[sheepdog] [PATCH v3 3/7] sheep: teach journal file to handle epoch and config write

[Liu Yuan]

On 04/05/2013 04:33 PM, Hitoshi Mitake wrote:
> At Thu,  4 Apr 2013 20:49:38 +0800,
> Liu Yuan wrote:
>>
>> From: Liu Yuan <tailai.ly at taobao.com>
>>
>> journal_write_epoch and journal_write_confi is a placeholder for future use.
> 
> This commit message is obsolete. You should update it.
> 
> BTW, confi for expressing config files would be confusing. I think
> config is more suitable.
> 
>>  static int replay_journal_entry(struct journal_descriptor *jd)
>>  {
>>  	char path[PATH_MAX];
>> @@ -136,20 +165,17 @@ static int replay_journal_entry(struct journal_descriptor *jd)
>>  	void *buf;
>>  	char *p = (char *)jd;
>>  
>> -	sd_dprintf("%"PRIx64", size %"PRIu64", off %"PRIu64", %d", jd->oid,
>> -		   jd->size, jd->offset, jd->create);
>> -
>>  	if (jd->create)
>>  		flags |= O_CREAT;
>> -	snprintf(path, sizeof(path), "%s/%016" PRIx64, get_object_path(jd->oid),
>> -		 jd->oid);
>> +
>> +	journal_get_path(jd, path);
>>  	fd = open(path, flags, def_fmode);
>>  	if (fd < 0) {
>>  		sd_eprintf("open %m");
>>  		return -1;
>>  	}
>>  
>> -	if (jd->create) {
>> +	if (jd->create && jd->flag == JF_STORE) {
>>  		ret = prealloc(fd, get_objsize(jd->oid));
>>  		if (ret < 0)
>>  			goto out;
>> @@ -315,22 +341,15 @@ retry:
>>  		panic("%s", strerror(err));
>>  }
> 
> replay_journal_entry() lacks free(buf);
> 

Good catch.

>>  }
>> +
>> +int journal_write_store(uint64_t oid, const char *buf, size_t size,
>> +			off_t offset, bool create)
>> +{
>> +	struct journal_descriptor jd = {
>> +		.magic = JOURNAL_DESC_MAGIC,
>> +		.flag = JF_STORE,
>> +		.offset = offset,
>> +		.size = size,
>> +		.oid = oid,
> 
> The compile error still rises.

Oops, forgot to move this line.


> 
>>  	if (uatomic_is_true(&sys->use_journal) &&
>> -	    journal_file_write(oid, iocb->buf, iocb->length, iocb->offset,
>> -			       false)
>> +	    journal_write_store(oid, iocb->buf, iocb->length, iocb->offset,
>> +				false)
>>  	    != SD_RES_SUCCESS) {
>>  		sd_eprintf("turn off journaling");
>>  		uatomic_set_false(&sys->use_journal);
> 
> Is this behavior, turning off journaling and continuing, suitable?
> From my perspective, at least a behavior which causes death of sheep
> when writing journal fails should be supported as an optional one.
> 

Your advice is reasonable, but the error handling would be better
considered when journaling is set as default. Let's keep this patch set
small.

Thanks,
Yuan