On Thu, 2 Sep 2021 14:24:24 GMT, Daniel Fuchs <dfu...@openjdk.org> wrote:
>> Please review this change to the Unix implementations of >> sun.nio.ch.sctp.Sctp*ChannelImpl#implCloseSelectableChannel() >> to be same as SocketChannelImpl at JDK-7118373. (The preClose is missing a >> check for the ST_KILLED state.) > > I have run this change on one of our machines that support SCTP. I did get > some intermittent failures with the other SCTP tests - they don't seem much > stable - but the new proposed test was failing all the time. I suspect that > using `lsof` to figure out whether the file descriptor was closed is not > reliable/portable enough. I also tried to modify the test to use /othervm - > which is probably a good idea if you don't want your results to be polluted > by whatever other test might have run previously/concurrently in the agent VM > - but to no avail: the test was still failing 100% of the time. > @masyano could you figure another way to detect whether the file descriptor > has been released? @dfuch I tried to count `/proc/<PID>/fd`, but there are recorded all fds in the process. I don't know how to pick up only active fds in `/proc/<PID>/fd`. So, I will change to call lsof last one time only (like https://github.com/openjdk/jdk/pull/4621), and counts all unreleased fds. Could you try this test case? ------------- PR: https://git.openjdk.java.net/jdk/pull/5274
