actually test it and then, if we do both of those potentially we get. The whole question of who owns recapture off our plane and back onto fully managed is great because it's like no longer are we using reCAPTCHA for any some related purposes, and then we don't have to actually touch this issue at all. So maybe we just would move this issue to the backlog and create those new to new issues for those on 12.1 or 12.2 and go forward with those. How does that sound? Winning.

A

B

Are you okay to create those issues? I will.

A

Create them also, potentially a third issue to leverage a kismet should that be necessary. Oh.

B

Yes, to enhance, like you know, the make it smarter what we're doing with a kiss make you mean yeah.

A

Because if we, if we switch, that switch that feature flag and realize now we're getting a whole lot of spam because of disabling recapture as as a prevention, then we would still be without in solution. For that. Yes,.

B

And the other thing is that we know that if we change that feature 5 right now, we would see a bunch of spam in the spam logs. That would be irrelevant because it would just be people checking boxes because, like we get the capture, so we know that a PS mr. texting that a spam- and we don't want that to happen so yeah. We need three new issues. We need one to make. This fan check smarter.

B

We need one for the no spam check contacts list update and we need another one to add a feature flag to always allow and content tech to spam to be submitted and then deal with it reactively.

B

So, oh yes, like you know, when you're you have like some discussion of forums that are pre moderated site or mailing lists or whatever that are pre moderated where you have to get like the message approved before it goes through and then other ones where it can be removed afterwards. But not you can always submit the message.

A

B

A

Make that an ordered list, because we want to have no sanitation, fotopolous first I guess this is annoying.

B

A

I wanted to have this unordered, but this order, oh.

B

Right: okay, well,.

A

Yeah, hey awesome.

D

A

Change it in with first second, so we want no spam detection for Hustler's first to not pollute the lock. Then we want to have the feature flag for disabling recapture.

B

What is an allowing spam through, because yeah I can just type.

B

A

See issue has potential, did we then split into multiple, smaller ones, depending yeah on the discussion yeah.

B

I think I think that one will probably split up, because at the moment we don't. We just finished discussing like what we could do there. We don't know what the right option is, but I think the first two might get us to a point where we don't care about the third one, so much I think so they're in the right order. As you have the map and.

A

Should should we realize from be that we indeed need recapture? Then we probably need another issue to replace recapture with another cap capture library.

A

B

A

See, but even if we make a Christmas martyr, we would still include recapture in the front end, so we would still have the licensing issue.

A

Yeah yeah, so we keep that existing issue limited to the icing problem. Yes, a fear, it is not at this discussion. It's not sorry.

B

A

A fear, the current discussion in that issue is not limited to know.

B

And I get why it's not because, like I said at the start, the thing is when your issue, it's a Texas time. What you see it's a recapture. You don't see anything about the hizmet, it's like for a user perspective. That's that's why you get there. There is a.

B

Internal customer there and my issue is confidential, so I mention it here, but that's being the internal customer issues being taken by manage, so it probably also makes managers life easier. If we can remove the plan element of this, but I think we can also possibly say that that other issue isn't blocked because.

B

So we can just like unblock that other issue, basically and say that I think I think that's fair enough room, because whatever we would replace it with our front end would be kind of a drop-in replacement type thing. Ideally, right like it wouldn't be like a rewrite everything it would be. We replace recapture with this solution that we've done all that we've got somewhere else. That has the other license a license that we can distribute and.

A

For a and B you want them to put them in 12.

A

B

Could put them in 12.1, certainly a because this issue is already in 12.1 and then we kind of said it was blocked. So I don't see a problem and pretty I don't know Donald's nothing. So yeah I don't see a problem in putting at least one issue, man, because it's just a straight swap basically.

A

And then so, I'll I said the most.

A

And this to 12-1.

B

Yeah- and we then need to move this issue out this milestone and.

A

This is probably I mean this will be any in a way o'clock until it's further split or we come up with a solution today, yeah.

B

And B&C need to be discussed with the abuse team and comment support team.

B

Okay, anything else we wanted to talk about I, think I feel like we got somewhere at least yeah.

A

I wrote expecially, imagine some use. You awesome.

B

I'll also post this video on YouTube thanks very much. Everybody have a great day, yeah.

A

You do thanks. Everyone.