Hacker News Reader

Comment by tedunangst | Hacker News Reader

Just from a technical perspective, what is Apollo doing that requires more than 100 queries per minute?

(Okay, so apparently client ID isn't user ID, but app ID. Explains a lot.)

Apollo uses an average of 324 requests per user per day, according to the developer's farewell announcement posted yesterday [0].

I hadn't heard anything about the "Free Data API" limit of 100 qpm before this either, so I assume the issues is that the free API is filtered and does not include any "mature content".

> Effective July 5, 2023, we will limit access to mature content via our Data API as part of an ongoing effort to provide guardrails to how explicit content and communities on Reddit are discovered and viewed. This change will not impact any moderator bots or extensions. In our conversations with moderators and developers, we heard two areas of feedback we plan to address.

0: https://old.reddit.com/r/apolloapp/comments/144f6xm/apollo_w...

Let's say Apollo can cut down to 100 requests per user per day. at 100 qpm, assuming equal distribution, they now can have one user for each minute in a day. Somehow I don't think having 1440 users is going to cut it.

I think the free limit is supposed to be per user, since he says that "over 90% of apps" make rates at requests within the free limit.

... if they meant per user, they'd say per user. They don't, but explicitly say that it is per client id (=app).

> since he says that "over 90% of apps" make rates at requests within the free limit.

>90% of apps are people that played around with the API a bit for fun or run a small hobby-bot or something. Of course those stay below the limit (and for many of those, 100 qpm is indeed quite generous). It's also entirely irrelevant to the small number of apps that non-dev end users use.

> ... if they meant per user, they'd say per user. They don't, but explicitly say that it is per client id (=app).

You may be right, the reddit post has been edited multiple times. I assumed when it referenced OAuth that it was per end user, but it does seem that "client_id" is global to the application. I wish I could find any place else that lays out the details on the limit besides just this one reddit post...

Currently, it says the limit is "100 queries per minute per OAuth client id if you are using OAuth authentication"

And it's pretty shitty if he's including every single toy project within the "90% of apps".

> Important note: currently, our rate limit response headers indicate counts by client id/user id combination. These headers will update to reflect this new policy based on client id only on July 1.

https://reddit.com/r/redditdev/comments/13wsiks/api_update_e...

Let's say Apollo has 1 million users (I think the actual # is significantly higher). Reddit doesn't have a push API for notifications (like if someone messages you or replies to your comment), so you have to poll. Just checking those inboxes once per week gets you to 100 queries per minute. In reality, most people want inbox notifications more frequently than that.

And in the meantime, users are also upvoting posts, commenting, scrolling, and doing other things that cut into the API quota

As far as I understand, its 100 qpm per oauth user. Each logged-in user will have 100 qpm, so they can actually query inbox once per minute and still be left with 99 queries for that minute alone. I thought the problem was that, if you exceed this limit even for some users (e.g. particularly active ones) then you are no longer eligible for the free-tier and get kicked out of the club and pay price even for the free tier.

It's 100 qpm per oauth client ID, which is tied to the developer's account in the API portal (at least that's how it worked when I used the API in the past). I suppose someone could make an app that lets users provide their own API keys, but most people aren't going to go through the hassle of setting up a dev account and generating API keys. That might also be a TOS violation?

No, it's not. It's 100 qpm per client id, i.e. app.

That was my initial question too, but note it's per oauth application. It could be that per-user it could get under 100, but in aggregate the only way they'd get there is if only one person used Apollo at a time.

I wonder why didn't Apollo just have each user register a new OAuth app at https://old.reddit.com/prefs/apps, the details of which would then be used per-user. It could even be done automatically upon entering one's credentials into the Apollo app. Wouldn't this have solved the problem?

if you assume that reddit won't just ban or at least take API access away from any user doing that, sure. I would not assume that.

Probably because the Apollo dev was sick of this shit and just wants to be done, especially after being slandered by the CEO of the company

Background reading you may be interested in:

https://old.reddit.com/r/apolloapp/comments/13ws4w3/had_a_ca...

https://news.ycombinator.com/item?id=36256167

As far as I understand, that's 100 shared queries across all users.

Why can't Apollo use per-user OAuth and not Apollo's own API keys?

https://www.reddit.com/r/redditdev/comments/13wsiks/api_upda...

Wouldn't that require each user to go in the developer settings for their account and set up their own OAuth app integration? That's probably not feasible for end-users and I would imagine breaks the spirit of the API terms if not the letter of them.

"100 queries per minute per OAuth client id". An app like Apollo uses a single client id, AFAIK.

No idea. Also, I suspect Apollo made a judgment error in looking at their usage stats through the lens of the mean, rather than their median and 99% percentile, usage.

They likely have power users greatly pulling that mean up that, in all cases, should probably be capped.

EDIT: Ooof, people dislike this thought. Has the Apollo dev spoken out about their approach? I'll be the first to admit I haven't seen reddit's entire debacle, but thought I was reasonably following the play by play.

re: your edit:

See: https://news.ycombinator.com/item?id=36261736

Thanks!

The reddit post is what I was referencing by looking at mean vs. median/99%

Reply on news.ycombinator.com