Ondřej Vašík reported that coreutils-8.3's tail-2/inotify-hash-abuse test is hanging when run via koji. You may recall that that system uses xen with a relatively old kernel on the host, yet quite new user-space in each guest.
At first I thought it must be hanging in one of the the until... loops, but he added an exec 2>&9 at the beginning of that script so we could see what's going on, and found that the wait was causing the hang. http://koji.fedoraproject.org/koji/getfile?taskID=1908889&name=build.log Here are the last few lines: + grep '^x$' out + tail -qF 1 2 3 4 5 6 7 8 9 + : + grep '^x$' out + mv 1 f + grep inaccessible out + echo a + dead=0 ++ seq 10 + for i in '$(seq 10)' + kill -0 544 + grep 'has appeared;' out + break + test 0 = 0 + kill -HUP 544 + wait 544 That means kill -HUP is failing to kill the tail process, so the wait hangs forever. I confirmed that tail doesn't do any signal manipulation itself, other than "kill" to check whether a parent is alive when --pid=PID is used. Ondřej confirmed that it hung also with -INT, though the ultimate test would be to use -TERM. Eventually someone will debug it...
