[Sheepdog] [PATCH] sheep: get vdi bitmap correctly in join phase

[MORITA Kazutaka]

At Sun, 11 Sep 2011 00:49:20 +0800,
Liu Yuan wrote:
> 
> From: Liu Yuan <tailai.ly at taobao.com>
> 
> After the new node joins, command *collie-vdi-list*
> cannot list any vdi information, that are created
> before this node is joined. This is because in the
> join phase, get_vdi_bitmap_from_all() cannot get corret
> vdi bitmap *if* there is no node in sd_node_list.
> This patch moves this function after the function
> which updates the sd_node_list.
> 
> With this patch, collie-vdi-list works as expected in
> newly added nodes.
> 
> Signed-off-by: Liu Yuan <tailai.ly at taobao.com>
> ---
>  sheep/group.c |   13 ++-----------
>  1 files changed, 2 insertions(+), 11 deletions(-)
> 
> diff --git a/sheep/group.c b/sheep/group.c
> index eb0c4e2..22c4f66 100644
> --- a/sheep/group.c
> +++ b/sheep/group.c
> @@ -863,17 +863,6 @@ static void __sd_deliver(struct cpg_event *cevent)
>  			break;
>  		}
>  	}
> -
> -	if (m->state == DM_FIN) {
> -		switch (m->op) {
> -		case SD_MSG_JOIN:
> -			if (((struct join_message *)m)->cluster_status == SD_STATUS_OK)
> -				if (sys->status != SD_STATUS_OK)
> -					get_vdi_bitmap_from_all();
> -			break;
> -		}
> -	}
> -
>  }
>  
>  static void send_join_response(struct work_deliver *w)
> @@ -902,6 +891,8 @@ static void __sd_deliver_done(struct cpg_event *cevent)
>  		switch (m->op) {
>  		case SD_MSG_JOIN:
>  			update_cluster_info((struct join_message *)m);
> +			if (((struct join_message *)m)->cluster_status == SD_STATUS_OK)
> +					get_vdi_bitmap_from_all();
>  			break;
>  		case SD_MSG_LEAVE:
>  			node = find_node(&sys->sd_node_list, m->nodeid, m->pid);

Thanks for your contribution, but this doesn't work.

We cannot sleep in __sd_deliver_done (the main thread), so we cannot
call get_vdi_bitmap_from_all() in it; if all the sheep daemons call
get_vdi_bitmap_from_all() at the same time, the cluster will stuck.

I think the right way is doing the followings in __sd_deliver (the
worker thread):

 - call get_vdi_bitmap_from_all()
 - extract a new node from join_message, and get a vdi bitmap from the node


Thanks,

Kazutaka