Re: [PATCH v6 09/10] ext4: temporarily elevate commit thread priority

[Jan Kara <jack@suse.cz>]

On Wed 29-05-24 01:20:02, Harshad Shirwadkar wrote:
> Unlike JBD2 based full commits, there is no dedicated journal thread
> for fast commits. Thus to reduce scheduling delays between IO
> submission and completion, temporarily elevate the committer thread's
> priority to match the configured priority of the JBD2 journal
> thread.
> 
> Signed-off-by: Harshad Shirwadkar <harshadshirwadkar@gmail.com>

This makes some sense although I'd note that io priority is getting less
and less use these days since IO is now mostly controlled through cgroup
controllers and they don't give a damn about IO priority. E.g. blk-iocost
controller uses bio_issue_as_root_blkg() (which boils down to bio->bi_opf &
(REQ_META | REQ_SWAP)) to determine whether it should avoid throttling IOs
to avoid priority inversion (exactly the case of fast-commit). So I think
properly annotating journal IO with REQ_META will bring much more tangible
benefit in common configurations that bother to control IO and then this
needn't be even needed. But I'm not really opposed either so feel free to
add:

Reviewed-by: Jan Kara <jack@suse.cz>

								Honza

> ---
>  fs/ext4/ext4.h        |  4 +++-
>  fs/ext4/fast_commit.c | 13 +++++++++++++
>  fs/ext4/super.c       |  5 ++---
>  3 files changed, 18 insertions(+), 4 deletions(-)
> 
> diff --git a/fs/ext4/ext4.h b/fs/ext4/ext4.h
> index 3721daea2890..d52df8a85271 100644
> --- a/fs/ext4/ext4.h
> +++ b/fs/ext4/ext4.h
> @@ -2287,10 +2287,12 @@ static inline int ext4_forced_shutdown(struct super_block *sb)
>  #define EXT4_DEFM_NODELALLOC	0x0800
>  
>  /*
> - * Default journal batch times
> + * Default journal batch times and ioprio.
>   */
>  #define EXT4_DEF_MIN_BATCH_TIME	0
>  #define EXT4_DEF_MAX_BATCH_TIME	15000 /* 15ms */
> +#define EXT4_DEF_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
> +
>  
>  /*
>   * Minimum number of groups in a flexgroup before we separate out
> diff --git a/fs/ext4/fast_commit.c b/fs/ext4/fast_commit.c
> index 35c89bee452c..55a13d3ff681 100644
> --- a/fs/ext4/fast_commit.c
> +++ b/fs/ext4/fast_commit.c
> @@ -1205,6 +1205,7 @@ int ext4_fc_commit(journal_t *journal, tid_t commit_tid)
>  	int subtid = atomic_read(&sbi->s_fc_subtid);
>  	int status = EXT4_FC_STATUS_OK, fc_bufs_before = 0;
>  	ktime_t start_time, commit_time;
> +	int old_ioprio, journal_ioprio;
>  
>  	if (!test_opt2(sb, JOURNAL_FAST_COMMIT))
>  		return jbd2_complete_transaction(journal, commit_tid);
> @@ -1212,6 +1213,7 @@ int ext4_fc_commit(journal_t *journal, tid_t commit_tid)
>  	trace_ext4_fc_commit_start(sb, commit_tid);
>  
>  	start_time = ktime_get();
> +	old_ioprio = get_current_ioprio();
>  
>  restart_fc:
>  	ret = jbd2_fc_begin_commit(journal, commit_tid);
> @@ -1242,6 +1244,15 @@ int ext4_fc_commit(journal_t *journal, tid_t commit_tid)
>  		goto fallback;
>  	}
>  
> +	/*
> +	 * Now that we know that this thread is going to do a fast commit,
> +	 * elevate the priority to match that of the journal thread.
> +	 */
> +	if (journal->j_task->io_context)
> +		journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
> +	else
> +		journal_ioprio = EXT4_DEF_JOURNAL_IOPRIO;
> +	set_task_ioprio(current, journal_ioprio);
>  	fc_bufs_before = (sbi->s_fc_bytes + bsize - 1) / bsize;
>  	ret = ext4_fc_perform_commit(journal);
>  	if (ret < 0) {
> @@ -1256,6 +1267,7 @@ int ext4_fc_commit(journal_t *journal, tid_t commit_tid)
>  	}
>  	atomic_inc(&sbi->s_fc_subtid);
>  	ret = jbd2_fc_end_commit(journal);
> +	set_task_ioprio(current, old_ioprio);
>  	/*
>  	 * weight the commit time higher than the average time so we
>  	 * don't react too strongly to vast changes in the commit time
> @@ -1265,6 +1277,7 @@ int ext4_fc_commit(journal_t *journal, tid_t commit_tid)
>  	return ret;
>  
>  fallback:
> +	set_task_ioprio(current, old_ioprio);
>  	ret = jbd2_fc_end_commit_fallback(journal);
>  	ext4_fc_update_stats(sb, status, 0, 0, commit_tid);
>  	return ret;
> diff --git a/fs/ext4/super.c b/fs/ext4/super.c
> index 77173ec91e49..18d9d2631559 100644
> --- a/fs/ext4/super.c
> +++ b/fs/ext4/super.c
> @@ -1833,7 +1833,6 @@ static const struct fs_parameter_spec ext4_param_specs[] = {
>  	{}
>  };
>  
> -#define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
>  
>  #define MOPT_SET	0x0001
>  #define MOPT_CLEAR	0x0002
> @@ -5211,7 +5210,7 @@ static int __ext4_fill_super(struct fs_context *fc, struct super_block *sb)
>  
>  	/* Set defaults for the variables that will be set during parsing */
>  	if (!(ctx->spec & EXT4_SPEC_JOURNAL_IOPRIO))
> -		ctx->journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
> +		ctx->journal_ioprio = EXT4_DEF_JOURNAL_IOPRIO;
>  
>  	sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
>  	sbi->s_sectors_written_start =
> @@ -6471,7 +6470,7 @@ static int __ext4_remount(struct fs_context *fc, struct super_block *sb)
>  			ctx->journal_ioprio =
>  				sbi->s_journal->j_task->io_context->ioprio;
>  		else
> -			ctx->journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
> +			ctx->journal_ioprio = EXT4_DEF_JOURNAL_IOPRIO;
>  
>  	}
>  
> -- 
> 2.45.1.288.g0e0cd299f1-goog
> 
> 
-- 
Jan Kara <jack@suse.com>
SUSE Labs, CR