Design for getting Twitter friends list for large user base and managing rate limiting

Assume there's a mobile app and a server.

I have question about rate limiting and hoping someone can give some advice on a design as I'm banging my head on how to navigate around rate limit. There must be something I"m missing because the 150 unauthenticated rate limit per IP per hour is extremely low.

Imagine the scenario I want to build is the following (simplified into a trivial example for this discusion). Assume user is signed into Twitter for this entire discussion to remove discussion about oAuth.

Mobile talks to our service to show users twitter friends list. Every time the mobile app is loaded, it will show the entire friends list, and highlighting the new friends that were added within the last 2 days.

That's it. But the trick is that I want to ensure that the friends list is always up to date in the client, which means our server has to have the most recent up to date friends list.

Periodically, I want my server to automatically scan the Twitter friends list for every user of my app to see if new friends have been added.

Our initial design was getting our server to do all the work with this flow:

1. New User signs in on client, gives access token to server
2. Server makes call to Twitter REST APIs to get initial friends lists
3. Server stores the Twitter Friends IDs and shows responds to the client with that list.

Periodically (e.g. every 48 hours), server checks Twitter REST APIs for friends list for each user and compares it to our cached Twitter friends list we have for them to see who is new and to highlight in the mobile app.

The good thing about this is that all the interaction with twitter to get friends list, compare and peridiocally refresh is on the server. Mobile client just makes a single call to my server and gets friends list.

The problem with this design is that it will work for a single user, but since the rate limit is 150 per hour on un-authenticated calls, I w开发者_Python百科ill hit my limit as soon as 151 users user my service (which has a fixed IP).

The only solution I can see is to have the client do the work for each user, then send me the friends list which my server caches. This takes care of Step #2 above. However, for Step #4, I'd have to build something into the client to auto refresh twitter friends and send back to the server.

This is super clumsy to have the client involved at all in this Twitter friends list operation.

At first I thought I was crazy and the public unauthenticated APIs like getting friends lists wouldn't be subject to rate limiting. However, according to their docs, it is.

Am I missing something obvious or is the only way to solve this is to put heavy logic into the client?

With whitelisting gone for those that aren't grandfathered or Twitter business partners, I don't think you have any alternative but to have your mobile app do the Twitter API calls from the handset.

Having the handset call Twitter isn't a bad thing by any means. Pretty much every Twitter client in the world does it. One benefit will be that the user will be authenticated to Twitter, and thus her full 350 calls per hour will be available to you. Keep in mind, however, that you should minimize your calls since the user may have other Twitter-aware applications installed on her handset eating into your call allotment, and vice versa.

Now to the solution. The way I would implement your use case would be to first fetch the complete list of friends for your user by calling the friends/ids method.

http://api.twitter.com/1/friends/ids.json?screen_name=yourUsersName

The above call will return the most recent 5,000 friend IDs, in order followed, for @yourUsersName. If you want to fetch more friend IDs than the first 5,000, you'll need to specify the cursor parameter to initiate paging.

Next, I would check the latest list of friends we just fetched against the list on the handset, syncing them by removing any IDs that are no longer present, while adding any that are new.

If we only need the friend IDs, then we're done at a cost of one API call per 5,000 friend IDs. If, however, we need to get user info for these new friends as well, then I would call users/lookup and pass in the list of all new users that we discovered while syncing friend IDs. You can request up to 100 user objects at a time.

http://api.twitter.com/1/users/lookup.json?user_id=123123,5235235,456243,4534563

You user must be authenticated in order to make the above request, but the call can fetch any Twitter user profiles you wish -- not just those that are friends of the authenticated user.

So, let's say for example that a user has 2,500 friends and has never used your app before. In that case, she would burn one call to fetch all of the friend IDs, and 25 calls for her friends' information. That's not too bad to get the app populated with data.

Subsequent calls should be more streamlined with probably only two calls burned (one for the IDs, and one to get the new friends).

Finally, once the data has been updated on the handset, the deltas for the IDs and user data can be gathered up and pushed to your server.

It may even be possible that your server application won't even have to interface with Twitter at all, and that should alleviate the 150 user limit you are encountering.

Some final notes:

1. Be sure to note in your app's privacy policy that you sync your user's friend list with your server.
2. I recommend specifying JSON as the return format for all Twitter API calls. It is a much more lightweight document format than XML, and you will typically transfer only about 1/3 to 1/2 as much data over the wire.
3. Pick a Twitter framework appropriate for your mobile device and your programming language. Twitter access is a commodity these days, and there's little to no reason to reinvent how to access the Twitter API.

I answered a similar question about an approach for efficiently fetching followers here.

Since you are making request on behalf of users you should make those requests be authenticated as those users. Then requests will count against each users own pool of 350 requests/hour.