java version "1.4.0"
Java(TM) 2 Runtime Environment, Standard Edition (build 1.4.0-b92)
Java HotSpot(TM) Client VM (build 1.4.0-b92, mixed mode)

Apologies if this is a dup, but I found no other obvious bug.

If you run the following program with JDK1.4 it spins in poll:


class Sleep {
    public static void main(String args[]) {
        while (true) {
            System.out.println("Going to sleep");
            try {
                Thread.sleep(60 * 1000);
            } catch (Exception e) {
            }
            System.out.println("Woke up");
        }
    }
}

Truss shows:
poll(0xFA281A38, 0, 10) = 0
poll(0xFA281A38, 0, 10) = 0
poll(0xFA281A38, 0, 10) = 0
lwp_cond_wait(0x0002B000, 0x0002AFE8, 0xFE581AD8) Err#62 ETIME
poll(0xFA281A38, 0, 10) = 0
poll(0xFA281A38, 0, 10) = 0
poll(0xFA281A38, 0, 10) = 0
poll(0xFA281A38, 0, 10) = 0
poll(0xFA281A38, 0, 10) = 0
poll(0xFA281A38, 0, 10) = 0
poll(0xFA281A38, 0, 10) = 0
poll(0xFA281A38, 0, 10) = 0

Looks like some thread is polling with a 10ms timeout.

I also tried JDK1.3.0 and it had the same behavior.

JDK1.2.1 does not have this behavior.

While this doesn't have too much impact on the system, it is annoying and
can interfere with debugging application code.

For kicks I took a look with dbx and it shows the main sleep thread that
is blocked in poll(), and another thread that is also in poll()
(presumeably spinning). Not sure what this watcher thread is, but it looks
like it is sleeping for 10ms.


(/opt/SUNWspro/bin/../WS5.0/bin/sparcv9/dbx) threads
 > t@1 a l@1 ?() running in _poll()
      t@2 b l@2 ?() running in _signotifywait()
      t@3 ?() sleep on (unknown) in _reap_wait()
      t@4 b l@5 _start() running in ___lwp_cond_wait()
      t@5 b l@6 _start() running in ___lwp_cond_wait()
      t@6 b l@7 _start() running in ___lwp_cond_wait()
      t@7 b l@8 _start() running in _poll()
      t@8 b l@9 _start() running in ___lwp_cond_wait()
      t@9 b l@10 _start() sleep on 0xfe46c178 in _lwp_sema_wait()
     t@10 b l@11 _start() running in ___lwp_cond_wait()
(/opt/SUNWspro/bin/../WS5.0/bin/sparcv9/dbx) thread t@7
t@7 (l@8) stopped in _poll at 0xff299884
0xff299884: _poll+0x0004: ta 0x8
(/opt/SUNWspro/bin/../WS5.0/bin/sparcv9/dbx) where
current thread: t@7
=>[1] _poll(0xf9981aa8, 0x0, 0xa, 0x0, 0x0, 0x0), at 0xff299884
  [2] _libc_select(0x0, 0x0, 0xff2bb1dc, 0xff2bb1dc, 0xff2bb1dc, 0xf9981aa8), at 0xff24c968
  [3] select(0x0, 0xa, 0x1, 0x0, 0x0, 0x0), at 0xff36b470
  [4] os::sleep(0xd1838, 0x0, 0xa, 0x0, 0x0, 0x0), at 0xfdd57688
  [5] WatcherThread::run(0xd1838, 0x0, 0x0, 0x0, 0x0, 0x0), at 0xfde667c8
  [6] _start(0xd1838, 0xff37f690, 0x1, 0x1, 0xff37e000, 0x0), at 0xfdd5377c




###@###.### 2002-05-13
The debugging issue is that when you use tools such as "truss" to try and
figure out what an application is doing you're flooded with polls. It also
seems counter-intuitive that a sleeping program is actually spinning.
In any case I probably can't make a compelling argument that this needs to be
fixed.