How to correctly read Flux<DataBuffer> and convert it to a single inputStream

This is really not as complicated as other answers imply.

The only way to stream the data without buffering it all in memory is to use a pipe, as @jin-kwon suggested. However, it can be done very simply by using Spring's BodyExtractors and DataBufferUtils utility classes.

Example:

private InputStream readAsInputStream(String url) throws IOException {
    PipedOutputStream osPipe = new PipedOutputStream();
    PipedInputSteam isPipe = new PipedInputStream(osPipe);

    ClientResponse response = webClient.get().uri(url)
        .accept(MediaType.APPLICATION.XML)
        .exchange()
        .block();
    final int statusCode = response.rawStatusCode();
    // check HTTP status code, can throw exception if needed
    // ....

    Flux<DataBuffer> body = response.body(BodyExtractors.toDataBuffers())
        .doOnError(t -> {
            log.error("Error reading body.", t);
            // close pipe to force InputStream to error,
            // otherwise the returned InputStream will hang forever if an error occurs
            try(isPipe) {
              //no-op
            } catch (IOException ioe) {
                log.error("Error closing streams", ioe);
            }
        })
        .doFinally(s -> {
            try(osPipe) {
              //no-op
            } catch (IOException ioe) {
                log.error("Error closing streams", ioe);
            }
        });

    DataBufferUtils.write(body, osPipe)
        .subscribe(DataBufferUtils.releaseConsumer());

    return isPipe;
}

If you don't care about checking the response code or throwing an exception for a failure status code, you can skip the block() call and intermediate ClientResponse variable by using

flatMap(r -> r.body(BodyExtractors.toDataBuffers()))

instead.

A slightly modified version of Bk Santiago's answer makes use of reduce() instead of collect(). Very similar, but doesn't require an extra class:

Java:

body.reduce(new InputStream() {
    public int read() { return -1; }
  }, (s: InputStream, d: DataBuffer) -> new SequenceInputStream(s, d.asInputStream())
).flatMap(inputStream -> /* do something with single InputStream */

Or Kotlin:

body.reduce(object : InputStream() {
  override fun read() = -1
}) { s: InputStream, d -> SequenceInputStream(s, d.asInputStream()) }
  .flatMap { inputStream -> /* do something with single InputStream */ }

Benefit of this approach over using collect() is simply you don't need to have a different class to gather things up.

I created a new empty InputStream(), but if that syntax is confusing, you can also replace it with ByteArrayInputStream("".toByteArray()) instead to create an empty ByteArrayInputStream as your initial value instead.

I was able to make it work by using Flux#collect and SequenceInputStream

@Override
public Mono<T> extract(ClientHttpResponse response, BodyExtractor.Context context) {
  Flux<DataBuffer> body = response.getBody();
  return body.collect(InputStreamCollector::new, (t, dataBuffer)-> t.collectInputStream(dataBuffer.asInputStream))
    .map(inputStream -> {
      try {
        JaxBContext jc = JaxBContext.newInstance(SomeClass.class);
        Unmarshaller unmarshaller = jc.createUnmarshaller();

        return (T) unmarshaller.unmarshal(inputStream);
      } catch(Exception e){
        return null;
      }
  }).next();
}

InputStreamCollector.java

public class InputStreamCollector {
  private InputStream is;

  public void collectInputStream(InputStream is) {
    if (this.is == null) this.is = is;
    this.is = new SequenceInputStream(this.is, is);
  }

  public InputStream getInputStream() {
    return this.is;
  }
}