RFR: 8312433: HttpClient request fails due to connection being considered idle and closed [v3]

[Daniel Jeliński]

On Thu, 27 Jul 2023 10:03:06 GMT, Jaikiran Pai <jpai at openjdk.org> wrote:

>> Can I please get a review of this change which proposes to fix the issue noted in https://bugs.openjdk.org/browse/JDK-8312433?
>> 
>> In JDK 20, we introduced a way to identify and close idle HTTP2 connections in the HttpClient implementation through https://bugs.openjdk.org/browse/JDK-8288717. Whenever a HTTP2 stream gets closed and if there are no other streams on the connection, we create a idle timeout event that times out after a (configurable) duration. After the event has been registered, if any more streams get opened on that connection, then the event is cancelled. If no streams are created during that period, then the event fires and we close the connection.
>> We have a race condition in this implementation, as noticed in https://bugs.openjdk.org/browse/JDK-8312433. When a HTTP2 connection is created, we pool that connection. Any subsequent request against the same target server will reuse this pooled connection if it's still open. When user code issues a request, we checkout a connection from this pool (if relevant) and then use that connection to create a HTTP2 stream and issue the request. This sequence has a small window of time, where a pooled connection which has a idle timeout event scheduled (because it had no active streams) is handed out of the pool and before it can create a new HTTP2 stream, gets closed (concurrently by the idle connection timeout event). This results in the HTTP2 stream creation to fail with an exception that effectively fails the user initiated request.
>> 
>> The commit in this PR addresses it by introducing an internal method `tryReserveForPoolCheckout()` on the `Http2Connection` which is responsible for co-ordinating with the idle connection management to ensure that if this connection does get handed out of the pool, then the idle connection event doesn't end up closing it.
>> 
>> A new jtreg test has been added to test this issue. This test consistently reproduces the issue without this fix and passes with this fix. Ideally, a new test method could have been added to the existing `test/jdk/java/net/httpclient/http2/IdleConnectionTimeoutTest.java` but that test has multiple `@run` with different timeout values ranging all the way upto 30 seconds and it focuses more on the timeout value parsing and honouring that value. Introducing this test method in that existing test would end up testing this against all those values for no reason, plus would also end up increasing the duration of that test. So I decided to introd...
>
> Jaikiran Pai has updated the pull request with a new target base due to a merge or a rebase. The incremental webrev excludes the unrelated changes brought in by the merge/rebase. The pull request contains five additional commits since the last revision:
> 
>  - minor - add a code comment in the test method explaining the use of the magic number
>  - merge latest from master branch
>  - Daniel's review - no need for volatile
>  - minor cleanup to simplify the code
>  - 8312433: HttpClient request fails due to connection being considered idle and closed

LGTM. Thanks!

-------------

Marked as reviewed by djelinski (Reviewer).

PR Review: https://git.openjdk.org/jdk/pull/15012#pullrequestreview-1549487433