On Thu, 9 Jun 2022 11:27:47 GMT, Conor Cleary <ccle...@openjdk.org> wrote:
>> **Issue** >> It was observed that when the httpclient sends a POST request with the >> `Expect: 100 Continue` header set and the server replies with a response >> code `417 Expectation Failed` that the httpclient hangs indefinitely when >> the version of Http used is HTTP/2. However, it was also seen that the issue >> persisted with HTTP/1_1 with the same usage. >> >> This was caused by an implementation in ExchangeImpl that resulted in two >> calls to readBodyAsync() in this case, where the second call causes the >> indefinite hanging (as if there was a respomse body, it has already been >> read). >> >> **Solution** >> When ExchangeImpl::expectContinue() detects that a response code 417 is >> received, two things occur. Firstly, a flag is set which ensures that the >> connection is closed locally in this case. Secondly, the response is >> returned to the client as a failed future, A unit test was added to ensure >> that this usage of the httpclient does not cause hanging. > > Conor Cleary has updated the pull request incrementally with two additional > commits since the last revision: > > - 8286171: Package-protected access for method > - 8286171: Added checks for correct response codes test/jdk/java/net/httpclient/ExpectContinueTest.java line 102: > 100: // Due to limitations of the above Http1 Server, a manual > approach is taken to test the hanging with the > 101: // httpclient using Http1 so that the correct response header > can be returned for the test case > 102: http1HangServer = new Http1HangServer(saHang); It took me a while to understand why we have special server implementation for this. The comment helped :) It reminded me about the `ServerImpl` used by the `HttpServer`, which unconditionally sends a `100` response code when it sees a `Expect: 100-continue` in the request header. ------------- PR: https://git.openjdk.java.net/jdk/pull/9093
