A bit late, but still, some more thoughts about the entry cache... Let
me add some comments in this mail to be sure I understood what you have
in mind...
Le 4/8/12 9:16 PM, Selcuk AYA a écrit :
I am about to revisit the logical caches issue. My plan is to do the
following to handle all these caches in a generic way:
- a singe version number is kept for all caches.
The latest, I guess.
- a thread starting a txn read locks an internal readwrite lock.
fine.
- when a thread needs to modify a cache, it ugrades its lock to
exclusive lock.
It will block all the read on the cache until the cache update is done,
right ?
If it detects a version change during this time, it
throws a conflict exception. If no, it bumps up the version number and
changes the cache.
as the write lock will be exclusive, I assume that the cache
modification will be done by one single thread. Now, there is one race
condition that can occur if the thread modifying the cache has a
revision number lower than the current revision number. That means the
cache has been changed by anothe rthred. The timeline for such a case
would be :
time arrow --->
T(r1) o-------------[r1] modify cache
T(r2) o-----[r2] modify cache
When t(r1) tries to modify the cache, the cache already has a higher
revion in it (r2), even if the T(r1) thread has been started before.
In this case, we will throw a conflict exception on T(r1)
Is that what you have in mind ?
- After committing, thread releases the lock.
-If thread aborts its txn, then it notifies interceptors in its
interceptor chain of the abort. Any interceptor can then rebuild its
cache from what is on disk at this point. I am assuming this is
possible for all logical caches.
What about aggregating all the cache update we do in all the
interceptors in one single CacheInterceptor, responsible for the update
of all the caches ? The idea would be to globally lock the cache one
single time instead of doing so in many places. Accessing the caches
will be done through an helper class masking the access to internal
caches, with proper locks shared by all the threads.
Sounds good ?
--
Regards,
Cordialement,
Emmanuel Lécharny
www.iktek.com
