On Mon, 2018-09-24 at 16:56 +0200, Jack van Ooststroom wrote:
> Recently we switched from the Apple's legacy Binary Provider API to
> the
> newer HTTP/2 based API in order to send our notifications using APNs.
> We
> decided on using Apache's HttpClient 5 in order to meet the HTTP/2
> requirement. However, our current support for this seems unstable as
> we
> are seeing the occasional H2StreamResetExceptions and
> TimeoutExceptions.
>
> We are currently creating the HttpAsyncClient as follows:
>
> HttpAsyncClients.custom().
> setVersionPolicy(HttpVersionPolicy.FORCE_HTTP_2).
> setConnectionManager(
> PoolingAsyncClientConnectionManagerBuilder.create().
> setTlsStrategy(
> new H2TlsStrategy(
> SSLContexts.custom().
> loadKeyMaterial(_keyStore, _keyStorePassword).
> build(),
> new String[]
> {"TLSv1.2"},
> null, null, new DefaultHostnameVerifier()
> )
> ).
> build()
> ).
> build()
>
> This HttpAsyncClient is shared among a configurable amount of
> threads
> with each thread having its own HttpClientContext instance. Each
> thread
> is sending our HTTP/2 requests as follows:
>
> Future _future =
> getHTTPClient().execute(_request, getHTTPContext(), null);
> try {
> SimpleHttpResponse _response = _future.get(3,
> TimeUnit.MINUTES);
> ...
> } catch (final CancellationException exception) { // If
> the
> computation was cancelled.
> ...
> } catch (final ExecutionException exception) { // If the
> computation threw an exception.
> ...
> } catch (final InterruptedException exception) { // If
> the
> current thread was interrupted while waiting.
> ...
> } catch (final TimeoutException exception) { // If the
> wait
> timed out.
> ...
> }
>
> On occasion we see and log the H2StreamResetException, wrapped inside
> an
> ExecutionException, for now. The details of this exception is as
> follows:
>
> java.util.concurrent.ExecutionException:
> org.apache.hc.core5.http2.H2StreamResetException: Connection
> terminated by the peer
> at
> org.apache.hc.core5.concurrent.BasicFuture.getResult(BasicFuture.
> java:71)
> at
> org.apache.hc.core5.concurrent.BasicFuture.get(BasicFuture.java:1
> 02)
> ...
> Caused by: org.apache.hc.core5.http2.H2StreamResetException:
> Connection terminated by the peer
> at
> org.apache.hc.core5.http2.impl.nio.AbstractHttp2StreamMultiplexer
> .consumeFrame(AbstractHttp2StreamMultiplexer.java:974)
> at
> org.apache.hc.core5.http2.impl.nio.AbstractHttp2StreamMultiplexer
> .onInput(AbstractHttp2StreamMultiplexer.java:419)
> at
> org.apache.hc.core5.http2.impl.nio.AbstractHttp2IOEventHandler.in
> putReady(AbstractHttp2IOEventHandler.java:63)
> at
> org.apache.hc.core5.http2.impl.nio.ClientHttp2IOEventHandler.inpu
> tReady(ClientHttp2IOEventHandler.java:38)
> at
> org.apache.hc.core5.reactor.InternalDataChannel.onIOEvent(Interna
> lDataChannel.java:117)
> at
> org.apache.hc.core5.reactor.InternalChannel.handleIOEvent(Interna
> lChannel.java:50)
> at
> org.apache.hc.core5.reactor.SingleCoreIOReactor.processEvents(Sin
> gleCoreIOReactor.java:173)
> at
> org.apache.hc.core5.reactor.SingleCoreIOReactor.doExecute(SingleC
> oreIOReactor.java:123)
> at
> org.apache.hc.core5.reactor.AbstractSingleCoreIOReactor.execute(A
> bstractSingleCoreIOReactor.java:80)
> at
> org.apache.hc.core5.reactor.IOReactorWorker.run(IOReactorWorker.j
> ava:44)
> ... 1 more
Please note that HttpClient signals the fact that the connection has
been terminated by the opposite endpoint. By itself this does not mean
there is anything wrong with the client side. H2StreamResetException
includes an error code that would help understand the reason for
connection termination.
>
> As we are using a PoolingAsyncClientConnectionManager and this
> exception
> is indicating "Connection terminated by the peer", is there something
> we
> need to do on our end in order to deal with this exception and to
> recover from it?
>
Depending on the error code and the request method it might be
sufficient to re-try the request.
> Also on occassion we see and log the TimeoutException for now. The
> details of this exception is as