Datasheet
Caelum
Caelum combines SMB/NFS storage and archiving to a qualified S3 target, with on-demand recall. Delay depends on your environment. GaLiLo is Laurent’s personal project dedicated to Caelum software; it is not a registered company.
OpenZFS for local data
- ✓ OpenZFS storage using single disks, mirrors or RAIDZ1/2/3, depending on configuration. SMB, NFS and Active Directory integration have been tested in a qualification scenario; this does not validate every permissions and client combination.
- ✓ ZFS volume quotas apply to SMB/NFS access; transient overruns are possible.
- ✓ Permissions: distinguish the historical model from the “applied permissions” mode. NFS AUTH_SYS does not guarantee equivalence with rich NTFS ACLs; validate owner, DENY, inheritance and groups.
- ✓ Snapshots are read-only, administrable and subject to retention. They do not replace independent backups.
- ✓ Windows Previous Versions workflows depend on configured snapshots and archived-file restoration limitations.
- ✓ Other storage options and client/identity combinations require qualification for your deployment.
S3 archiving and on-demand reads
- ✓ Archived files are replaced by stubs at their original paths. Reading depends on the protocol, active modules and object-storage availability; delay varies with the network and the data to retrieve.
- ✓ Qualify the S3 target against its implementation and the reading paths used.
- ✓ Tasks resolve their scope and destination. The first run uses simulation by default; this default can be overridden and is not a universal approval lock.
- ✓ Range recall can avoid downloading the entire file before reading, depending on the access path; it does not guarantee instant opening.
- ✓ Stubs carry an object identifier and a descriptor with a SHA-256 digest. These attributes are not signed and do not provide WORM protection.
- ✓ Purge accounts for live references and Caelum retention. Manage the version lifecycle separately on versioned buckets.
High availability: unqualified beta
- ✓ Two distinct architectures are available. Validate them outside production before relying on them; this overview does not guarantee service continuity or recovery objectives.
- ✓ Shared storage: Pacemaker orchestration of the primary pool, shares and VIP. The setup script includes fencing, but it cannot be configured in the wizard; failover conditions still require qualification.
- ✓ Replication: asynchronous ZFS snapshots with manual promotion. Effective RPO depends on the age of the last successfully received snapshot, not just the configured interval.
- ✓ Before manual promotion, isolate the peer and verify that it is no longer serving data. The code promotes locally before requesting remote demotion; a demotion failure does not undo promotion.
- ✓ Risk acknowledgements do not fence the peer. Pool coverage depends on the selected mode and the datasets actually replicated.
- ✓ Entitlements and activation for each appliance must be specified in the pilot terms.
Prepare a migration from SMB
- ✓ Incremental copy allows a transfer to be rerun. Plan final synchronization and access changes within a cutover window suited to your environment.
- ✓ Copy warnings can remain even when a job is marked complete; review the report before acceptance.
- ✓ Permissions may be reproduced only partially. Validate owner, inheritance, DENY and groups on representative samples; do not equate NFS access with Windows ACLs.
- ✓ The described workflow does not deliver an independent, exhaustive post-copy hash verification pass. Plan explicit integrity checks.
- ✓ Plan duration, potential interruption and DNS/UNC changes; uninterrupted business cutover is not guaranteed.
SMB filtering and recovery points
- ✓ SMB veto filtering matches names and extensions; it does not detect every attack. Snapshots can restore an earlier state when a suitable recovery point is available, alongside independent backups.
- ✓ Known limitation in 0.65.9 — T-236: the veto also affects legitimate files such as .lock and .enc. It can reject creation or make existing files invisible and unreadable over SMB, without deleting them from disk. A fix is planned for 0.65.10; it is not delivered in the version described.
- ✓ ZFS snapshots are read-only, subject to administrative permissions and retention. They are neither a WORM guarantee nor an offline backup.
- ✓ Recovery requires a point created before the incident. For archived files, follow the restoration and catalog replay limitations of the installed version.
- ✓ Alerting and logging integrations must be described in terms of the events, formats and destinations actually tested.
Validate your deployment network
- ✓ Define administration, file, replication and object-storage traffic before the pilot. This page does not qualify every network or SAN option.
- ✓ Prepare a per-function validation plan covering versions, hardware and relevant failure scenarios.
- ✓ Check S3 connectivity and the selected HA mode’s prerequisites separately.
- ✓ Document a way back to the previous configuration before changing network settings.
Web console and REST API
- ✓ The v2 web console and documented REST API provide administration access. This overview does not claim exhaustive UI/CLI feature parity.
- ✓ Updates: stable and beta channels; installation from a local package on a compiled appliance, without the appliance downloading that package.
- ✓ Returning to a code version is separate from data restoration. Restoring a recovery point discards writes made after that point.
- ✓ Assistant off by default. Two modes, each explicitly activated: a Mistral key provided by GaLiLo (once enabled, content sent is transmitted to Mistral AI) and BYOK, a screen planned for 0.65.10 and not delivered here.
- ✓ Offline update-package installation does not make online services available offline. Data sent to the assistant and its recipients must be documented; do not enter confidential information before that verification.
Deployment
Installation from an ISO image. The documented test environment uses Proxmox; validate your chosen platform, drivers and sizing before deployment.
Hardware requirements
| Platform | Documented test environment uses Proxmox; validate other environments |
|---|---|
| Resources | Measure sizing for workload, storage and concurrent readers |
| Validation | Test drivers, access and failure scenarios before deployment |
Try Caelum
Pilot access after approval; scope, duration and terms must be specified before installation.