lakeFS Mount v0.14.0¶

Changelog¶

✨ New features and improvements

• Content-addressed disk cache. Everest addresses its disk cache by object content instead of by open file handle. Downloaded data can be reused across reads, and eviction treats data read from local disk as in use.
• Cache size limit. The combined metadata and data disk caches stay within the --cache-size budget.
• Group-writable directories on FUSE. When a FUSE mount has --mount-gid set and a directory has no stored POSIX mode, the directory is group-writable.

🐛 Bug fixes

• Paths with a double slash (//) are skipped because filesystems cannot represent them. Skipping them lets you read the other files in the repository.

• Commands other than mount and mount-server now reject --lakectl-* flags and no longer read lakectl configuration. The flags are --lakectl-config, --lakectl-access-key-id, --lakectl-secret-access-key and --lakectl-server-url.

• Reading a small part of a large file no longer drops the whole file from the cache. The cache now accounts for the chunks actually downloaded.

• After a commit, an old FUSE directory handle returns ESTALE. Before, it could accept changes that never appeared in everest diff or the next commit.

• The mount server now exits when its FUSE connection closes and handles SIGTERM. CSI Mount Pods can finish and be deleted.

• Linux NFS mounts no longer use the user option, which caused an unintended noexec behavior.

• An NFS directory read no longer deadlocks when another thread writes to that directory.
• everest commit retries on failure. A mount with no changes now succeeds and adds an empty commit to the branch. Before, it failed with "no changes".

Assets¶