On Tue, 8 Dec 2020 16:27:19 GMT, Xue-Lei Andrew Fan <xue...@openjdk.org> wrote:
>> The test sun/security/ssl/SSLSocketImpl/SSLSocketLeak.java caused sporadic >> noise because sometimes it opens more file handles than expected. It was >> moved to a manual test to quiesce this >> ([JDK-8257670](https://bugs.openjdk.java.net/browse/JDK-8257670)) >> >> It would be good, however, to have this test as an automatic test to be able >> to spot potential regressions. The threshold for windows should be adapted. > > test/jdk/sun/security/ssl/SSLSocketImpl/SSLSocketLeak.java line 61: > >> 59: System.out.println("FDs in the end: " + fds_end); >> 60: >> 61: if ((fds_end - fds_start) > (NUM_TEST_SOCK / >> OPEN_HANDLE_GROWTH_THRESHOLD)) { > > It looks like the test may still fail intermittently. But let's do it for > now and see if it is reliable to be an automation test case. The failing cases I've seen in SAP's infrastructure have all been on Windows and the number of handles was marginally above the 10 percent threshold. Do you have seen otherwise in Oracle's tests? The test prints out the handle count before and after... ------------- PR: https://git.openjdk.java.net/jdk/pull/1686
