shaba/lvm2
1
0
Fork 0
mirror of git://sourceware.org/git/lvm2.git synced 2025-01-02 01:18:26 +03:00
Code Releases Activity

clvmd: singlenode signals only when lock changes

Browse Source 
There is no benefit in waking-up all the waiters
when there is no actual change in lock state.
This avoid some unnecessarily ping-pong effects like:

 Resource V_LVMTEST15724vg retrying lock in mode:WRITE...
 Resource V_LVMTEST15724vg already locked lockid=40, mode:WRITE
 Resource V_LVMTEST15724vg retrying lock in mode:WRITE...
 Resource V_LVMTEST15724vg already locked lockid=40, mode:WRITE
This commit is contained in:
Zdenek Kabelac 2015-04-08 23:12:29 +02:00
parent f32973c78e
commit 4459413225
2 changed files with 3 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
WHATS_NEW
View File

@ -1,5 +1,6 @@
Version 2.02.119 -
==================================
More efficient clvmd singlenode locking emulation.
Don't skip invalidation of cached orphans if vg write lck is held (2.02.118).
Log relevant PV tags when using cling allocation.
Add str_list_add_list() to combine two lists.

4
daemons/clvmd/clvmd-singlenode.c
View File

@ -208,8 +208,6 @@ static int _lock_resource(const char *resource, int mode, int flags, int *lockid
pthread_mutex_lock(&_lock_mutex);
retry:
pthread_cond_broadcast(&_lock_cond); /* to wakeup waiters */
if (!(head = dm_hash_lookup(_locks, resource))) {
if (flags & LCKF_CONVERT) {
/* In real DLM, lock is identified only by lockid, resource is not used */
@ -269,12 +267,14 @@ retry:
dm_list_add(head, &lck->list);
}
out:
pthread_cond_broadcast(&_lock_cond); /* to wakeup waiters */
pthread_mutex_unlock(&_lock_mutex);
DEBUGLOG("Locked resource %s, lockid=%d, mode=%s\n",
resource, lck->lockid, _get_mode(lck->mode));
return 0;
bad:
pthread_cond_broadcast(&_lock_cond); /* to wakeup waiters */
pthread_mutex_unlock(&_lock_mutex);
DEBUGLOG("Failed to lock resource %s\n", resource);