Hello, I am trying to implement a HTTP2 server in netty, only h2c is needed, no TLS. The code is based on the sample found in io.netty.example.http2.helloworld.server. Right now, I am struggling with how to correctly send back a response asynchronously from when a request was received in a Http2FrameListener. See my current implementation at:
https://github.com/rrva/netty-http2/blob/main/src/main/kotlin/main.kt it is a small hello world server. For GET requests, I read the request in Http2FrameListener.onHeadersRead and simulate that a response is computed asynchronously by some other thread pool than the netty event loop, and when the response CompletableFuture completes I want to send the response back in the correct context. Here I tried to execute the sending on the response on the same executor as the context event loop. The implementation I linked above is obviously not correct. With this implementation, I sometimes get a correct response sent back to the client and sometimes only headers are sent back but no data frame. I assume this is because I am using HTTP2FrameListener and the context in the wrong way and the frame encoder encounters some race condition or similar. In the code, if the request path is /graphql then I try this async response handling, but on other paths like / the response is sent back directly and that works fine. How would a better implementation look like if I still want to execute the computation of the response in another thread pool? to send responses I use: private fun sendResponse( ctx: ChannelHandlerContext, streamId: Int, payload: ByteBuf, status: HttpResponseStatus = HttpResponseStatus.OK ) { val headers = DefaultHttp2Headers().status(status.codeAsText()) encoder().writeHeaders(ctx, streamId, headers, 0, false, ctx.newPromise()) encoder().writeData(ctx, streamId, payload, 0, true, ctx.newPromise()) ctx.flush() } in Http2FrameListener.onHeadersRead I have a graphResponseFuture which is the future for my response which I then invoke sending the response on like this (see github link above for full code): graphResponseFuture.whenCompleteAsync({ graphResponse, throwable -> if (throwable != null) { log.error("Error processing data", throwable) val errorPayload = ctx.alloc().buffer().writeBytes("Internal server error". toByteArray()) sendResponse(ctx, streamId, errorPayload, HttpResponseStatus. INTERNAL_SERVER_ERROR) } else { val responseBuffer = ctx.alloc().buffer() serializeGraphResponseToByteBuf(graphResponse, responseBuffer) sendResponse(ctx, streamId, responseBuffer) } }, ctx.channel().eventLoop()) When testing, I reproduce problems by repeatedly requesting like this: curl -v --http2-prior-knowledge "http://localhost:9093/graphql"; but I never see issues with curl -v --http2-prior-knowledge "http://localhost:9093/"; the difference being that in the second request the response is sent synchronously when the request handler is invoked. -- You received this message because you are subscribed to the Google Groups "Netty discussions" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. To view this discussion on the web visit https://groups.google.com/d/msgid/netty/ead5cf45-c78c-4965-8c73-ea7b1b253b35n%40googlegroups.com.
