ArgosLinux sees files.
The FUSE frontend handles file I/O, directories, links, ownership, timestamps, extended attributes, and permissions. Byte-oriented filenames are preserved without assuming UTF-8.
Browse the FUSE frontendUnder the hood
One filesystem.
Four working layers.
A Linux FUSE frontend connects ordinary file operations to an erasure-coded data path, persistent metadata, and a health-aware controller.
Disks hold shards.
Writes are split into stripes, compressed and optionally encrypted, then encoded into data and parity shards. Checksums detect corruption; surviving shards support degraded reads and repair.
Browse the data pathChanges leave a record.
Metadata copies, checksummed journals, replay, and checkpoints support recovery. The current implementation still serializes logical metadata as JSON; keyed-page helpers and raw checkpoint indexes are steps toward a more scalable store.
Metadata design & current limitsMaintenance has a plan.
Disk health, capacity, latency, and workload signals guide placement and maintenance. The controller can explain its choices, limit background work, and verify state after an action.
Read the controller modelThe control loop
Observe. Plan. Act. Verify.
Each stage makes the controller’s behavior inspectable. Dry-run mode lets you review a plan before any maintenance is executed.
Observe
Collect available SMART counters, capacity pressure, measured latency, and workload information. Missing telemetry is part of the state.
Plan
Score candidate actions with confirmation gates, cooldowns, redundancy checks, and capacity constraints. Keep rejection reasons visible.
Act
Run bounded scrub, drain, repair, or rebalance work. Maintenance budgets and throttling limit how much work a cycle can do.
Verify
Check journal, filesystem, and shard state after maintenance. Verification failures can downgrade the automation mode.
Storage backends
Choose what
sits underneath.
Host directories
Shard objects are ordinary host files. This is the simplest way to explore the CLI and mounted behavior.
Loop images
Image files use the project’s block layout and allocator. Useful for testing pool and rootfs behavior without assigning physical disks.
Raw devices
Dedicated block devices use the same raw format. Device selection, recovery, and replacement need operator validation.
Data-path options
Configure the trade-offs.
zstd, lz4, or none
Per-stripe compression is configurable. Choose the CPU and storage trade-off that fits the data you actually store.
Authenticated payloads
Optional XChaCha20-Poly1305 protects stripe data with Argon2id-derived keys. Encryption is configured explicitly.
Buffered by default
RAM and persistent L2 caches accelerate reads. Optional direct I/O and io_uring paths have buffered fallbacks.