Is it correct to convert a CompletableFuture<Stream<T>> to a Publisher<T>?

Question

To allow multiple iterations on the resulting stream from a CompletableFuture<Stream<String>> I am considering one of the following approaches:

1. Convert the resulting future to CompletableFuture<List<String>> through: teams.thenApply(st -> st.collect(toList()))

2. Convert the resulting future to Flux<String> with cache: Flux.fromStream(teams::join).cache();

Flux<T> is the implementation of Publisher<T> in project reactor.

Use case:

I would like to get a sequence with the premier league teams names (eg Stream<String> ) from a data source which provides a League object with a Standing[] (based on football-data RESTful API, eg http://api.football-data.org/v1/soccerseasons/445/leagueTable ). Using AsyncHttpClient and Gson we have:

CompletableFuture<Stream<String>> teams = asyncHttpClient
    .prepareGet("http://api.football-data.org/v1/soccerseasons/445/leagueTable")
    .execute()
    .toCompletableFuture()
    .thenApply(Response::getResponseBody)
    .thenApply(body -> gson.fromJson(body, League.class));
    .thenApply(l -> stream(l.standings).map(s -> s.teamName));

To re-use the resulting stream I have two options:

1. CompletableFuture<List<String>> res = teams.thenApply(st -> st.collect(toList()))

2. Flux<String> res = Flux.fromStream(teams::join).cache()

Flux<T> is less verbose and provides all that I need. Yet, is it correct to use it in this scenario?

Or should I use CompletableFuture<List<String>> instead? Or is there any other better alternative?

UPDATED with some thoughts (2018-03-16) :

CompletableFuture<List<String>> :

• [PROS] The List<String> will be collected in a continuation and when we need to proceed with the result of the future, maybe it is already completed.
• [CONS] Declaration verbosity.
• [CONS] If we just want to use it once, then we did not need to collect those items in a List<T> .

Flux<String> :

• [PROS] Declaration conciseness
• [PROS] If we just want to use it once, then we can omit .cache() and forward it to the next layer, which can take advantage of the reactive API, eg web flux reactive controller, eg @GetMapping(produces =MediaType.TEXT_EVENT_STREAM) public Flux<String> getTeams() {…}
• [CONS] If we want to reuse that Flux<T> we have to wrap it in a cacheable Flux<T> ( ….cache() ) which in turn will add overhead on the first traversal, because it has to store the resulting items in an internal cache.

Answer 1

    CompletableFuture<Stream<String>> teams = ...;
    Flux<String> teamsFlux = Mono.fromFuture(teams).flatMapMany(stream -> Flux.fromStream(stream));

EDIT:

Flux.fromStream(teams::join) is a code smell because it's holding a thread to fetch the result from CompletableFuture which is running on another thread.

Answer 2

Once you have downloaded the league table, and the team names are extracted from this table, I'm not sure you need a back-pressure ready stream to iterate over these items. A conversion of the stream to a standard list (or array) should be good enough, and should probably have better performance, no?

For instance:

String[] teamNames = teams.join().toArray(String[]::new);