Hacker News Reader

Comment by alexdowad | Hacker News Reader

I tried btrfs and got bitten by bugs; not doing that again. Judging from this move by RH, it looks like I wasn't the only one.

View on HN

What bugs?

I recently setup a software mirroring raid with btrfs and I'm loving features like checksumming. It makes me feel my data is quite safe and can't bit rot anymore. So far it is working fine.

Hi.

Did you notice that the official description of RAID-1 is "Mostly working"? Are you aware if one of your drive fails, you have one chance to re-mirror it, before the remaining drive can no longer be mounted read-write and you need to dump the filesystem and re-create from scratch?

What do you mean by one chance to re-mirror? Does this mean that if the resilver fails you can't try it again? Is this documented? With ZFS or regular RAID as long as you have one good disk in the mirror you can resilver, is this not the case for BTRFS? If so this is quite disappointing.

There's a reason my server is running ZFS on FreeBSD. I also love jails, which let me have as many virtual servers as I want without any virtualization overhead.

> Does this mean that if the resilver fails you can't try it again? Is this documented?

It's documented on the status Wiki. RAID 1 is "mostly working". If a mirror drops to having one disk, you can mount it once as a read-write volume (required for resilvering); after that, you have to trash it and start again.

Tbh I don't care if I can't mount it read-write as long as I can mount RO and get the data to a new filesystem. The drives are new so nothing to worry about for years (probably). And if I get a chance ("one chance") to re-mirror, even better.

It might be an inconvenient restriction, but when a drive fails I'm already happy that there will be no data loss.

RAID1 meaning you get one chance. Just move to RAID10 if you're worried. (/s)

It has exactly the same bug with RAID10.

That's just how raid mirroring works. Now sure why you mention it specifically for btrfs?

No, it isn't. You can run a ZFS VDEV or a Linux mdraid as a single-disk RAID 1 unit until the remaining disk fails. You have an arbitary number of reboots/remounts to fix the problem.

So it turns out there's some nuance here. You can remount it multiple times, as long as you don't write to it, which I was familiar with. The moment you write again and don't explicitly remove the extra volume, you'll lose the ability to mount r/w, which is a pretty bad bug :(

I had tried xfs recently and I hated it. I have a thin laptop which would just lose power if I grabbed it wrong and very quickly my xfs partition got corrupted and I was unable to fix it. So, I had to re-install my os, and went back to ext4. Which even with the same power offs, never an issue since several months. This is largely anecdotal. But have had a coworker with the same issue too...

I also had issues with a vms using xfs too.

But, I do use xfs on ssd raids(in our servers, being used for testing) and never had an issue there.

We lost an entire openstack host due to an xfs glitch last year. It needed rebuilding from scratch.

Reply on news.ycombinator.com