RFR: 8291637: HttpClient default keep alive timeout not followed if server sends invalid value

[Jaikiran Pai]

On Thu, 4 Aug 2022 20:41:33 GMT, Michael McMahon <michaelm at openjdk.org> wrote:

> Hi,
> 
> Some new keep alive tests are exposing some old bugs. In this case if the server sends an invalid timeout (say -20 seconds) we accept it creating a timeout in the past. So, the first time the keep alive thread wakes up it will close the connection.
> The correct behavior is to ignore the invalid parameter and fallback to the default timeout or the timeout set by the relevant system property.
> 
> Thanks,
> Michael

src/java.base/share/classes/sun/net/www/http/HttpClient.java line 902:

> 900:                             responses.findValue("Keep-Alive"));
> 901:                         /* default should be larger in case of proxy */
> 902:                         keepAliveConnections = p.findInt("max", usingProxy?50:5);

Hello Michael, should we do something similar for this `max` parameter value too and (re)set the `keepAliveConnections` to the defaults, if the server sends a negative value?

test/jdk/sun/net/www/http/KeepAliveCache/B8291637.java line 27:

> 25:  * @test
> 26:  * @bug 8291637
> 27:  * @run main/othervm -Dhttp.keepAlive.time.server=20 -esa -ea B8291637

Is it intentional that we are setting the -esa and -ea options here?

test/jdk/sun/net/www/http/KeepAliveCache/B8291637.java line 103:

> 101:                 }
> 102:             } catch (IOException e) {
> 103:                 System.err.println("Server exception terminating");

It looks to me that if an exception occurs (not just `IOException`) then this test might hang (and timeout with the jtreg timeout) at the `passed.get()` call in the `main()` method. Should we perhaps catch `Throwable` here and do `passed.completeExceptionally`?

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

PR: https://git.openjdk.org/jdk/pull/9755