[sheepdog] [PATCH v2] recovery: fix incomplete recovery because of faulty oid scheduling

[MORITA Kazutaka]

At Mon, 29 Apr 2013 20:31:02 +0800,
Liu Yuan wrote:
> 
> From: Liu Yuan <tailai.ly at taobao.com>
> 
> tests/010 will fail sometimes to catch this bug.
> 
> When auto-recover is disabled, if we prepare_schedule_oid on the same oid
> multiple times, it will break finish_schedule_oid to wrongly squeeze some
> victim oid out of rw->oids array, then this node will never have a chance to
> recover the ejected oids.
> 
> Signed-off-by: Liu Yuan <tailai.ly at taobao.com>
> ---
>  sheep/recovery.c |   11 ++++-------
>  1 file changed, 4 insertions(+), 7 deletions(-)
> 
> diff --git a/sheep/recovery.c b/sheep/recovery.c
> index 23babe0..d1085aa 100644
> --- a/sheep/recovery.c
> +++ b/sheep/recovery.c
> @@ -237,18 +237,14 @@ static inline void prepare_schedule_oid(uint64_t oid)
>  				   oid);
>  			return;
>  		}
> -	/*
> -	 * When auto recovery is enabled, the oid is currently being
> -	 * recovered
> -	 */
> -	if (!sys->disable_recovery && rw->oids[rw->done] == oid)
> +	/* When auto-recovery is enabled, oid is currently being recovered */
> +	if (!rw->suspended && rw->oids[rw->done] == oid)
>  		return;

The comment message is no longer correct because sheep exits this
function here even when auto-recovery is disabled.

s/auto-recovery is enabled/recovery is not suspended/ ?

> +
>  	rw->nr_prio_oids++;
>  	rw->prio_oids = xrealloc(rw->prio_oids,
>  				 rw->nr_prio_oids * sizeof(uint64_t));
>  	rw->prio_oids[rw->nr_prio_oids - 1] = oid;
> -	resume_suspended_recovery();
> -
>  	sd_dprintf("%"PRIx64" nr_prio_oids %d", oid, rw->nr_prio_oids);
>  }
>  
> @@ -291,6 +287,7 @@ bool oid_in_recovery(uint64_t oid)
>  	}
>  
>  	prepare_schedule_oid(oid);
> +	resume_suspended_recovery();

Why do you move resume_suspended_recovery()?  If the oid is not
scheduled, we don't have to call this function.

Thanks,

Kazutaka