Skip to content
Projects
Groups
Snippets
Help
Loading...
Help
Support
Keyboard shortcuts
?
Submit feedback
Contribute to GitLab
Sign in / Register
Toggle navigation
M
mariadb
Project overview
Project overview
Details
Activity
Releases
Repository
Repository
Files
Commits
Branches
Tags
Contributors
Graph
Compare
Issues
0
Issues
0
List
Boards
Labels
Milestones
Merge Requests
0
Merge Requests
0
Analytics
Analytics
Repository
Value Stream
Wiki
Wiki
Snippets
Snippets
Members
Members
Collapse sidebar
Close sidebar
Activity
Graph
Create a new issue
Commits
Issue Boards
Open sidebar
Kirill Smelkov
mariadb
Commits
23fa28bb
Commit
23fa28bb
authored
Jan 07, 2014
by
John Esmet
Browse files
Options
Browse Files
Download
Email Patches
Plain Diff
fixes #135 Don't hold the write list lock while evicting a node
parent
96afd28f
Changes
1
Hide whitespace changes
Inline
Side-by-side
Showing
1 changed file
with
22 additions
and
38 deletions
+22
-38
ft/cachetable.cc
ft/cachetable.cc
+22
-38
No files found.
ft/cachetable.cc
View file @
23fa28bb
...
...
@@ -651,39 +651,6 @@ static void cachetable_free_pair(PAIR p) {
ctpair_destroy
(
p
);
}
// Maybe remove a pair from the cachetable and free it, depending on whether
// or not there are any threads interested in the pair. The flush callback
// is called with write_me and keep_me both false, and the pair is destroyed.
// The sole purpose of this function is to remove the node, so the write_me
// argument to the flush callback is false, and the flush callback won't do
// anything except destroy the node.
//
// on input, pair_list's write lock is held and PAIR's mutex is held
// on exit, only the pair_list's write lock is still held
//
static
void
cachetable_maybe_remove_and_free_pair
(
pair_list
*
pl
,
evictor
*
ev
,
PAIR
p
)
{
// this ensures that a clone running in the background first completes
if
(
p
->
value_rwlock
.
users
()
==
0
&&
p
->
refcount
==
0
)
{
// assumption is that if we are about to remove the pair
// that no one has grabbed the disk_nb_mutex,
// and that there is no cloned_value_data, because
// no one is writing a cloned value out.
assert
(
nb_mutex_users
(
&
p
->
disk_nb_mutex
)
==
0
);
assert
(
p
->
cloned_value_data
==
NULL
);
cachetable_remove_pair
(
pl
,
ev
,
p
);
pair_unlock
(
p
);
cachetable_free_pair
(
p
);
}
else
{
pair_unlock
(
p
);
}
}
// assumes value_rwlock and disk_nb_mutex held on entry
// responsibility of this function is to only write a locked PAIR to disk
// and NOTHING else. We do not manipulate the state of the PAIR
...
...
@@ -2458,10 +2425,10 @@ static void remove_pair_for_close(PAIR p, CACHETABLE ct, bool completely) {
assert
(
p
->
dirty
==
CACHETABLE_CLEAN
);
assert
(
p
->
refcount
==
0
);
if
(
completely
)
{
// TODO: maybe break up this function
// so that write lock does not need to be held for entire
// free
cachetable_
maybe_remove_and_free_pair
(
&
ct
->
list
,
&
ct
->
ev
,
p
);
cachetable_remove_pair
(
&
ct
->
list
,
&
ct
->
ev
,
p
);
pair_unlock
(
p
);
//
TODO: Eventually, we should not hold the write list lock during
free
cachetable_
free_pair
(
p
);
}
else
{
// if we are not evicting completely,
...
...
@@ -4235,8 +4202,25 @@ void evictor::evict_pair(PAIR p, bool for_checkpoint) {
nb_mutex_unlock
(
&
p
->
disk_nb_mutex
);
// at this point, we have the pair list's write list lock
// and we have the pair's mutex (p->mutex) held
cachetable_maybe_remove_and_free_pair
(
m_pl
,
this
,
p
);
// this ensures that a clone running in the background first completes
bool
removed
=
false
;
if
(
p
->
value_rwlock
.
users
()
==
0
&&
p
->
refcount
==
0
)
{
// assumption is that if we are about to remove the pair
// that no one has grabbed the disk_nb_mutex,
// and that there is no cloned_value_data, because
// no one is writing a cloned value out.
assert
(
nb_mutex_users
(
&
p
->
disk_nb_mutex
)
==
0
);
assert
(
p
->
cloned_value_data
==
NULL
);
cachetable_remove_pair
(
m_pl
,
this
,
p
);
removed
=
true
;
}
pair_unlock
(
p
);
m_pl
->
write_list_unlock
();
// do not want to hold the write list lock while freeing a pair
if
(
removed
)
{
cachetable_free_pair
(
p
);
}
}
//
...
...
Write
Preview
Markdown
is supported
0%
Try again
or
attach a new file
Attach a file
Cancel
You are about to add
0
people
to the discussion. Proceed with caution.
Finish editing this message first!
Cancel
Please
register
or
sign in
to comment