Turning paginated requests into an Observable stream with RxJs

问题

I have a service which returns data in pages. The response to one page contains details on how to query for the next page.

My approach is to return the response data and then immediately concat a deferred call to the same observable sequence if there are more pages available.

function getPageFromServer(index) {
  // return dummy data for testcase
  return {nextpage:index+1, data:[1,2,3]};
}

function getPagedItems(index) {
  return Observable.return(getPageFromServer(index))
    .flatMap(function(response) {
      if (response.nextpage !== null) {
        return Observable.fromArray(response.data)
          .concat(Observable.defer(function() {return getPagedItems(response.nextpage);}));
      }

      return Observable.fromArray(response.data);
    });
}

getPagedItems(0).subscribe(
  function(item) {
    console.log(new Date(), item);
  },
  function(error) {
    console.log(error);
  }
)

This must be the wrong approach, because within 2 seconds you get:

      throw e;
            ^
RangeError: Maximum call stack size exceeded
    at CompositeDisposablePrototype.dispose (/Users/me/node_modules/rx/dist/rx.all.js:654:51)

What is the correct approach to pagination?

回答1:

EDIT Ah! I see the problem you're facing. A bit of tail call optimization should fix you up:

function mockGetPageAjaxCall(index) {
  // return dummy data for testcase
  return Promise.resolve({nextpage:index+1, data:[1,2,3]});
}

function getPageFromServer(index) {
  return Observable.create(function(obs) {
    mockGetPageAjaxCall(index).then(function(page) {
      obs.onNext(page);
    }).catch(function(err) {
      obs.onError(err)
    }).finally(function() {
      obs.onCompleted();
    });
  });
}

function getPagedItems(index) {
    return Observable.create(function(obs) {
        // create a delegate to do the work
        var disposable = new SerialDisposable();
        var recur = function(index) {
            disposable.setDisposable(getPageFromServer(index).retry().subscribe(function(page) {
                obs.onNext(page.items);
                if(page.nextpage === null) {
                  obs.onCompleted();   
                }

                // call the delegate recursively
                recur(page.nextpage);
            }));
        };

        // call the delegate to start it
        recur(0);

        return disposable;
    });
}

getPagedItems(0).subscribe(
  function(item) {
    console.log(new Date(), item);
  },
  function(error) {
    console.log(error);
  }
)

回答2:

Looking at the OP code, it is really the right method. Just need to make your mock service asynchronous to simulate real conditions. Here is a solution that avoids the stack exhaustion (I also made getPageFromServer actually return a cold observable instead of requiring the caller to wrap it).

Note that if you really expect your service requests to complete synchronously in the real application and thus you need to ensure your code does not exhaust the stack when this happens, just have getPagedItems() invoke the currentThread scheduler. The currentThread scheduler schedules tasks using a trampoline to prevent recursive calls (and stack exhaustion). See the commented out line at the end of getPagedItems

function getPageFromServer(index) {
    // return dummy data asynchronously for testcase
    // use timeout scheduler to force the result to be asynchronous like
    // it would be for a real service request
    return Rx.Observable.return({nextpage: index + 1, data: [1,2,3]}, Rx.Scheduler.timeout);

    // for real request, if you are using jQuery, just use rxjs-jquery and return:
    //return Rx.Observable.defer(function () { return $.ajaxAsObservable(...); });
}

function getPagedItems(index) {
    var result = getPageFromServer(index)
        .retry(3) // retry the call if it fails
        .flatMap(function (response) {
            var result = Rx.Observable.fromArray(response.data);
            if (response.nextpage !== null) {
                result = result.concat(getPagedItems(response.nextpage));
            }
            return result;
        });

    // If you think you will really satisfy requests synchronously, then instead
    // of using the Rx.Scheduler.timeout in getPageFromServer(), you can
    // use the currentThreadScheduler here to prevent the stack exhaustion...

    // return result.observeOn(Rx.Scheduler.currentThread) 
    return result;
}

回答3:

Here is a more concise & IMHO clean answer without any recursion. It's using ogen(~46 loc) to transform any generator into an observable.

It has a custom built next function that will emit data anytime your function yield something.

nb: The original article is worth reading

function getPagedItems({offset=0, limit=4}) {
    paginatedQueryGenerator = function*(someParams offset, limit) {
        let hasMore = true
        while(hasMore) {
            const results = yield YOUR_PROMISE_BASED_REQUEST(someParams, limit, offset)
            hasMore = results && results.nextpage !== null 
            offset += limit
        }
    }

    return ogen(paginatedQueryGenerator)(someParams, offset, limit)
} 

回答4:

Another one solution is to use retryWhen

getAllData() {
    let page = 0;
    let result = [];

    const getOnePage = () => {
        return of(0).pipe(mergeMap(() => getPaginatedData(page++)));
    };

    return getOnePage()
        .pipe(
            map(items => {
                result = result.concat(items);
                if (templates.length === PAGE_SIZE) {
                    throw 'get next page';
                }
            }),
            retryWhen(e => e.pipe(
                takeWhile(er => er === 'get next page'))
            ),
            map(e => result)
        )
        .subscribe(items => {
            console.log('here is all data', items);
        });

}

来源：https://stackoverflow.com/questions/27514310/turning-paginated-requests-into-an-observable-stream-with-rxjs