File cache is not persistent across remounts

[nbdd0121]

Describe the issue

When GCSFuse unmounts and remounts, the file cache populated doesn't seem to be accessed.

n https://cloud.google.com/storage/docs/gcsfuse-cache#persistence says that the file cache should be reused when the metadata cache has been populated, but it doesn't seem to be the case.

A brief grep through the codebase suggests that contentcache have the code that handles persistence of the cache, but this functionality is not present in the filecache code. The contentcache seems completely unused currently so the functionality is lost.

Reproduction

The following experiment is conducted:

• Have a 1.7G file on a GCS bucket
• Mount with the following options gcsfuse -o ro --file-mode=755 --config-file=gcs.yaml BUCKET_NAME /mnt with the config file below:

  file-cache:
  max-size-mb: -1
  cache-file-for-range-read: true
  metadata-cache:
  stat-cache-max-size-mb:  32
  ttl-secs: 3600
  type-cache-max-size-mb: 4
  cache-dir: ~/.cache/gcsfuse

• vmtouch -t /mnt/large_file takes ~18 seconds.
• Subsequent vmtouch calls are also instant to complete.
• umount /mnt and remount.
• vmtouch -t /mnt/large_file still takes ~18 seconds.

System (please complete the following information):

• OS: NixOS 24.05
• Platform: Bare-metal
• Version: 2.3.1

Additional context Add any other context about the problem here.

SLO: We strive to respond to all bug reports within 24 business hours.

[charith87]

Thanks @nbdd0121 for reaching out. This is a known limitation at this point that the file cache is not persistent across mounts.

The documentation could have been more clearer here. It actually tries to say the file cache will be re-used for subsequent operations once the metadata cache is filled. It assumes that file cache is also filled in the current session. The lines together with the previous line is what might be misleading. Will work with the documentation team to change this appropriately.

The reason for the limitation right now is the file cache doesn't hold a marker of the generation(version) of the object on the disk ( which the metadata cache holds). There's no way to reconcile if the local file (cached already) and the new metadata (filled in the new mount session) are the same without additional changes. This is a roadmap item in the future but no timelines at this point.

cc @marcoa6