Re: Safety of raid1 vs raid10

[Scoopta <mlist@scoopta.email>]

On 12/9/24 9:12 PM, Zygo Blaxell wrote:

> On Mon, Dec 09, 2024 at 06:44:14PM -0800, Scoopta wrote:
>> Ok, this is what I thought but I have seen so many articles that imply
>> raid10 groups drives and I have found no evidence for that. I was always
>> under the impression that neither raid 1 nor 10 grouped drives together
>> intentionally and any grouping was circumstantial.
> In traditional, non-btrfs raid1/raid10 setups, this can easily be the
> case, and people might be assuming this is also true on btrfs without
> checking.

Yeah I figured that's where the misconception was coming from but I 
wanted to verify that it was a misconception. It didn't make sense to me 
that btrfs would pair up raid10 drives but not raid1 drives. I figured 
they'd all behave the same way and that btrfs was free to organize data 
however it wanted provided the redundancy "on the tin", so to speak, was 
met.


> Traditional setups are much less flexible on space--you can't have
> different sizes of drives--but you can organize the drives such that
> each of them is entirely mirroring one other drive.  This can tolerate
> up to N/2 failures, as long as each failure only removes one drive from
> each pair--but it can also fail completely with only 2 drive failures,
> if they're both mirrors of the same stripe.
>
>> On 12/9/24 6:42 PM, Zygo Blaxell wrote:
>>> On Mon, Dec 09, 2024 at 06:26:24PM -0800, Scoopta wrote:
>>>> I've read online that btrfs raid10 is theoretically safer than raid1 because
>>>> raid10 groups drives together into mirrored pairs making the filesystem more
>>>> likely to successfully survive a multi-drive failure event. I can't find any
>>>> documentation that says this to be the case. Is it true that btrfs pairs
>>>> drives together for raid10 but not raid1, if this is the case what's the
>>>> reasoning for it?
>>> It is _possible_ for raid10 and raid1 to be arranged such that multiple drive
>>> losses are possible.  e.g. if all odd numbered devices are paired with all
>>> even numbered devices, then any odd numbered device can be lost and the
>>> filesystem still survives.
>>>
>>> This is not _guaranteed_, it is only _possible_.
>>>
>>> In a filesystem with an odd number of drives, or with drives of varying
>>> sizes, the block groups will only guarantee that one drive loss can be
>>> tolerated in each block group, in order to have the flexibility needed
>>> to fill all available space.  In such cases it is common that block
>>> groups are arranged in such a way that loss of any two drives will break
>>> the filesystem.