Re: [PATCH] jbd2: do not start commit when t_updates does not back to zero

[Andreas Dilger]

On Mar 29, 2019, at 3:25 PM, Jan Kara <jack@xxxxxxx> wrote:
> 
> On Sun 24-03-19 11:38:35, Liu Song wrote:
>> When t_updates back to zero, it guaranteed wake up process which
>> waiting on j_wait_updates. If we triggered a commit start without
>> considered t_updates, the commit thread wakes up and find t_updates
>> is not zero, it have to wait on it once again. So, add checking code
>> to avoid this happen.
>> 
>> Signed-off-by: Liu Song <liu.song11@xxxxxxxxxx>
> 
> Do I understand correctly that this is a performance improvement? If yes,
> did you measure any benefit of the patch? Because I have some doubts that
> t_updates == 0 case is very common.

In the past there have been times when periodic journal commits keep a disk
from spinning down, but I don't know if that is the case here...

Cheers, Andreas

>> ---
>> fs/jbd2/transaction.c | 5 +++--
>> 1 file changed, 3 insertions(+), 2 deletions(-)
>> 
>> diff --git a/fs/jbd2/transaction.c b/fs/jbd2/transaction.c
>> index 79a028a7a579..e0499fd73b1e 100644
>> --- a/fs/jbd2/transaction.c
>> +++ b/fs/jbd2/transaction.c
>> @@ -144,12 +144,13 @@ static void wait_transaction_locked(journal_t *journal)
>> 	__releases(journal->j_state_lock)
>> {
>> 	DEFINE_WAIT(wait);
>> -	int need_to_start;
>> +	int need_to_start = 0;
>> 	tid_t tid = journal->j_running_transaction->t_tid;
>> 
>> 	prepare_to_wait(&journal->j_wait_transaction_locked, &wait,
>> 			TASK_UNINTERRUPTIBLE);
>> -	need_to_start = !tid_geq(journal->j_commit_request, tid);
>> +	if (!atomic_read(&journal->j_running_transaction->t_updates))
>> +		need_to_start = !tid_geq(journal->j_commit_request, tid);
>> 	read_unlock(&journal->j_state_lock);
>> 	if (need_to_start)
>> 		jbd2_log_start_commit(journal, tid);
>> --
>> 2.19.1
>> 
>> 
> --
> Jan Kara <jack@xxxxxxxx>
> SUSE Labs, CR


Cheers, Andreas

Attachment: signature.asc
Description: Message signed with OpenPGP