17 comments

[ 5.1 ms ] story [ 48.3 ms ] thread
I thought NTFS and ZFS are the enterprise filesystems of choice? We merely tolerate others. That has been my experience over the last 15 years or so.
NTFS because you have to, ZFS has a little restriction on it that makes it difficult to be 'of choice' and EXT4 needs to go away for huge volumes.

Linux with a huge volume? Use XFS.

Or use FreeBSD with ZFS or buy a NAS appliance.
i recall XFS had some serious corruption issues if a failure happens when not on a UPS (or the UPS fails catastastrophically). Has this been dealt with ?
From personal experience I can say that I haven't seen it recently. RHEL 7 went to XFS as their default filesystem so hopefully they've looked into it.
Yes, it had but that bug was fixed like 10 years ago. Stop spreading FUD.
Possibly fixed. I've had one incident of an XFS filesystem refuse to mount after it replayed the journal after a power outage. xfs_repair worked but I shouldn't have to piss around in single user mode in 2015.

Kit was a HP DL380p, SmartArray 420i SAS array, CentOS 7 about 3 months ago.

Outage caused by a large UPS explosion that took out 100 machines. No NTFS or ext3/4 trouble at all.

Could we add a "(2013)" to the title?

Also, can anyone speak to the current state of affairs with respect to ext4, xfs, and btrfs? I'd love to hear about the tooling and code quality in these file systems from an informed (and recent) point of view. Has Redhat been putting out fires since the switch? Did it go unnoticed? Are there measurable performance and/or reliability gains?

..no actual content besides "We at SuSE have been doing this for aaaaaaaages"...
xfs is my first choice for data loss. Poweroff in suspend and the root fs was bricked, twice. Never ever again.
Do you know what kernel version you were running?

Was this years ago, or more recently?

Last try was in maybe ~2012.

I learned from a collegue that if the hard drive implements barriers incorrectly, this can cause easy XFS corruptions. Maybe this was the reason, especially SSD firmwares are very complex these days...

XFS better than EXT4 because EXT4 needs fsync()? XFS needs fsync too or you'll end up with data files full of zeroes.

It happened to me.

XFS would be perfect if there was a way to disable delayed allocation. Has that changed recently? I still use it on a number of machines because it tends to work better with large volumes. But that hasn't kept me from losing data with it, repeatedly. Of course I've lost data with lots of other linux file systems too, so in that regard it isn't much worse. These days if I really think the data is important I still use a bunch of little ext3 partitions with data=journal, barrier=1 and all the disk caches set to write through mode.

https://en.wikipedia.org/wiki/Allocate-on-flush