wofferl/proxmox-backup-arm64 3.3.5-1

Version 3.3.5-1 (Debian/Bookworm)

on GitHub

These are unofficial binary packages of Proxmox Backup Server 3.x for debian/bookworm.

The proxmox-backup*,promox-mini-journalreader and pve-xtermjs packages are cross build using the source from https://git.proxmox.com/ with the build.sh script and the github action docker buildx workflow. The other arch independent packages were downloaded from http://download.proxmox.com/debian/pbs/dists/bookworm/pbs-no-subscription/binary-amd64/.

Use at your own risk.

Official Changelog

rust-proxmox-backup (3.3.5-1) bookworm; urgency=medium

• api: config: use guard for unmounting on failed datastore creation

• client: align description for backup specification to docs, using
  archive-name and type over label and ext.

• client: read credentials from CREDENTIALS_DIRECTORY environment variable
  following the "System and Service Credentials" specification. This allows
  users to use native systemd capabilities for credential management if the
  proxmox-backup-client is used in systemd units or, e.g., through a wrapper
  like systemd-run.

• fix #3935: datastore/api/backup: move datastore locking to '/run' to avoid
  that lock-files can block deleting backup groups or snapshots on the
  datastore and to decouple locking from the underlying datastore
  file-system.

• api: fix race when changing the owner of a backup-group.

• fix #3336: datastore: remove group if the last snapshot is removed to
  avoid confusing situations where the group directory still exists and
  blocks re-creating a group with another owner even though the empty group
  was not visible in the web UI.

• notifications: clean-up and add dedicated types for all templates as to
  allow declaring that interface stable in preparation for allowing
  overriding them in the future (not included in this release).

• tape: introduce a tape backup job worker-thread option for restores.
  Depending on the underlying storage using more threads can dramatically
  improve the restore speed. Especially fast storage with low penalty for
  random access, like flash-storage (SSDs) can profit from using more
  worker threads. But on file systems backed by spinning disks (HDDs) the
  performance can even degrade with more threads. This is why for now the
  default is left at a single thread and the admin needs to tune this for
  their storage.

• garbage collection: generate index file list via datastore iterators in a
  structured manner.

• fix #5331: garbage collection: avoid multiple chunk atime updates by
  keeping track of the recently marked chunks in phase 1 of garbage to avoid
  multiple atime updates via relatively expensive utimensat (touch) calls.
  Use a LRU cache with size 32 MiB for tracking already processed chunks,
  this fully covers backup groups referencing up to 4 TiB of actual chunks
  and even bigger ones can still benefit from the cache. On some real-world
  benchmarks of a datastore with 1.5 million chunks, and original data
  usage of 120 TiB and a referenced data usage of 2.7 TiB (high
  deduplication count due to long-term history) we measured 21.1 times less
  file updates (31.6 million) and a 6.1 times reduction in total GC runtime
  (155.4 s to 22.8 s) on a ZFS RAID 10 system consisting of spinning HDDs
  and a special device mirror backed by datacenter SSDs.

• logging helper: use new builder initializer – not functional change
  intended.

-- Proxmox Support Team support@proxmox.com Wed, 02 Apr 2025 19:42:38 +0200