Skip to content

v0.61.0

Choose a tag to compare

@github-actions github-actions released this 15 Nov 06:58
· 19 commits to main since this release
v0.61.0

filex v0.61.0

Self-hosted file manager - Go single binary + multi-framework frontend.

Download a binary below, or pull a Docker image:

docker pull ghcr.io/brf-tech/filex:slim-v0.61.0
docker pull ghcr.io/brf-tech/filex:full-v0.61.0

What changed

Added

  • The storage form says what a scan interval would cost, measured on the
    storage's last scan
    (task #353). Under Scan every (minutes) filex says
    how long the last full scan took, for how many files and folders, and the
    share of the time scans at the typed interval keep the storage busy; from a
    quarter of the time it suggests a rounder, longer interval - every 5 min,
    about 80% of the time; suggested 50 min
    - with a button that puts it in the
    field, and a scan that outlasts the interval is said to scan all the time.
    When the last scan took filex to 80% of the memory it was given (the
    container's limit, else the machine's RAM) it warns that the system would
    stop filex. The server judges it while you type
    (GET /api/admin/storages/{id}/interval-advice); any interval still saves.
    Saving a storage answers with the same judgement (interval_advice), so an
    agent using the admin_storages_create / _update MCP tools hears it too.
    A full scan records the most memory the process held while it walked
    (peak_mem_bytes on the run, migration 00139; Peak memory in the
    storage's run table).
    STORAGE.md → How long a scan takes.
  • Switch to live watching, in one click (task #351). A local storage that
    polls - every one made before this release - is marked in the storage list
    and offered the switch on its page; the click saves the mode alone.
    STORAGE.md → Which mode a new storage gets.
  • filex-watch: changes made on another machine reach filex in seconds
    (task #352). A watch only sees what its own machine writes, so a storage whose
    disk is elsewhere - a NAS share mounted over the network, the folder an SFTP
    storage points at - learned of a change made there at its next sync.
    filex-watch is a small program for that machine (Linux, Windows and macOS on
    amd64 and arm64, a systemd unit, a launchd daemon, a Windows service, the
    image ghcr.io/brf-tech/filex-watch) that watches the folder and reports the
    folders that change to POST /api/storages/{id}/watch/events; filex looks at
    those folders alone - a Live watching storage reconciles each one, any other
    mode rescans its subtree - and a start, a lost connection, a full queue or
    a filex that started again since its last answer asks for one resync of the
    storage. The storage's page has a card for it (not for an object store such as S3, whose driver says there is nothing to watch):
    make a key for a machine (shown once), see what each key's agent last said,
    copy the install commands, revoke a key; the admin MCP tools
    admin_storages_watch, admin_storages_watch_key_create and
    admin_storages_watch_key_delete do the same. The key is an API key of its
    own kind: it opens that storage's change reports and nothing else - no file
    API, no protocol, no admin route - and reports as the administrator who made
    it, as they are now. filex-watch has a release line of its own
    (filex-watch-v0.1.0), never GitHub's "latest" release, and neither the
    update check nor the Releases page lists it.
    docs/FILEX-WATCH.md.

Changed

  • A new local storage is watched live instead of scanned every 15 minutes
    (task #351). A file copied into a storage's folder without filex - over the
    NAS's own share, by a download client, by another container - stayed
    invisible until the next scan, and a folder listed from the disk during the
    first scan looked empty again once it ended. A storage nobody chose a mode
    for now gets its driver's default, decided in one place and read by the
    environment seed every store package uses (CasaOS, Umbrel, Unraid, TrueNAS),
    the admin API, filex storage add and the storage form: live watching
    (sync_mode: lazy) for local, so the opened folder is read from the disk
    and watched, and poll for every other driver. The scan interval stays as
    the safety net. Existing storages keep their mode.
    STORAGE.md → Which mode a new storage gets.
  • On a network filesystem nothing is watched (task #351). A watch only hears
    what passes through this machine, so on NFS, SMB/CIFS, 9P or a FUSE mount
    (Unraid's /mnt/user included) a live-watched storage places none and reads
    every opened folder from the disk; the storage page says so and names
    filex-watch, which reports changes from the machine
    the disk is attached to.
    A storage set to Watch the whole storage (fsnotify) on such a
    filesystem is polled at its own interval instead, the way it already was on
    a driver that is not local: the server logs it once at start and the storage
    page says it is scanned at its interval.
    LAZY-CATALOGUE.md → Network filesystems.
  • The sync modes have new names on screen (task #351): lazy is Live
    watching
    (it was Lazy catalog) and fsnotify is Watch the whole
    storage
    (it was Live watch, the name the new default now carries). The
    values in the API and the configuration do not change.

Fixed

  • On MySQL a scan no longer ends before it began. A sync run's start
    is kept to the microsecond and its end was written to the whole second, so
    a run that started and ended within one second was stored as ending first.
    The end is now as precise as the start.

  • Selecting an archive member whose name holds a line break no longer does
    nothing.
    7-Zip 25.01 and newer (every version filex accepts, the Docker
    image's included) list a line break in a member's name as _, so a 7z, RAR
    or encrypted ZIP showed line_break.txt for line + line break +
    break.txt, and extracting that selection wrote nothing and reported
    nothing. filex now finds the member the listed name stands for, and it is
    skipped with name_not_allowed and named in the result, as when the whole
    archive is extracted.
    ARCHIVES.md → Safety model.

Full changelog entry


Verify: sha256sum -c checksums.txt