How to return an already-gzipped stream

[aylwyne]

I have a service built on RestEasy Classic. This service has quarkus.resteasy.gzip.enabled = true and most REST resources return JSON and use the @GZIP annotation to automatically compress the responses.

I am now trying to integrate with a backend that is returning to me an already GZIPped data stream and I want to be able to return that from a Quarkus REST endpoint without unzipping it.

I am struggling with getting this to work. It seems I get double-gzipped responses. I have a REST resource without the @GZIP annotation. I have tried a basic test with the following method in my resource:

@GET
@Path("/test")
@Produces(MediaType.APPLICATION_JSON)
public Response test() {

    String json = "{\"foo\":\"bar\"}";
    try (ByteArrayOutputStream baos = new ByteArrayOutputStream()) {
        try (GZIPOutputStream gzip = new GZIPOutputStream(baos)) {
            gzip.write(json.getBytes(StandardCharsets.UTF_8));
        }
        baos.flush();
        return Response.ok(baos.toByteArray()).encoding("gzip").build();
    } catch (Exception e) {
        e.printStackTrace();
    }
    return null;
}

The number of bytes of the gzipped baos is 33 yet this results in 49 bytes returned. If I remove the .encoding("gzip") then I correctly get back 33 bytes but without the content-encoding header so the browser doesn't unzip it. However if I manually unzip that 33 bytes I do get the expected JSON.

So it seems like Quarkus is also doing its automatic gzipping on top of my own due to the quarkus.resteasy.gzip.enabled config.

Is there a way to have an endpoint return an already gzipped byte array unmodified?

Also, I did test with quarkus.resteasy.gzip.enabled = false and the above code works as expected. So I guess it boils down to needing to mix automatically gzipped endpoints with manually gzipped ones.

[cescoffier]

\CC @mkouba

[mkouba]

@aylwyne Are you using the RESTEasy Classic extension (quarkus-resteasy)? Because quarkus.resteasy.gzip.enabled should not be honored by Quarkus REST (quarkus-rest, formerly RESTEasy Reactive).

While reading our docs it seems that the @org.jboss.resteasy.annotations.GZIP annotation should be still needed to enable compression for a resource method.

If you were using Quarkus REST I'd recommend you to simply annotate your resource method with @Uncompressed but that would not help for RESTEasy Classic.

[gsmet]

Just throwing random ideas but if you can add a filter than runs after the Gzip encoder, you could add the appropriate header without having it encoded?

It could be triggered by a custom header that you could also drop in the filter for it to be transparent.

[aylwyne]

That's a good idea @gsmet. I'll pursue something like that. And thanks @mkouba for your help.

[gsmet]

If you end up with a working solution, please present it here, it will be interesting for the future readers.

[aylwyne]

The suggestion of intercepting the response after the GZIPEncodingInterceptor did the trick nicely. I invented a custom header that I use in the REST resource. When set, this interceptor removes that custom header and adds the normal Content-Encoding: gzip one.

Here's my working code (fyi, the 'pf' prefix on things is just an internal thing here):

Interceptor:

@Provider
@Priority(Priorities.USER)
public class PfGzipEncodingInterceptor implements AsyncWriterInterceptor {

    public static final String PF_GZIP_HEADER = "x-pf-gzip";
    private static final String GZIP_ENCODING = "gzip";

    @Override
    public CompletionStage<Void> asyncAroundWriteTo(AsyncWriterInterceptorContext context) {
        processHeaders(context.getHeaders());
        return context.asyncProceed();
    }

    @Override
    public void aroundWriteTo(WriterInterceptorContext context) throws IOException, WebApplicationException {
        processHeaders(context.getHeaders());
        context.proceed();
    }

    private void processHeaders(MultivaluedMap<String, Object> headers) {
        if (Boolean.TRUE.equals(headers.getFirst(PF_GZIP_HEADER))) {
            headers.put(HttpHeaders.CONTENT_ENCODING, List.of(GZIP_ENCODING));
            headers.remove(PF_GZIP_HEADER);
        }
    }

}

Test REST Resource:

@Path("/gzip")
public class GzipTestResource {

    @Path("/manual")
    @GET
    @Produces(MediaType.APPLICATION_JSON)
    public Response manualGzip() {
        String json = "{\"key\":\"value\"}";
        try (ByteArrayOutputStream baos = new ByteArrayOutputStream()) {
            try (GZIPOutputStream gzip = new GZIPOutputStream(baos)) {
                gzip.write(json.getBytes(StandardCharsets.UTF_8));
            }
            baos.flush();
            return Response.ok(baos.toByteArray()).header(PfGzipEncodingInterceptor.PF_GZIP_HEADER, true).build();
        } catch (Exception e) {
            e.printStackTrace();
            return Response.serverError().build();
        }
    }

    @Path("/automatic")
    @GET
    @GZIP
    @Produces(MediaType.APPLICATION_JSON)
    public Response automaticGzip() {
        Map<String, String> data = Map.of("key", "value");
        return Response.ok(data).build();
    }
}

[gsmet]

Awesome, thanks for sharing!