Hacker News Reader

Comment by andy800 | Hacker News Reader

This advice makes me extremely unpopular in NAS forums, but I strongly recommend you never use any other RAID option than RAID-1. When a disk goes bad, you have a single mirror disk that you can directly and immediately connect to your computer and recover from. With any of the NAS proprietary RAIDs, or something like RAID-5, you'll need to somehow hook up 3 or 5 disks to your computer to have any chance of a recovery.

The NAS gurus always say "RAID is not a backup", and it is true you should have additional backups. But no backup solution is perfect, none get updated every day, especially if you plan on keeping the backup off-site. For most users, the NAS is the backup. All I'm saying is that when a disk fails on your NAS, you'll be in panic mode and will want the easiest, most direct path to data recovery, and there is no RAID option in this scenario preferable than RAID-1.

View on HN

RAID-1, RAID-10, RAID-5, RAID-6, etc. none offer protection against an accidental `rm -rf`.

With that in mind, the NAS is not the backup. Once a file is deleted, it's gone.

With regard to recovering from a RAID-5 dual disk failure, typically a USB 2x sata toaster is enough to connect the failed disk and the replacement disk to a system that can run a ddrescue to clone the failed disk. Often times disks that are ejected from a RAID array still have enough life left to copy data off very slowly (over several tries/passes)

For backup, a simple solution is connecting very large external USB3 disk(s) and running daily rsync hardlink backups. There are several scripts to do this, and it's quite low effort to maintain after its set up.

Yep, exactly, all the home users know what a sata toaster is and a ddrescue and a rsync hardlink script. No problem!

That's why I would always recommend a Synology (or similar) NAS to non tech nerds (or tech nerds who want something that just works without putting effort into it), because backups are easily managed to various storage destinations and work reliable

> With that in mind, the NAS is not the backup. Once a file is deleted, it's gone.

Snapshots can be used to catch that failure mode, so I don't think this is universally true. Whether there actually are NASes that would have snapshottable file systems out of the box... Is there?

There is!

https://www.truenas.com/truenas-mini/

Synology supports snapshots if you choose to use BTRFS, but it also lets you enable recycle bins so you can recover things you deleted.

Note that not all the Synology products support BTRFS....

Right, but the recycle bin is universal, just to clarify.

For most people I agree with that. Heck, that's basically what I'm doing with my vdevs, since they're all mirrors.

But I think there's a semi-common exception: people who are using ZFS where they are maxing out the drives their enclosure can hold (or have a plan for that expansion). In that case, sure, fine, use raidz1 or raidz2 (with a hot or cold spare), because the rebuild commands are the same.

Mirroring is a good recommendation for simplicity, but you're half touching upon two different aspects.

1. Think about the contents of your bare disks, in terms of what is required to read them. Lots of RAID-1 will stick headers at the start of the disk (including Linux software raid by default, iirc). Using ZFS mirroring will necessitate the use of ZFS to read it, and so on. I prefer using only Free software raid for this reason, while opting to make myself reliant on Linux/ZFS.

2. Mirroring is much simpler for the sake of rebuilds/performance, and if you're looking for a simple hassle-free setup that's the way to go. I've got a 4+3 raidz3 for my main array, not even for capacity's sake but rather for redundancy's sake (it was going to be a 3+3 until I realized how ZFS deals out chunks). But I realize that's going to be a hassle if I do start getting correlated disk failures (I've also got true backups).

What I use on my NAS is 2 data disks that are backed up to on-demand mounted disk which keeps versioned backups of changed files. For my use case this gives me a copy of each file that I care about and in case of malware encryption, corruption or deletion of the file, I have old version available. This will not protect from a fire, theft or from loss of anything that was modified since the last backup but I feel this works well for all the scenarios I care about. Irreplaceable photos can always be uploaded to Google or similar for additional layer of security.

With RAID-5 your system still functions when a drive breaks. With RAID-6, two drives can fail without causing problems.

RAID-1 always survives a single drive failure too, and it can also survive 2 (or more) disks failing, if they aren't two mirrors of each other (not that I would wait to allow that to happen). But what about when the NAS itself fails (and they almost all do, just read the forums)?

This is why I've always used either Linux RAID or ZFS.

If the RAID is in software I can slap those disks into anything (like my desktop) and keep going.

If the RAID is NAS specific or on hardware, well, now you need something specific.

Of course you should have backups too, but why add something that will make your life more complicated in an already stressful time?

Synology at least just uses mdadm and lvm under the hood, so any Linux box can read its volumes.

How about Synology SHR? Do you need DSM to rebuild?

What SHR does is take the smallest disk size and create partitions of that size across all disks and raid them together, then take the second smallest disk size and create partitions across all disks with at least that much space and raid them together, then finds the third smallest, etc. [0]

It's all still just lvm partitions and mdadm.

> Do you DSM to rebuild?

Depends what you mean. If a disk fails, I hot-swap a replacement in and press the volume rebuild button in the GUI.

[0] https://kb.synology.com/en-my/DSM/tutorial/What_is_Synology_...

Ah, I see you've updated the question:

> Do you need DSM to rebuild?

I don't think so. I've never had to do it without DSM, but since it is just mdadm and lvm you should be able to connect all the remaining disks to and Linux box and run the appropriate mdadm commands to fix the array.

Ok, thanks!

Yes, I already figured out based on your previous response (and modified the question for others).

But will the array survive a rebuild? The calculators I've used gave me significant probability of uncorrectable read errors during this process. It only got worse as capacity increased.

> you'll need to somehow hook up 3 or 5 disks to your computer to have any chance of a recovery

If you are using only 2 disks, the only reasonable options are 0 or 1. So, you are saying that 0 is too unreliable, what I think nobody will disagree.

But when you have enough disks, RAID 6 is a no-brainier.

> All I'm saying is that when a disk fails on your NAS, you'll be in panic mode …

I have 3 + 2 disks Truenas setup, so when a disk fails, i’m still another failure away from single copy of data state. No panic, just relaxed shopping for replacement drive.

I think the authors point was that if the NAS itself fails, with RAID1 you can hook a drive up to a computer and get the data. RAID5 hardware RAID will generally require the same controller for recovery, making the operation more difficult.

ZFS or MDADM don’t have the same limitation, but there are other caveats there.

Seems like very reasonable advice to me.

I backup to raid 1 and use raid 6 for performance.

Reply on news.ycombinator.com