Hi VIFF guys,
I've implemented a VIFF application involving two binary searches. A
typical trace from one of the players during excecution looks like
this:
search for p1
search for p1: [-1, 4000]
search for p2
search for p2: [0, 4001]
search for p1: [-1, 4000]
search for p2: [0, 4001]
search for p1: [-1, 1999]
search for p2: [0, 2000]
search for p1: [999, 1999]
search for p2: [1000, 2000]
search for p1: [1499, 1999]
search for p2: [1500, 2000]
search for p1: [1499, 1749]
search for p2: [1500, 1750]
search for p1: [1499, 1624]
search for p2: [1500, 1625]
search for p1: [1499, 1561]
search for p2: [1500, 1562]
search for p1: [1530, 1561]
search for p2: [1531, 1562]
search for p1: [1530, 1545]
search for p2: [1531, 1546]
search for p1: [1537, 1545]
search for p2: [1538, 1546]
search for p1: [1541, 1545]
search for p2: [1542, 1546]
search for p1: [1541, 1543]
search for p2: [1542, 1544]
The two searches are independend of each other and they are hence
correctly scheduled to run in parallel by VIFF as indicated by the
trace.
This works all fine with VIFF version e89fb02c5e3d. However, if I
update VIFF to the current tip (8ec45943c12a), strange things start to
happen:
Player 1:
search for p1
search for p1: [-1, 4000]
search for p2
search for p2: [0, 4001]
search for p1: [-1, 4000]
search for p2: [0, 4001]
search for p1: [-1, 1999]
search for p2: [0, 2000]
search for p1: [999, 1999]
search for p2: [1000, 2000]
search for p1: [1499, 1999]
search for p2: [1500, 2000]
search for p1: [1749, 1999]
search for p2: [1750, 2000]
search for p1: [1874, 1999]
search for p2: [1750, 1875]
search for p1: [1936, 1999]
Player 2:
search for p1
search for p1: [-1, 4000]
search for p2
search for p2: [0, 4001]
search for p1: [-1, 4000]
search for p2: [0, 4001]
search for p1: [-1, 1999]
search for p2: [0, 2000]
search for p1: [999, 1999]
search for p2: [1000, 2000]
search for p1: [1499, 1999]
search for p1: [1749, 1999]
search for p1: [1874, 1999]
search for p1: [1936, 1999]
search for p1: [1936, 1967]
search for p1: [1951, 1967]
search for p1: [1951, 1959]
Player 3:
search for p1
search for p1: [-1, 4000]
search for p2
search for p2: [0, 4001]
search for p1: [-1, 4000]
search for p2: [0, 4001]
search for p2: [0, 2000]
search for p2: [1000, 2000]
search for p2: [1500, 2000]
search for p2: [1750, 2000]
search for p2: [1875, 2000]
search for p2: [1937, 2000]
search for p2: [1968, 2000]
search for p2: [1984, 2000]
search for p2: [1984, 1992]
search for p2: [1988, 1992]
search for p2: [1988, 1990]
It seems from the above that one of the players continue with one of
the searches (search for p1) while another player continues with the
other search (search for p2). This doesn't make sense since each step
in the binary search is synchronized, e.g. the players have to engage
in a comparison of a secret share in each step. The above trace is
just an example; it varies for each time I run the program.
Also, after the above trace, the players now deadlock where they used
to continue with other parts of the program.
It seems that the new code causes the two independent callback chains
for each binary search to somehow mingle in a wrong way.
Note also that I haven't enabled the new reactor, just installed the
new VIFF tip code. If I enable the new reactor, e.g. by including
import viff.reactor
viff.reactor.install()
in the code, I get the following exception:
Traceback (most recent call last):
File C:\Python25\Lib\site-packages\viff\runtime.py, line 401, in
inc_pc_wrapper
return method(self, *args, **kwargs)
File C:\Python25\Lib\site-packages\viff\passive.py, line 109, in open
self.activate_reactor()
File C:\Python25\Lib\site-packages\viff\runtime.py, line 838, in
activate_reactor
reactor.doIteration(0)
File C:\Python25\Lib\site-packages\viff\reactor.py, line 42, in doIteration
self.runUntilCurrent()
--- exception caught here ---
File C:\Python25\Lib\site-packages\twisted\internet\base.py, line
729, in runUntilCurrent
f(*a, **kw)
File C:\Python25\lib\site-packages\twisted\internet\defer.py, line
243, in callback
self._startRunCallbacks(result)
File C:\Python25\lib\site-packages\twisted\internet\defer.py, line
298, in _startRunCallbacks
raise AlreadyCalledError
twisted.internet.defer.AlreadyCalledError:
whereafter the players continue with the strange search as shown in
the above traces.
Since the only change between the right and the wrong behaviour of my
binary search algorithm described above was the VIFF update - my own
code remained unchanged - I guess that this is caused by a bug in the
new VIFF code. Do anyone have similar experiences? Anyone having a
clue of this bug?
Best regards,
Thomas
___
viff-devel mailing list (http://viff.dk/)
viff-devel@viff.dk
http://lists.viff.dk/listinfo.cgi/viff-devel-viff.dk
