Can you tell us about your experience with the XFS file system?

Question

linux

Hello!

I have long heard about the XFS file system and its advantages, however, I would like to know how it serves others, and what people experience when using it. I'm going to upgrade disks on my server soon and expand, currently using Ext4 file system on a 4 terabyte disk, however, despite its relative survivability, there are disadvantages that I may encounter in the future: a limit on the total number of files in section 4 billion, possible difficulties in the operation of the disk more than 4 terabytes due to extended addresses. I heard that the XFS file system is specially designed for large disks and that it works very well with them.

However, I have a fear due to the following shortcomings:
- A sudden power outage (if not a sudden kernel panic) is more likely to kill the XFS file system in general (I have never seen this with ext4, at most only that some single files are broken), although on the other hand, I also heard that there is a setting, which will improve the reliability of XFS during such emergencies.
- Very inefficient work with small files (but vice versa, very efficient with large ones)
- I also heard that deleting files takes much longer than usual.

I would like to know how you used XFS, what you encountered while using it, did you feel slowdowns, lags, what happened in cases of sudden power outages or cold reboots? Did files/partitions survive such incidents, etc.?

Answer 1

Used XFS on my home torrent download for a little less than 7 years 24/7. FS as FS. Power outages were normal. For all the time, there were no problems with anything. It's hard to say for the speed, because. access was exclusively through the network.

Answer 2

Of the benefits:
Depicts df as the user wanted all the time; shows 0 if there are no files.
Never stepped on the lack of inodes.
There is no mandatory fsck time flag. On large FS, it is a surprise that the old server gets stuck for a couple of hours on fsck ext4 when rebooting.
Fsck is needed very rarely, could not cope with older versions or in clusters on shared disks.
Increasing the size on the fly (well, almost everyone already does this)
Cons: It is
impossible to reduce the size.

Answer 3

From what I encountered in real life on prod - this is slow work with a large number of small files. These were millions of pictures scattered across thousands of directories with subdirectories. Compared to ext4, the performance hit was noticeable. Otherwise, there were no complaints.

A sudden power outage (if not a sudden kernel panic) is more likely to kill the XFS filesystem altogether

Refers to journaled file systems, so there is a probability of loss, but not high. With a sudden power outage, it is more likely to kill the disk than the FS itself.
On the other hand, there is really no alternative. XFS under the wing of RH and more or less develops, you can’t say the same about ext4. Fashionable COW FS are either comparable in performance (zfs), or you can use it at your own peril and risk (btrfs).