[Cluster-devel] [GFS2 PATCH] GFS2: Rework gfs2_logd daemon

[Steven Whitehouse]

Hi,


On 30/06/17 17:45, Bob Peterson wrote:
> Hi,
>
> This patch reorganizes some of the hokey logic in gfs2_logd.
> It also tries to only fetch the logd_secs tunable only once per
> second to avoid too many spinlock conflicts.
>
> Signed-off-by: Bob Peterson <rpeterso at redhat.com>
Again, please explain what the benefit is. The tunable should only every 
be accessed by gfs2_logd and by a userland process that is updating the 
tunable. If there is contention here, something is very wrong. The patch 
appears to remove the check for whether a journal or ail flush is 
required before gfs2_logd sleeps, so this appears that it would likely 
make performance worse,

Steve.

> ---
> diff --git a/fs/gfs2/log.c b/fs/gfs2/log.c
> index 32aa1f0..3f18db4 100644
> --- a/fs/gfs2/log.c
> +++ b/fs/gfs2/log.c
> @@ -915,15 +915,12 @@ int gfs2_logd(void *data)
>   	struct gfs2_sbd *sdp = data;
>   	unsigned long t = 1;
>   	DEFINE_WAIT(wait);
> -	bool did_flush;
> +	unsigned long tune_time = 0;
>   
>   	while (!kthread_should_stop()) {
> -
> -		did_flush = false;
>   		if (gfs2_jrnl_flush_reqd(sdp) || t == 0) {
>   			gfs2_ail1_empty(sdp);
>   			gfs2_log_flush(sdp, NULL, NORMAL_FLUSH);
> -			did_flush = true;
>   		}
>   
>   		if (gfs2_ail_flush_reqd(sdp)) {
> @@ -931,26 +928,22 @@ int gfs2_logd(void *data)
>   			gfs2_ail1_wait(sdp);
>   			gfs2_ail1_empty(sdp);
>   			gfs2_log_flush(sdp, NULL, NORMAL_FLUSH);
> -			did_flush = true;
>   		}
>   
> -		if (!gfs2_ail_flush_reqd(sdp) || did_flush)
> -			wake_up(&sdp->sd_log_waitq);
> -
> -		t = gfs2_tune_get(sdp, gt_logd_secs) * HZ;
> +		wake_up(&sdp->sd_log_waitq);
>   
>   		try_to_freeze();
>   
> -		do {
> -			prepare_to_wait(&sdp->sd_logd_waitq, &wait,
> -					TASK_INTERRUPTIBLE);
> -			if (!gfs2_ail_flush_reqd(sdp) &&
> -			    !gfs2_jrnl_flush_reqd(sdp) &&
> -			    !kthread_should_stop())
> -				t = schedule_timeout(t);
> -		} while(t && !gfs2_ail_flush_reqd(sdp) &&
> -			!gfs2_jrnl_flush_reqd(sdp) &&
> -			!kthread_should_stop());
> +		if (kthread_should_stop())
> +			break;
> +
> +		if (time_after(jiffies, tune_time + HZ)) {
> +			t = gfs2_tune_get(sdp, gt_logd_secs) * HZ;
> +			tune_time = jiffies;
> +		}
> +		prepare_to_wait(&sdp->sd_logd_waitq, &wait,
> +				TASK_INTERRUPTIBLE);
> +		t = schedule_timeout(t);
>   		finish_wait(&sdp->sd_logd_waitq, &wait);
>   	}
>