diff options
author | unknown <guilhem@gbichot3.local> | 2006-12-18 17:24:02 +0100 |
---|---|---|
committer | unknown <guilhem@gbichot3.local> | 2006-12-18 17:24:02 +0100 |
commit | 7199c905590391f64802913369aab7d288eff4c8 (patch) | |
tree | d38393d634c6dc8b3886863dbdab2d348526eb45 /BUILD/compile-pentium-valgrind-max | |
parent | 71b404973c1f6343e9e63d3179c65f3642aade9a (diff) | |
download | mariadb-git-7199c905590391f64802913369aab7d288eff4c8.tar.gz |
WL#3071 Maria checkpoint
- cleanups, simplifications
- moving the construction of the "dirty pages table" into the
pagecache where it belongs (because it's the pagecache which knows
dirty pages). TODO: do the same soon for the "transactions table".
- fix for a small bug in the pagecache (decrementation of "changed_blocks")
include/pagecache.h:
prototype
mysys/mf_pagecache.c:
m_string.h moves up for LEX_STRING to be known for pagecache.h.
In pagecache_delete_page(), we must decrement "blocks_changed" even
if we just delete the page without flushing it.
A new function pagecache_collect_changed_blocks_with_LSN()
(used by the Checkpoint module), which stores information about the
changed blocks (a.k.a. "the dirty pages table") into a LEX_STRING.
This function is not tested now, it will be when there is a Checkpoint.
storage/maria/ma_checkpoint.c:
refining the checkpoint code: factoring functions, moving the
construction of the "dirty pages table" into mf_pagecache.c
(I'll do the same with the construction of the "transactions table"
once Serg tells me what's the best way to do it).
storage/maria/ma_least_recently_dirtied.c:
Simplifying the thread which does background flushing of
least-recently-dirtied pages:
- in first version that thread will not flush, just do checkpoints
- in 2nd version, flushing should re-use existing page cache functions
like flush_pagecache_blocks().
unittest/mysys/test_file.h:
m_string.h moves up for LEX_STRING to be known in pagecache.h
Diffstat (limited to 'BUILD/compile-pentium-valgrind-max')
0 files changed, 0 insertions, 0 deletions