No mount options are set on XFS volumes

[James Hebden]

The swift-storage charm currently sets no mount options when mounting created XFS volumes, and provides no option to do so.

Given the nature of Swift, and the necessary amount of directory and file walking that occurs as part of general operations, it is often necessary to adjust mount options for XFS to ensure the underlying block devices and any available caches are used efficiently.

I would propose, based on recent testing in a large (1,000,000,000+ small objects) environment I have found that setting "noatime,nodiratime,inode64,logbsize=262144" greatly reduced write performance during normal patrol reads, and allowed a much higher replicator concurrency, due to better caching of inode data in the filesystem cache, and in this case, bcache cache devices.

I would suggest these as safe and sane default mount options for any XFS filesystem prepared by this charm. The charm should also allow the default mount options to be set via a charm option, if this default needs to be modified.

The code in question is in `lib/swift_storage_utils.py` in the `setup_storage()` function. Time willing, I will be pushing some changes through the review process to change this shortly.

[James Page (james-page)]

Sounds like a sensible optimization of the swift-storage charm; triaging as medium as this has the potential to benefit deployments with large numbers of objects.