I'm using the new java.net.http
classes to handle asynchronous HTTP request+response exchanges, and I'm trying to find a way to have the BodySubscriber handle different encoding types such as gzip.
However, mapping a BodySubsriber<InputStream>
so that the underlying stream is wrapped by a GZIPInputStream
(when "Content-Encoding: gzip" is found in the response header) leads to a hang. No exceptions, just a total cessation of activity.
The code which maps the BodySubscriber
looks like this:
private HttpResponse.BodySubscriber<InputStream> gzippedBodySubscriber(
HttpResponse.ResponseInfo responseInfo) {
return HttpResponse.BodySubscribers.mapping(
HttpResponse.BodySubscribers.ofInputStream(),
this::decodeGzipStream);
}
private InputStream decodeGzipStream(InputStream gzippedStream) {
System.out.println("Entered decodeGzipStream method.");
try {
InputStream decodedStream = new GZIPInputStream(gzippedStream);
System.out.println(
"Created GZIPInputStream to handle response body stream.");
return decodedStream;
} catch (IOException ex) {
System.out.println("IOException occurred while trying to create GZIPInputStream.");
throw new UncheckedIOException(ex);
}
}
Receiving an HTTP response which has "gzip" encoding leads to the console showing just this:
Entered EncodedBodyHandler.apply method.
Entered decodeGzipStream method.
Nothing more is seen, so the line after the call to the GZIPInputStream
constructor is never executed.
Does anyone know why this attempt to wrap the InputStream
from a BodySubscriber<InputStream>
in a GZIPInputStream
is hanging?
Note: the equivalent method for unencoded (raw text) HTTP response bodies contains simply a call to BodySubscribers.ofInputStream()
with no mapping, and this allows the response to be received and displayed without problem.
This is indeed a bug. I have logged JDK-8217264. I can suggest two work-arounds:
Workaround one
Do not use BodySubscribers.mapping
- but transform the InputStream
into a GZIPInputStream
after getting the HttpResponse's body:
GZIPInputStream gzin = new GZIPInputStream(resp.getBody());
Workaround two
Have the mapping function return a Supplier<InputStream>
instead, taking care not to create the GZIPInputStream
until Supplier::get
is called
static final class ISS implements Supplier<InputStream> {
final InputStream in;
GZIPInputStream gz;
ISS(InputStream in) {
this.in = in;
}
public synchronized InputStream get() {
if (gz == null) {
try {
gz = new GZIPInputStream(in);
} catch (IOException t) {
throw new UncheckedIOException(t);
}
}
return gz;
}
}
Encountered the exact same problem. I tried the example in the Javadoc of the BodySubscribers.mapping
method. Same behavior, the application hangs without any errors.
Could be a bug, because this is an official example from the Javadoc.
public static <W> BodySubscriber<W> asJSON(Class<W> targetType) {
BodySubscriber<InputStream> upstream = BodySubscribers.ofInputStream();
BodySubscriber<W> downstream = BodySubscribers.mapping(
upstream,
(InputStream is) -> {
try (InputStream stream = is) {
ObjectMapper objectMapper = new ObjectMapper();
return objectMapper.readValue(stream, targetType);
} catch (IOException e) {
throw new UncheckedIOException(e);
}
});
return downstream;
} }
来源:https://stackoverflow.com/questions/53379087/wrapping-bodysubscriberinputstream-in-gzipinputstream-leads-to-hang