| Message ID | 20240520131831.2910790-1-yi.zhang@huaweicloud.com (mailing list archive) |
|---|---|
| State | New |
| Headers | show |
| Series | [v2] jbd2: speed up jbd2_transaction_committed() | expand |
On Mon 20-05-24 21:18:31, Zhang Yi wrote: > From: Zhang Yi <yi.zhang@huawei.com> > > jbd2_transaction_committed() is used to check whether a transaction with > the given tid has already committed, it holds j_state_lock in read mode > and check the tid of current running transaction and committing > transaction, but holding the j_state_lock is expensive. > > We have already stored the sequence number of the most recently > committed transaction in journal t->j_commit_sequence, we could do this > check by comparing it with the given tid instead. If the given tid isn't > smaller than j_commit_sequence, we can ensure that the given transaction > has been committed. That way we could drop the expensive lock and > achieve about 10% ~ 20% performance gains in concurrent DIOs on may > virtual machine with 100G ramdisk. > > fio -filename=/mnt/foo -direct=1 -iodepth=10 -rw=$rw -ioengine=libaio \ > -bs=4k -size=10G -numjobs=10 -runtime=60 -overwrite=1 -name=test \ > -group_reporting > > Before: > overwrite IOPS=88.2k, BW=344MiB/s > read IOPS=95.7k, BW=374MiB/s > rand overwrite IOPS=98.7k, BW=386MiB/s > randread IOPS=102k, BW=397MiB/s > > After: > overwrite IOPS=105k, BW=410MiB/s > read IOPS=112k, BW=436MiB/s > rand overwrite IOPS=104k, BW=404MiB/s > randread IOPS=111k, BW=432MiB/s > > CC: Dave Chinner <david@fromorbit.com> > Suggested-by: Dave Chinner <david@fromorbit.com> > Link: https://lore.kernel.org/linux-ext4/ZjILCPNZRHeazSqV@dread.disaster.area/ > Signed-off-by: Zhang Yi <yi.zhang@huawei.com> Thanks! The patch looks good. Feel free to add: Reviewed-by: Jan Kara <jack@suse.cz> Honza > --- > v1->v2: > - Add READ_ONCE and WRITE_ONCE to access ->j_commit_sequence > concurrently. > - Keep the jbd2_transaction_committed() helper. > > fs/jbd2/commit.c | 2 +- > fs/jbd2/journal.c | 12 +----------- > 2 files changed, 2 insertions(+), 12 deletions(-) > > diff --git a/fs/jbd2/commit.c b/fs/jbd2/commit.c > index 5e122586e06e..8244cab17688 100644 > --- a/fs/jbd2/commit.c > +++ b/fs/jbd2/commit.c > @@ -1108,7 +1108,7 @@ void jbd2_journal_commit_transaction(journal_t *journal) > > commit_transaction->t_state = T_COMMIT_CALLBACK; > J_ASSERT(commit_transaction == journal->j_committing_transaction); > - journal->j_commit_sequence = commit_transaction->t_tid; > + WRITE_ONCE(journal->j_commit_sequence, commit_transaction->t_tid); > journal->j_committing_transaction = NULL; > commit_time = ktime_to_ns(ktime_sub(ktime_get(), start_time)); > > diff --git a/fs/jbd2/journal.c b/fs/jbd2/journal.c > index b6c114c11b97..cc586e3c4ee1 100644 > --- a/fs/jbd2/journal.c > +++ b/fs/jbd2/journal.c > @@ -789,17 +789,7 @@ EXPORT_SYMBOL(jbd2_fc_end_commit_fallback); > /* Return 1 when transaction with given tid has already committed. */ > int jbd2_transaction_committed(journal_t *journal, tid_t tid) > { > - int ret = 1; > - > - read_lock(&journal->j_state_lock); > - if (journal->j_running_transaction && > - journal->j_running_transaction->t_tid == tid) > - ret = 0; > - if (journal->j_committing_transaction && > - journal->j_committing_transaction->t_tid == tid) > - ret = 0; > - read_unlock(&journal->j_state_lock); > - return ret; > + return tid_geq(READ_ONCE(journal->j_commit_sequence), tid); > } > EXPORT_SYMBOL(jbd2_transaction_committed); > > -- > 2.39.2 >
Zhang Yi <yi.zhang@huaweicloud.com> writes: > From: Zhang Yi <yi.zhang@huawei.com> > > jbd2_transaction_committed() is used to check whether a transaction with > the given tid has already committed, it holds j_state_lock in read mode > and check the tid of current running transaction and committing > transaction, but holding the j_state_lock is expensive. > > We have already stored the sequence number of the most recently > committed transaction in journal t->j_commit_sequence, we could do this > check by comparing it with the given tid instead. If the given tid isn't > smaller than j_commit_sequence, we can ensure that the given transaction > has been committed. That way we could drop the expensive lock and > achieve about 10% ~ 20% performance gains in concurrent DIOs on may > virtual machine with 100G ramdisk. > > fio -filename=/mnt/foo -direct=1 -iodepth=10 -rw=$rw -ioengine=libaio \ > -bs=4k -size=10G -numjobs=10 -runtime=60 -overwrite=1 -name=test \ > -group_reporting > > Before: > overwrite IOPS=88.2k, BW=344MiB/s > read IOPS=95.7k, BW=374MiB/s > rand overwrite IOPS=98.7k, BW=386MiB/s > randread IOPS=102k, BW=397MiB/s > > After: > overwrite IOPS=105k, BW=410MiB/s > read IOPS=112k, BW=436MiB/s > rand overwrite IOPS=104k, BW=404MiB/s > randread IOPS=111k, BW=432MiB/s I was surprised to see that even the read and randread performance is improved with this patch which should theoritically only impact write workloads given based on such checks we are just setting IOMAP_F_DIRTY flag to report to iomap. But then I came across these two patches [1] [2]. It seems the change [1] to set IOMAP_F_DIRTY was initially only done for IOMAP_WRITE path. But patch [2] moved some logic to fs-dax core and filesystems were left to always just reports if there is any dirty metadata, irrespective of reads or writes. [1]: https://lore.kernel.org/all/20171101153648.30166-17-jack@suse.cz/ [2]: https://lore.kernel.org/all/151062258598.8554.8157038002895095232.stgit@dwillia2-desk3.amr.corp.intel.com/ Ohh - could this patch be that reason of peformance regression when ext4 DIO moved to iomap? Should we CC: stable to when ext4 DIO was moved to iomap atleast - which I believe was v5.5? Looks good to me. Reviewed-by: Ritesh Harjani (IBM) <ritesh.list@gmail.com> > > CC: Dave Chinner <david@fromorbit.com> > Suggested-by: Dave Chinner <david@fromorbit.com> > Link: https://lore.kernel.org/linux-ext4/ZjILCPNZRHeazSqV@dread.disaster.area/ aah. This link is helpful too to understand the context. Thanks! > Signed-off-by: Zhang Yi <yi.zhang@huawei.com> > --- > v1->v2: > - Add READ_ONCE and WRITE_ONCE to access ->j_commit_sequence > concurrently. > - Keep the jbd2_transaction_committed() helper. > > fs/jbd2/commit.c | 2 +- > fs/jbd2/journal.c | 12 +----------- > 2 files changed, 2 insertions(+), 12 deletions(-) > > diff --git a/fs/jbd2/commit.c b/fs/jbd2/commit.c > index 5e122586e06e..8244cab17688 100644 > --- a/fs/jbd2/commit.c > +++ b/fs/jbd2/commit.c > @@ -1108,7 +1108,7 @@ void jbd2_journal_commit_transaction(journal_t *journal) > > commit_transaction->t_state = T_COMMIT_CALLBACK; > J_ASSERT(commit_transaction == journal->j_committing_transaction); > - journal->j_commit_sequence = commit_transaction->t_tid; > + WRITE_ONCE(journal->j_commit_sequence, commit_transaction->t_tid); > journal->j_committing_transaction = NULL; > commit_time = ktime_to_ns(ktime_sub(ktime_get(), start_time)); > > diff --git a/fs/jbd2/journal.c b/fs/jbd2/journal.c > index b6c114c11b97..cc586e3c4ee1 100644 > --- a/fs/jbd2/journal.c > +++ b/fs/jbd2/journal.c > @@ -789,17 +789,7 @@ EXPORT_SYMBOL(jbd2_fc_end_commit_fallback); > /* Return 1 when transaction with given tid has already committed. */ > int jbd2_transaction_committed(journal_t *journal, tid_t tid) > { > - int ret = 1; > - > - read_lock(&journal->j_state_lock); > - if (journal->j_running_transaction && > - journal->j_running_transaction->t_tid == tid) > - ret = 0; > - if (journal->j_committing_transaction && > - journal->j_committing_transaction->t_tid == tid) > - ret = 0; > - read_unlock(&journal->j_state_lock); > - return ret; > + return tid_geq(READ_ONCE(journal->j_commit_sequence), tid); > } > EXPORT_SYMBOL(jbd2_transaction_committed); > > -- > 2.39.2
On Mon, 20 May 2024 21:18:31 +0800, Zhang Yi wrote: > jbd2_transaction_committed() is used to check whether a transaction with > the given tid has already committed, it holds j_state_lock in read mode > and check the tid of current running transaction and committing > transaction, but holding the j_state_lock is expensive. > > We have already stored the sequence number of the most recently > committed transaction in journal t->j_commit_sequence, we could do this > check by comparing it with the given tid instead. If the given tid isn't > smaller than j_commit_sequence, we can ensure that the given transaction > has been committed. That way we could drop the expensive lock and > achieve about 10% ~ 20% performance gains in concurrent DIOs on may > virtual machine with 100G ramdisk. > > [...] Applied, thanks! [1/1] jbd2: speed up jbd2_transaction_committed() commit: 7c73ddb7589fb8ddb1136b6306dfb72089c81511 Best regards,
diff --git a/fs/jbd2/commit.c b/fs/jbd2/commit.c index 5e122586e06e..8244cab17688 100644 --- a/fs/jbd2/commit.c +++ b/fs/jbd2/commit.c @@ -1108,7 +1108,7 @@ void jbd2_journal_commit_transaction(journal_t *journal) commit_transaction->t_state = T_COMMIT_CALLBACK; J_ASSERT(commit_transaction == journal->j_committing_transaction); - journal->j_commit_sequence = commit_transaction->t_tid; + WRITE_ONCE(journal->j_commit_sequence, commit_transaction->t_tid); journal->j_committing_transaction = NULL; commit_time = ktime_to_ns(ktime_sub(ktime_get(), start_time)); diff --git a/fs/jbd2/journal.c b/fs/jbd2/journal.c index b6c114c11b97..cc586e3c4ee1 100644 --- a/fs/jbd2/journal.c +++ b/fs/jbd2/journal.c @@ -789,17 +789,7 @@ EXPORT_SYMBOL(jbd2_fc_end_commit_fallback); /* Return 1 when transaction with given tid has already committed. */ int jbd2_transaction_committed(journal_t *journal, tid_t tid) { - int ret = 1; - - read_lock(&journal->j_state_lock); - if (journal->j_running_transaction && - journal->j_running_transaction->t_tid == tid) - ret = 0; - if (journal->j_committing_transaction && - journal->j_committing_transaction->t_tid == tid) - ret = 0; - read_unlock(&journal->j_state_lock); - return ret; + return tid_geq(READ_ONCE(journal->j_commit_sequence), tid); } EXPORT_SYMBOL(jbd2_transaction_committed);