Re: [PATCH 1/5] btrfs: add flags to give an hint to the chunk allocator

[Goffredo Baroncelli <kreijack@libero.it>]

On 22/03/2022 18.56, Josef Bacik wrote:
> On Sun, Mar 06, 2022 at 07:14:39PM +0100, Goffredo Baroncelli wrote:
>> From: Goffredo Baroncelli <kreijack@inwind.it>
>>
>> Add the following flags to give an hint about which chunk should be
>> allocated in which disk:
>>
>> - BTRFS_DEV_ALLOCATION_HINT_DATA_PREFERRED
>>    preferred for data chunk, but metadata chunk allowed
>> - BTRFS_DEV_ALLOCATION_HINT_METADATA_PREFERRED
>>    preferred for metadata chunk, but data chunk allowed
>> - BTRFS_DEV_ALLOCATION_HINT_METADATA_ONLY
>>    only metadata chunk allowed
>> - BTRFS_DEV_ALLOCATION_HINT_DATA_ONLY
>>    only data chunk allowed
>>
>> Signed-off-by: Goffredo Baroncelli <kreijack@inwind.it>
>> ---
>>   include/uapi/linux/btrfs_tree.h | 16 ++++++++++++++++
>>   1 file changed, 16 insertions(+)
>>
>> diff --git a/include/uapi/linux/btrfs_tree.h b/include/uapi/linux/btrfs_tree.h
>> index b069752a8ecf..e0d842c2e616 100644
>> --- a/include/uapi/linux/btrfs_tree.h
>> +++ b/include/uapi/linux/btrfs_tree.h
>> @@ -389,6 +389,22 @@ struct btrfs_key {
>>   	__u64 offset;
>>   } __attribute__ ((__packed__));
>>   
>> +/* dev_item.type */
>> +
>> +/* btrfs chunk allocation hint */
>> +#define BTRFS_DEV_ALLOCATION_HINT_BIT_COUNT	2
>> +/* btrfs chunk allocation hint mask */
>> +#define BTRFS_DEV_ALLOCATION_HINT_MASK	\
>> +	((1 << BTRFS_DEV_ALLOCATION_HINT_BIT_COUNT) - 1)
>> +/* preferred data chunk, but metadata chunk allowed */
>> +#define BTRFS_DEV_ALLOCATION_HINT_DATA_PREFERRED	(0ULL)
> 
> Actually don't we have 0 set on type already?  So this will make all existing
> file systems have DATA_PREFERRED, when in reality they may not want that
> behavior.  So should we start at 1?  Thanks,

Yes, the default is 0 (now DATA_PREFERRED).
If we have all the disks set to DATA_PREFERRED (or METADATA_PREFERRED), all the disks
have the same priority so the chunks allocator works as usual.

If we have DATA_PREFERRED=1, it is not clear to me which would be the expected behavior when the allocator has to choice one of the followings disks:
- TYPE=0
- DATA_PREFERRED
- DATA_ONLY

It should ignore TYPE=0 ? or it should block the 'allocation_hint' policy ? Or TYPE=0 should have the lowest (or highest) priority ?

DATA_PREFERRED to me seems a safe default, because at worst we have only missing performance penalty (i.e. it is a faster disk where we should put METADATA)

BR

> 
> Josef


-- 
gpg @keyserver.linux.it: Goffredo Baroncelli <kreijackATinwind.it>
Key fingerprint BBF5 1610 0B64 DAC6 5F7D  17B2 0EDA 9B37 8B82 E0B5