Patricia Shanahan wrote:
On 9/11/2010 10:49 PM, Peter Firmstone wrote:
Patricia Shanahan wrote:
On 9/11/2010 7:56 PM, Peter Firmstone wrote:
...
3. Doesn't yet contain the TaskManager improvements.
...
I'll add the changes in 3 & 4, then run the tests again.
...
I'm worried about improving TaskManager without resolving a RetryTask
issue first. I think we may be being protected from concurrency bugs
by a tendency in TaskManager to go single runner thread when there are
runAfter dependencies.
Ok, I won't add the TaskManager change, so you can experiment with it.
What about the ServiceDiscoveryManager patch?
As far as I can tell, the SDM patch is a good thing, fixing a bug.
I have a general concern with the level of concurrency testing for
both SDM and JoinManager, but the only thing I can do about that is
try to write some more tests, which I am doing.
The problem is the status of a RetryTask that fails and is scheduled
for retry. As far as I can tell, it is off the TaskManager books, and
there is nothing preventing dependent tasks from getting ahead of it.
I don't know whether that is safe behavior.
There should be a FINEST log message whenever there is a retry. So
far, I have not seen it, though that may be a problem with my attempts
to turn up the logging level in RetryTask. I have no assurance that
retries are being tested.
Lets see what we can do to get some good logging output (I'm looking
into RetryTask's logging), then you can look at making some changes,
feel free to play around in the skunk/pepe branch, commit changes etc,
it's experimental, you can't hurt it.
I'm thinking of creating a branch for TaskManager work.
Ok, I'll be setting up a trunk replacement, but I may not get time until
next weekend. It'll be very similar to pepe, but without
org.apache.river.*, it'll include the original
net.jini.security.policy.DynamicPolicyProvider and be without the
StreamServiceRegistrar interfaces. I don't think any of these will
effect your work, though, with perhaps the exception of the
DynamicConcurrentPolicyProvider, reducing thread contention.
My current tests are running with FINEST enabled, they might show up
some RetryTask log messages, I'll check & let you know when they're
complete.
Excellent. Even an absence of retry messages would be useful
information, meaning that we are not testing retries.
I can run tests if you need me to, just svn commit your changes to
skunk, testing on different architectures might help to flush out the
concurrency issues.
In the mean time, I'll look at making a trunk replacement in another svn
directory, without experimental code, then if everyone's happy with it,
we can make it the official trunk.
Regards,
Peter.
