Jaroslav, Use fuser command to wait until file being closed is better way to do it
if fuser -s filename; then Busy; else Ready; fi -Dmitry On 2013-11-19 13:00, Jaroslav Bachorik wrote: > On 19.11.2013 09:45, Alan Bateman wrote: >> On 18/11/2013 23:29, Erik Gahlin wrote: >>> Could I have a review of this test fix. >>> >>> It appears it takes more than 10 s to write to a file on slow >>> machines. Instead of sleeping 10 s the test now polls for the file to >>> appear. There may be other issues with the test, but it's hard to know >>> without ruling this race out. >>> >>> Thanks >>> Erik >>> >>> Testing: >>> - I have run the test 20+ times locally on a highly saturated CPU >>> without failure >>> - JPRT. >>> >>> Bug: >>> https://bugs.openjdk.java.net/browse/JDK-6402201 >>> >>> Webrev: >>> http://cr.openjdk.java.net/~egahlin/6402201_1/ >> This looks okay to me. The only thing that comes to mind is whether it >> is possible for to observe the port file before the port is written. > > I think it is. Writing to a file does not guarantee that it's creation > and actual write are atomic :( > > What I used to do was to write the data to one file and the use a second > signaling file to notify the other party that the data has been > successfully written and can be read. > > -JB- > >> Keeping the sleep so that it happens after you've file the file without >> significantly reduce that possibility. >> >> -Alan. > -- Dmitry Samersoff Oracle Java development team, Saint Petersburg, Russia * I would love to change the world, but they won't give me the sources.
