[sheepdog] [PATCH] sheep: pass the old and new vnode_info to recovery

[Liu Yuan]

On 05/30/2012 10:39 PM, Christoph Hellwig wrote:

> Signed-off-by: Christoph Hellwig <hch at lst.de>
> 
> diff --git a/sheep/group.c b/sheep/group.c
> index f3b7d1f..ea5a69e 100644
> --- a/sheep/group.c
> +++ b/sheep/group.c
> @@ -652,6 +652,7 @@ static void update_cluster_info(struct join_message *msg,
>  		struct sd_node *joined, struct sd_node *nodes, size_t nr_nodes)
>  {
>  	struct node *n, *t;
> +	struct vnode_info *old_vnode_info;
>  
>  	eprintf("status = %d, epoch = %d, %x, %d\n", msg->cluster_status,
>  		msg->epoch, msg->result, sys->join_finished);
> @@ -662,7 +663,7 @@ static void update_cluster_info(struct join_message *msg,
>  	if (!sys->join_finished)
>  		finish_join(msg, joined, nodes, nr_nodes);
>  
> -	put_vnode_info(current_vnode_info);
> +	old_vnode_info = current_vnode_info;
>  	current_vnode_info = alloc_vnode_info(nodes, nr_nodes);
>  
>  	if (msg->cluster_status == SD_STATUS_OK ||
> @@ -699,9 +700,11 @@ static void update_cluster_info(struct join_message *msg,
>  	if (sys_can_recover() && msg->inc_epoch) {
>  		list_for_each_entry_safe(n, t, &sys->leave_list, list)
>  			list_del(&n->list);
> -		start_recovery(sys->epoch);
> +		start_recovery(current_vnode_info, old_vnode_info);
>  	}
>  
> +	put_vnode_info(old_vnode_info);
> +
>  	if (sys_stat_halt()) {
>  		if (current_vnode_info->nr_zones >= sys->nr_copies)
>  			sys_stat_set(SD_STATUS_OK);
> @@ -1044,6 +1047,7 @@ void sd_join_handler(struct sd_node *joined, struct sd_node *members,
>  void sd_leave_handler(struct sd_node *left, struct sd_node *members,
>  		size_t nr_members)
>  {
> +	struct vnode_info *old_vnode_info;
>  	int i;
>  
>  	dprintf("leave %s\n", node_to_str(left));
> @@ -1053,14 +1057,15 @@ void sd_leave_handler(struct sd_node *left, struct sd_node *members,
>  	if (sys_stat_shutdown())
>  		return;
>  
> -	put_vnode_info(current_vnode_info);
> +	old_vnode_info = current_vnode_info;
>  	current_vnode_info = alloc_vnode_info(members, nr_members);
>  
>  	if (sys_can_recover()) {
>  		uatomic_inc(&sys->epoch);
>  		log_current_epoch();
> -		start_recovery(sys->epoch);
> +		start_recovery(current_vnode_info, old_vnode_info);
>  	}
> +	put_vnode_info(old_vnode_info);
>  
>  	if (sys_can_halt()) {
>  		if (current_vnode_info->nr_zones < sys->nr_copies)
> diff --git a/sheep/recovery.c b/sheep/recovery.c
> index 69432ee..9ee876f 100644
> --- a/sheep/recovery.c
> +++ b/sheep/recovery.c
> @@ -631,29 +631,6 @@ again:
>  	return 0;
>  }
>  
> -/* setup node list and virtual node list */
> -static int init_rw(struct recovery_work *rw)
> -{
> -	struct sd_node nodes[SD_MAX_NODES];
> -	int nr_nodes;
> -	uint32_t epoch = rw->epoch;
> -
> -	nr_nodes = epoch_log_read_nr(epoch, (char *)nodes, sizeof(nodes));
> -	if (nr_nodes <= 0) {
> -		eprintf("failed to read epoch log for epoch %"PRIu32"\n", epoch);
> -		return -1;
> -	}
> -	rw->cur_vnodes = alloc_vnode_info(nodes, nr_nodes);
> -
> -	nr_nodes = epoch_log_read_nr(epoch - 1, (char *)nodes, sizeof(nodes));
> -	if (nr_nodes <= 0) {
> -		eprintf("failed to read epoch log for epoch %"PRIu32"\n", epoch - 1);
> -		return -1;
> -	}
> -	rw->old_vnodes = alloc_vnode_info(nodes, nr_nodes);
> -	return 0;
> -}
> -
>  static void do_recovery_work(struct work *work)
>  {
>  	struct recovery_work *rw = container_of(work, struct recovery_work, work);
> @@ -663,8 +640,6 @@ static void do_recovery_work(struct work *work)
>  	if (!sys->nr_copies)
>  		return;
>  
> -	init_rw(rw);
> -
>  	if (fill_obj_list(rw) < 0) {
>  		eprintf("fatal recovery error\n");
>  		rw->count = 0;
> @@ -672,7 +647,7 @@ static void do_recovery_work(struct work *work)
>  	}
>  }
>  
> -int start_recovery(uint32_t epoch)
> +int start_recovery(struct vnode_info *cur_vnodes, struct vnode_info *old_vnodes)
>  {
>  	struct recovery_work *rw;
>  
> @@ -682,15 +657,18 @@ int start_recovery(uint32_t epoch)
>  
>  	rw->state = RW_INIT;
>  	rw->oids = malloc(1 << 20); /* FIXME */
> -	rw->epoch = epoch;
> +	rw->epoch = sys->epoch;
>  	rw->count = 0;
>  
> +	rw->cur_vnodes = grab_vnode_info(cur_vnodes);
> +	rw->old_vnodes = grab_vnode_info(old_vnodes);
> +
>  	rw->work.fn = do_recovery_work;
>  	rw->work.done = do_recover_main;
>  
>  	if (sd_store->begin_recover) {
>  		struct siocb iocb = { 0 };
> -		iocb.epoch = epoch;
> +		iocb.epoch = rw->epoch;
>  		sd_store->begin_recover(&iocb);
>  	}
>  
> diff --git a/sheep/sheep_priv.h b/sheep/sheep_priv.h
> index e5f8f44..c31a484 100644
> --- a/sheep/sheep_priv.h
> +++ b/sheep/sheep_priv.h
> @@ -284,7 +284,8 @@ int set_cluster_ctime(uint64_t ctime);
>  uint64_t get_cluster_ctime(void);
>  int get_obj_list(const struct sd_list_req *, struct sd_list_rsp *, void *);
>  
> -int start_recovery(uint32_t epoch);
> +int start_recovery(struct vnode_info *cur_vnodes,
> +	struct vnode_info *old_vnodes);
>  void resume_recovery_work(void);
>  int is_recoverying_oid(uint64_t oid);
>  int is_recovery_init(void);


Ummm, I was reading code hard to find the code in the patch set to do
this as one of your patch advertised so, didn't get any clue.

Applied, thanks.
Yuan