Experimental Physics and
Industrial Control System

"Jeff Hill" <[email protected]> · Thu, 14 Dec 2006 11:02:44 -0700

> 
> As I recall, the crux of this issue is that a shutdown procedure for the
> db
> ca link code would need to destroy all of its channels first, before
> destroying the ca context, and that would require that the individual
> records have shutdown procedures in place also - a big change.
> 

Typically there isn't a problem if the ca context is destroyed and the
channels are not deleted first. However in an IOC context we have in memory
channels and the situation is more complex. In that situation if the
channels are not destroyed before the ca context is destroyed we will
continue to receive monitor callbacks as the scan tasks run, and the monitor
callbacks crash when using a ca context that no-longer exists.

I further recall that this could be fixed if:
O CA were to keep a list of all in-memory channels
O When the context is destroyed we could run this list of in-memory channels
and uninstall all of their subscriptions

This would cause some duplication (wasted storage) however because the
database is already storing pointers to all of the channels that it creates,
and presumably we will never get shutdown fixed properly until the db has
implemented an orderly shutdown. 

Jeff

Experimental Physics and Industrial Control System

Experimental Physics and
Industrial Control System