BufferManager page replacement logic is faulty
Closed this issue · 3 comments
GoogleCodeExporter commented
Currently, there is a chance that the LRU page selected for eviction is not the
right one, as a
preference is given to pages that aren't dirty.
Correct behavior would be to evict the LRU page, and if it is dirty, write it
to disk prior to eviction.
This used to be the behavior before the change was made to ensure that all page
writes are
performed by the BufferWriter thread.
Some other enhancements are possible:
Use condition to wait for IO to complete rather than the pause/retry logic
currently employed.
Original issue reported on code.google.com by d.majum...@gmail.com on 21 Sep 2008 at 3:21
GoogleCodeExporter commented
Fixed in svn.
Original comment by d.majum...@gmail.com on 9 Nov 2008 at 12:41
GoogleCodeExporter commented
Original comment by d.majum...@gmail.com on 27 Dec 2008 at 9:48
- Changed state: Started
GoogleCodeExporter commented
Fixed in 1.0.12-BETA
Original comment by d.majum...@gmail.com on 7 Apr 2009 at 9:42
- Changed state: Fixed