XFS if a "stable" (mature) file system that demonstrate performance close to ext4.

Although it can't be shrunk XFS can grow beyond 16 TiB and can be defragmented online.

We've evaluated XFS at the end of 2012 after watching Dave Chinner's exciting speech: Recent and Future Adventures in Filesystem Scalability.

First disappointment was to realise that XFS can't be used with external journal because it is impossible to mount or repair the partition without journal device.

Anyway it didn't take long to reveal another shocking problem with XFS — loss of data.

Consider cron job running every minute and saving an image taken from web camera to the folder on XFS partition. After ~20 hours we've rebooted the machine using "Raising Skinny Elephants Is Utterly Boring" 1,2 trick. Surprisingly there were no traces of current folder with data for last 20 hours to be found. xfs_check/xfs_repair didn't help and the data has gone forever even though folders with data for previous days were still there. Initially this accident was dismissed but in the few following weeks the same data loss was experienced two more times — more than enough to dismiss XFS as viable choice of file system.

There might be work-flows that do not endanger the data on XFS but in general it is unbelievably frustrating to experience data loss even once to never risk again.

Besides, xfs_check uses much more memory than fsck.ext4 for the same amount of data so another consideration for XFS is to have enough memory for xfs_repair to be able to operate.

Conclusion:

XFS is not safe and allow loss of data on hard reset or loss of power. There is no reason or excuse for modern file system with journal to lose data on sudden power loss.

Notes:

Testing was conducted on Debian x86_64 GNU/Linux_3.6.