[sheepdog] [PATCH] sheep: propagate deletion information via SD_OP_GET_VDI_COPIES

[Saeki Masaki]

Hi, Mitake-san.

Thank you to improve the process that did not have enough of consideration.

I tried the check this patch, and node additional test.
it's very desirable results were obtained.
and patch looks good to me.

Then I'll shows the results.

tested procedure is below.
#sorry for has become a different test condition,
#but I think that it is sufficient conditions for.

1. create 1000 snapshot
2. delete  999 snapshot

3. add new node at -p 7003
  # sheep -p 7003 -z 3 /var/lib/sheepdog/data3
( wait for complete recovery )

4. test vdi list to original node
  # dog vdi list -p 7000 -T
===> [test-case-1]
5. test vdi list to additional node
  # dog vdi list -p 7003 -T
===> [test-case-2]

before patch applied
[test-case-1]  Elapsed time: 0.083 seconds
[test-case-2] Elapsed time: 1.134 seconds

after patch applied
[test-case-1]  Elapsed time: 0.070seconds
[test-case-2]  Elapsed time: 0.167 seconds

Reviewed-by: Masaki Saeki <saeki.masaki at po.ntts.co.jp>
Tested-by: Masaki Saeki <saeki.masaki at po.ntts.co.jp>

Best regards,
Saeki.
(2014/11/26 16:54), Hitoshi Mitake wrote:
> At Wed, 26 Nov 2014 16:53:53 +0900,
> Hitoshi Mitake wrote:
>>
>> The commit f68feab7edc0de introduced significant optimization for dog
>> vdi list. But the deleted vid bitmap cannot be copied to newly joining
>> sheep. This patch let new sheeps copy the information via
>> SD_OP_GET_VDI_COPIES during joining process.
>>
>> Cc: Masaki Saeki <saeki.masaki at po.ntts.co.jp>
>> Signed-off-by: Hitoshi Mitake <mitake.hitoshi at lab.ntt.co.jp>
>> ---
>>   include/internal_proto.h |  1 +
>>   sheep/group.c            |  2 ++
>>   sheep/ops.c              |  4 +++-
>>   sheep/sheep_priv.h       |  1 +
>>   sheep/vdi.c              | 18 ++++++++++++++++++
>>   5 files changed, 25 insertions(+), 1 deletion(-)
>
> Saeki-san, could you test and review the patch if you have time?
>
> Thanks,
> Hitoshi
>
>>
>> diff --git a/include/internal_proto.h b/include/internal_proto.h
>> index faed3f6..ad2f3ed 100644
>> --- a/include/internal_proto.h
>> +++ b/include/internal_proto.h
>> @@ -353,6 +353,7 @@ struct vdi_state {
>>   	uint32_t vid;
>>   	uint8_t nr_copies;
>>   	uint8_t snapshot;
>> +	uint8_t deleted;
>>   	uint8_t copy_policy;
>>
>>   	uint32_t lock_state;
>> diff --git a/sheep/group.c b/sheep/group.c
>> index a45904e..2b98a9b 100644
>> --- a/sheep/group.c
>> +++ b/sheep/group.c
>> @@ -507,6 +507,8 @@ retry:
>>   	count = rsp->data_length / sizeof(*vs);
>>   	for (i = 0; i < count; i++) {
>>   		atomic_set_bit(vs[i].vid, sys->vdi_inuse);
>> +		if (vs[i].deleted)
>> +			atomic_set_bit(vs[i].vid, sys->vdi_deleted);
>>   		add_vdi_state(vs[i].vid, vs[i].nr_copies, vs[i].snapshot,
>>   			      vs[i].copy_policy);
>>   	}
>> diff --git a/sheep/ops.c b/sheep/ops.c
>> index 4b54780..e645e04 100644
>> --- a/sheep/ops.c
>> +++ b/sheep/ops.c
>> @@ -188,8 +188,10 @@ static int post_cluster_del_vdi(const struct sd_req *req, struct sd_rsp *rsp,
>>   	struct cache_deletion_work *dw;
>>   	int ret = rsp->result;
>>
>> -	if (ret == SD_RES_SUCCESS)
>> +	if (ret == SD_RES_SUCCESS) {
>>   		atomic_set_bit(vid, sys->vdi_deleted);
>> +		vdi_mark_deleted(vid);
>> +	}
>>
>>   	if (!sys->enable_object_cache)
>>   		return ret;
>> diff --git a/sheep/sheep_priv.h b/sheep/sheep_priv.h
>> index 3e6adc2..5fc6b90 100644
>> --- a/sheep/sheep_priv.h
>> +++ b/sheep/sheep_priv.h
>> @@ -333,6 +333,7 @@ int vdi_exist(uint32_t vid);
>>   int vdi_create(const struct vdi_iocb *iocb, uint32_t *new_vid);
>>   int vdi_snapshot(const struct vdi_iocb *iocb, uint32_t *new_vid);
>>   int vdi_delete(const struct vdi_iocb *iocb, struct request *req);
>> +void vdi_mark_deleted(uint32_t vid);
>>   int vdi_lookup(const struct vdi_iocb *iocb, struct vdi_info *info);
>>   void clean_vdi_state(void);
>>   int sd_delete_vdi(const char *name);
>> diff --git a/sheep/vdi.c b/sheep/vdi.c
>> index b03421a..1c8fb36 100644
>> --- a/sheep/vdi.c
>> +++ b/sheep/vdi.c
>> @@ -15,6 +15,7 @@ struct vdi_state_entry {
>>   	uint32_t vid;
>>   	unsigned int nr_copies;
>>   	bool snapshot;
>> +	bool deleted;
>>   	uint8_t copy_policy;
>>   	struct rb_node node;
>>
>> @@ -248,6 +249,7 @@ static struct vdi_state *fill_vdi_state_list_with_alloc(int *result_nr)
>>   		vs[i].vid = entry->vid;
>>   		vs[i].nr_copies = entry->nr_copies;
>>   		vs[i].snapshot = entry->snapshot;
>> +		vs[i].deleted = entry->deleted;
>>   		vs[i].copy_policy = entry->copy_policy;
>>   		vs[i].lock_state = entry->lock_state;
>>   		vs[i].lock_owner = entry->owner;
>> @@ -1576,6 +1578,22 @@ out:
>>   	return ret;
>>   }
>>
>> +void vdi_mark_deleted(uint32_t vid)
>> +{
>> +	struct vdi_state_entry *entry;
>> +
>> +	sd_write_lock(&vdi_state_lock);
>> +	entry = vdi_state_search(&vdi_state_root, vid);
>> +	if (!entry) {
>> +		sd_err("VID: %"PRIx32" not found", vid);
>> +		goto out;
>> +	}
>> +
>> +	entry->deleted = true;
>> +out:
>> +	sd_rw_unlock(&vdi_state_lock);
>> +}
>> +
>>   /* Calculate a vdi attribute id from sheepdog_vdi_attr. */
>>   static uint32_t hash_vdi_attr(const struct sheepdog_vdi_attr *attr)
>>   {
>> --
>> 1.8.3.2
>>
>
--