[sheepdog] [PATCH] sheep: exit when vdi bitmap collection is failed

[Saeki Masaki]

(2014/10/20 17:28), Hitoshi Mitake wrote:
> In our test environment, we found an inconsistency of vdi bitmap
> caused by failed colleciton process. The inconsistency is really
> critical so calling panic() is much safer.
> 
> Cc: Saeki Masaki <saeki.masaki at po.ntts.co.jp>
> Signed-off-by: Hitoshi Mitake <mitake.hitoshi at lab.ntt.co.jp>
> ---
>   sheep/group.c | 33 ++++++++++++++++++++++++---------
>   1 file changed, 24 insertions(+), 9 deletions(-)

Looks good to me.
Reviewed-by: Saeki Masaki <saeki.masaki at po.ntts.co.jp>

Regards,


> 
> diff --git a/sheep/group.c b/sheep/group.c
> index 14c106a..f1104c1 100644
> --- a/sheep/group.c
> +++ b/sheep/group.c
> @@ -525,9 +525,19 @@ static void do_get_vdis(struct work *work)
>   		sd_debug("try to get vdi bitmap from %s",
>   			 node_to_str(&w->joined));
>   		ret = get_vdis_from(&w->joined);
> -		if (ret != SD_RES_SUCCESS)
> -			sd_alert("failed to get vdi bitmap from %s",
> -				 node_to_str(&w->joined));
> +		if (ret != SD_RES_SUCCESS) {
> +			if (sys->cinfo.status == SD_STATUS_OK)
> +				/*
> +				 * SD_STATUS_OK means enough zones are gathered,
> +				 * so failed vdi bitmap collection isn't
> +				 * critical
> +				 */
> +				sd_alert("failed to get vdi bitmap from %s",
> +					 node_to_str(&w->joined));
> +			else
> +				panic("failed to get vdi bitmap from %s",
> +				      node_to_str(&w->joined));
> +		}
>   		return;
>   	}
>   
> @@ -538,12 +548,17 @@ static void do_get_vdis(struct work *work)
>   
>   		sd_debug("try to get vdi bitmap from %s", node_to_str(n));
>   		ret = get_vdis_from(n);
> -		if (ret != SD_RES_SUCCESS) {
> -			/* try to read from another node */
> -			sd_alert("failed to get vdi bitmap from %s",
> -				 node_to_str(n));
> -			continue;
> -		}
> +		if (ret != SD_RES_SUCCESS)
> +			/*
> +			 * It means this sheep has missing vdi bitmap, and
> +			 * reading bitmap from other sheep cannot be guaranteed
> +			 * to success.
> +			 *
> +			 * Inconsistency of vdi bitmap between nodes is deathly
> +			 * critical, so dying here is safer.
> +			 */
> +			panic("failed to get vdi bitmap from %s",
> +			      node_to_str(n));
>   
>   		/*
>   		 * TODO: If the target node has a valid vdi bitmap (the node has
>