Hi Andres,
Are you planning to continue working on this? Summarizing the discussion
so far:
* Robert listed a bunch of little cleanup tasks
(http://www.postgresql.org/message-id/CA+TgmoZShVVqjaKUL6P3kDvZVPibtgkzoti3M+Fvvjg5V+XJ0A@mail.gmail.com).
Amit posted yet more detailed commends.
* We talked about changing the file layout. I think everyone is happy
with your proposal here:
http://www.postgresql.org/message-id/20140702131729.GP21169@awork2.anarazel.de,
with an overview description of what goes where
(http://www.postgresql.org/message-id/CA+TgmoYuxfsm2dSy48=JgUTRh1mozrvmjLHgqrFkU7_WPV-xLA@mail.gmail.com)
* Talked about nested spinlocks. The consensus seems to be to (continue
to) forbid nested spinlocks, but allow atomic operations while holding a
spinlock. When atomics are emulated with spinlocks, it's OK to acquire
the emulation spinlock while holding another spinlock.
* Talked about whether emulating atomics with spinlocks is a good idea.
You posted performance results showing that at least with the patch to
use atomics to implement LWLocks, the emulation performs more or less
the same as the current spinlock-based implementation. I believe
everyone was more or less satisfied with that.
So ISTM we have consensus that the approach to spinlock emulation and
nesting stuff is OK. To finish the patch, you'll just need to address
the file layout and the laundry list of other little details that have
been raised.
- Heikki
