[Oz] Re: Tapping a watcher on the shoulder

[Erik Klintskog]

Peter Van Roy wrote:

> [For users members: the watcher model is part of Mozart's
> language-based fault detection, accessed through the Fault module.  A
> watcher is installed on a language entity, e.g., on a port, which can
> be on a remote site.  The watcher triggers a user-defined procedure
> when the port has a temporary or permanent failure.  Watchers are very
> useful for building fault tolerance abstractions in the language.]
>
> Metwally has found an interesting missing feature in the watcher model.
> An installed watcher will eventually trigger when the entity on which
> it watches fails.  During the recovery phase of the GlobalStore, having
> to wait this long to be informed of failed clients would make the
> recovery take very long. What Metwally wants to do is 'tap the watcher
> on the shoulder', i.e., give it a hint to check *now* whether its
> entity has failed.  Currently, he does this manually, by sending a kind
> of ``ping'' message, independent of the watchers.  He finds that it's
> much faster than the watchers. But, logically, what he is doing is too
> low level and it should be inside the watchers.  (From the viewpoint of
> language semantics, giving a hint is a no-operation.)

The wathcers are suposed to work the way you want them to do. You
should not have to perform operations on the entities to trigger them.

In the current system, the one that is to  be released, the message
passing
layer does a lot of pinging to uphold a consistent view of the system so
this
sounds strange to me.

Could you give me an example that gives this behavior you describe, so we
can have a look
at it. I assume that you are using the latest version of Mozart.

/Erik



-
Please send submissions to users at mozart-oz.org
and administriva mail to users-request at mozart-oz.org.
The Mozart Oz web site is at http://www.mozart-oz.org/.