Reorganize docs into v1 (beetfs) and v2 (new architecture)

docs/v1/ - Original beetfs documentation:
  - analysis.md, components.md, data-flow.md, drawbacks.md
  - features.md, modernization.md, rust-migration.md
  - benchmark-plan.md, benchmark-results.md, e2e-test-plan.md
  - README.md

docs/v2/ - New MusicFS architecture:
  - requirements.md: Full requirements spec (FR-1 to FR-25, NFR-1 to NFR-14)
    - P0: Multi-origin, plugins, CAS, control API
    - P1: Search, album art, prefetch, metadata sources
    - P3: HA, 10M+ files scalability
  - architecture.md: Google BlueDoc style design document
    - PlantUML diagrams for all components
    - Design requirements with quantitative targets
    - Alternatives considered, implementation plan
This commit is contained in:
Alexander
2026-05-12 16:46:37 +02:00
parent 3a6115cbab
commit 1374084135
14 changed files with 2248 additions and 276 deletions
-276
View File
@@ -1,276 +0,0 @@
# beetfs Architecture
## System Overview
beetfs implements a **metadata overlay filesystem** using FUSE. The key innovation is separating metadata storage (in beets SQLite database) from audio data storage (original files on disk).
```
┌─────────────────────────────────────────────────────────────────────────────┐
│ USER SPACE │
│ ┌─────────────┐ ┌─────────────────────────────────────────────────────┐ │
│ │ Application │ │ beetfs │ │
│ │ (VLC, etc) │ │ ┌─────────────┐ ┌──────────────┐ ┌────────────┐ │ │
│ │ │◄───┼──┤beetFileSystem│──│ FileHandler │──│ Interpol. │ │ │
│ │ │ │ │ (FUSE) │ │ │ │ FLAC/ID3 │ │ │
│ └─────────────┘ │ └─────────────┘ └──────────────┘ └────────────┘ │ │
│ │ │ │ │ │ │
│ │ ▼ ▼ ▼ │ │
│ │ ┌─────────────┐ ┌──────────────┐ ┌────────────┐ │ │
│ │ │ FSNode │ │ Beets │ │ Original │ │ │
│ │ │ (dir tree) │ │ Database │ │ Files │ │ │
│ │ └─────────────┘ └──────────────┘ └────────────┘ │ │
│ └─────────────────────────────────────────────────────┘ │
├─────────────────────────────────────────────────────────────────────────────┤
│ KERNEL SPACE │
│ ┌───────────────┐ │
│ │ FUSE VFS │ │
│ └───────────────┘ │
└─────────────────────────────────────────────────────────────────────────────┘
```
## Component Architecture
### 1. Plugin Layer
```python
class beetFs(BeetsPlugin):
"""Beets plugin hook - registers the 'mount' subcommand"""
def commands(self):
return [beetFs_command]
beetFs_command = Subcommand('mount', help='Mount a beets filesystem')
beetFs_command.func = mount
```
### 2. Initialization Flow
```
beet mount /mountpoint
┌───────────────────────────────────────────────────────────────┐
│ mount() function │
│ 1. Parse PATH_FORMAT template │
│ 2. Create FSNode root (directory_structure) │
│ 3. Iterate all items in beets library │
│ 4. For each item: │
│ - Build template substitution map │
│ - Add directories to FSNode tree │
│ - Add file entry (filename → item.id mapping) │
│ 5. Create beetFileSystem FUSE server │
│ 6. server.main() - enter FUSE event loop │
└───────────────────────────────────────────────────────────────┘
```
### 3. Virtual Directory Structure
The default path template:
```python
PATH_FORMAT = "$artist/$album ($year) [$format_upper]/$track - $artist - $title.$format"
```
Results in structure like:
```
/mountpoint/
├── Pink Floyd/
│ └── The Wall (1979) [FLAC]/
│ ├── 01 - Pink Floyd - In The Flesh?.flac
│ └── 02 - Pink Floyd - The Thin Ice.flac
└── Led Zeppelin/
└── IV (1971) [FLAC]/
└── 01 - Led Zeppelin - Black Dog.flac
```
### 4. FSNode Tree Structure
```python
class FSNode:
dirs: Dict[str, FSNode] # subdirectories
files: Dict[str, int] # filename → beets item ID
# Example tree:
FSNode(
dirs={
"Pink Floyd": FSNode(
dirs={
"The Wall (1979) [FLAC]": FSNode(
dirs={},
files={
"01 - Pink Floyd - In The Flesh?.flac": 42,
"02 - Pink Floyd - The Thin Ice.flac": 43
}
)
},
files={}
)
},
files={}
)
```
## Core Data Flow
### Read Operation
```
Application: read("/mount/Artist/Album/track.flac", offset=0, size=4096)
┌───────────────────────┐
│ beetFileSystem.read() │
│ Lines 1077-1106 │
└───────────┬───────────┘
┌───────────────┴───────────────┐
│ Get/Create FileHandler │
│ for this path │
└───────────────┬───────────────┘
┌───────────┴───────────┐
│ FileHandler.read() │
│ Lines 497-517 │
└───────────┬───────────┘
┌───────────────┴───────────────┐
▼ ▼
┌─────────────────────┐ ┌─────────────────────┐
│ offset < bound │ │ offset >= bound │
│ (in header area) │ │ (in audio area) │
└──────────┬──────────┘ └──────────┬──────────┘
│ │
▼ ▼
┌─────────────────────┐ ┌─────────────────────┐
│ Return modified │ │ Return original │
│ header from DB │ │ audio from file │
│ │ │ │
│ self.header[...] │ │ self.music_data[...]│
└─────────────────────┘ └─────────────────────┘
```
### Write Operation
```
Application: write("/mount/Artist/Album/track.flac", data, offset=100)
┌───────────────────────┐
│ beetFileSystem.write()│
│ Lines 1108-1135 │
└───────────┬───────────┘
┌───────────┴───────────┐
│ FileHandler.write() │
│ Lines 519-565 │
└───────────┬───────────┘
┌───────────────┴───────────────┐
▼ ▼
┌─────────────────────┐ ┌─────────────────────┐
│ offset < bound │ │ offset >= bound │
│ (in header area) │ │ (in audio area) │
└──────────┬──────────┘ └──────────┬──────────┘
│ │
▼ ▼
┌─────────────────────┐ ┌─────────────────────┐
│ 1. Patch header │ │ DISCARD │
│ 2. Parse new tags │ │ (audio writes │
│ 3. Extract values │ │ not allowed) │
│ 4. Update beets DB │ │ │
│ 5. Regenerate header│ │ │
└─────────────────────┘ └─────────────────────┘
```
## Memory Model
### FileHandler State
```python
class FileHandler:
# Paths
path: str # Virtual path in FUSE mount
real_path: str # Actual file on disk
# Beets integration
item: Item # Beets library item
lib: Library # Beets library reference
# File data
file_object: File # File handle (closed after init)
music_data: bytes # Audio data cached in memory
# Metadata
format: str # "flac" or "mp3"
inf: FLAC/ID3 # Interpolated metadata object
header: bytes # Generated header with DB metadata
bound: int # Byte offset where header ends
music_offset: int # Byte offset where audio starts in original
# Reference counting
instance_count: int # Number of open handles
```
### Memory Layout
```
Virtual File (as seen by application):
┌────────────────────────────────────────────────────────────────┐
│ HEADER (from DB) │ AUDIO (from file) │
│ [0 ... bound) │ [bound ... EOF) │
│ │ │
│ Generated by InterpolatedFLAC │ Cached in music_data │
│ Contains: title, artist, album, │ Original audio frames │
│ genre from beets DB │ Unchanged │
└────────────────────────────────────────────────────────────────┘
▲ ▲
│ │
self.header self.music_data
Original File (on disk):
┌────────────────────────────────────────────────────────────────┐
│ ORIGINAL HEADER │ AUDIO DATA │
│ [0 ... music_offset) │ [music_offset ... EOF) │
│ │ │
│ May have different │ Same as virtual file │
│ tag values │ │
└────────────────────────────────────────────────────────────────┘
```
## Threading Model
```python
server.multithreaded = 0 # Single-threaded mode
```
beetfs runs in **single-threaded mode** to avoid concurrency issues with:
- Shared `files` dictionary
- Beets library access
- File handle reference counting
## Global State
```python
# Module-level globals (set during mount)
structure_split: List[str] # PATH_FORMAT split by "/"
structure_depth: int # Number of path components
library: Library # Beets library instance
directory_structure: FSNode # Root of virtual directory tree
```
## Error Handling
| Situation | Response |
|-----------|----------|
| File not found | Return `-errno.ENOENT` |
| Permission denied | Return `-errno.EACCES` |
| Operation not supported | Return `-errno.EOPNOTSUPP` |
| Parse error | Log and return `-errno.ENOENT` |
## Limitations
1. **Format Support**: Only FLAC fully implemented; MP3 support is incomplete
2. **Memory Usage**: Entire audio portion cached in memory per open file
3. **Single-threaded**: No concurrent access optimization
4. **No Streaming**: Full file must be read into memory
5. **Python 2**: Uses deprecated language features
6. **fuse-python**: Old FUSE bindings, not maintained
+249
View File
@@ -0,0 +1,249 @@
# beetfs Feature Set
## Overview
beetfs is a FUSE filesystem plugin for [beets](https://beets.io/) that presents your music library as a virtual filesystem organized by metadata. Files appear with paths derived from their database metadata, and reading file headers returns metadata from the beets database rather than the actual file tags.
**Author**: Martin Eve (2010)
**License**: GPLv3
**Python**: 2.7 (uses fuse-python)
## Core Features
### 1. Virtual Metadata-Based Directory Structure
Files are presented in a configurable path format based on beets database fields:
```
$artist/$album ($year) [$format_upper]/$track - $artist - $title.$format
```
**Example**:
```
/mnt/beetfs/
├── Metallica/
│ └── 72 Seasons (2023) [FLAC]/
│ ├── 01 - Metallica - 72 Seasons.flac
│ ├── 02 - Metallica - Shadows Follow.flac
│ └── ...
├── Pink Floyd/
│ └── The Dark Side of the Moon (1973) [FLAC]/
│ └── ...
```
**Available template variables**:
- `$artist`, `$album`, `$title`, `$genre`, `$composer`, `$grouping`
- `$year`, `$month`, `$day`
- `$track`, `$tracktotal`, `$disc`, `$disctotal`
- `$format`, `$format_upper` (file extension)
- `$lyrics`, `$comments`, `$bpm`, `$comp`
### 2. Metadata Overlay (Read)
When you read a file through beetfs, the **metadata header is synthesized from the beets database**, not read from the actual file on disk.
**How it works**:
1. Open file → beetfs reads the real file from disk
2. Parse the audio format header (FLAC/MP3)
3. Replace metadata fields with values from beets database
4. Return synthesized header + original audio data
**Supported fields for overlay**:
- `title`, `artist`, `album`, `genre` (FLAC only currently)
**Use case**: Your files may have inconsistent or wrong tags, but beetfs presents them with the corrected metadata from your beets library.
### 3. Metadata Passthrough (Write)
When you write to file headers through beetfs, the **changes are saved to the beets database**, not to the actual file.
**How it works**:
1. Application writes new metadata to file header region
2. beetfs intercepts the write
3. Parses the new metadata values
4. Updates the beets database (`lib.store()`, `lib.save()`)
5. Regenerates the synthesized header
**Result**: Tag editors (Picard, Kid3, etc.) can edit metadata through beetfs, and changes persist in the beets database without modifying the original files.
### 4. Format Support
| Format | Read | Metadata Overlay | Write to DB |
|--------|------|------------------|-------------|
| FLAC | ✅ | ✅ Full | ✅ |
| MP3 | ✅ | ❌ Disabled | ❌ |
| Other | ❌ | ❌ | ❌ |
**FLAC Implementation**:
- Uses `InterpolatedFLAC` class extending mutagen
- Reconstructs Vorbis comment block with DB values
- Preserves audio data and other metadata blocks
**MP3 Implementation**:
- Passthrough only (no interpolation)
- `self.bound = 0` disables header replacement
### 5. File Caching
Open files are cached in `FileHandler` objects:
- First open: Load entire file into memory, parse headers
- Subsequent opens: Reuse cached `FileHandler`
- Reference counting for multiple opens
- Release when reference count reaches zero
**Memory impact**: Each open file consumes ~filesize RAM.
## FUSE Operations
### Implemented (Functional)
| Operation | Description |
|-----------|-------------|
| `getattr` | File/directory stat (size, mode, timestamps) |
| `access` | Permission checking |
| `opendir` | Open directory for listing |
| `readdir` | List directory contents |
| `releasedir` | Close directory |
| `open` | Open file for reading/writing |
| `read` | Read file contents |
| `write` | Write to file (header region only) |
| `release` | Close file |
| `fgetattr` | Stat with file handle |
| `statfs` | Filesystem statistics |
### Not Implemented (Return EOPNOTSUPP)
| Operation | Reason |
|-----------|--------|
| `create` | Read-only structure |
| `mknod` | Read-only structure |
| `mkdir` | Read-only structure |
| `unlink` | Read-only structure |
| `rmdir` | Read-only structure |
| `symlink` | Not needed |
| `link` | Not needed |
| `rename` | Would break DB consistency |
| `chmod` | Metadata-only FS |
| `chown` | Metadata-only FS |
| `truncate` | Would corrupt audio |
| `utime` | Metadata-only FS |
## Usage
### Mount
```bash
beet mount /mnt/beetfs
```
### Unmount
```bash
fusermount -u /mnt/beetfs
```
### Example Session
```bash
# Mount the filesystem
beet mount /mnt/music
# Browse by artist
ls /mnt/music/
# Metallica/ Pink Floyd/ The Beatles/ ...
# List an album
ls "/mnt/music/Metallica/72 Seasons (2023) [FLAC]/"
# 01 - Metallica - 72 Seasons.flac
# 02 - Metallica - Shadows Follow.flac
# ...
# Play through any music player
mpv "/mnt/music/Metallica/72 Seasons (2023) [FLAC]/01 - Metallica - 72 Seasons.flac"
# Edit tags (changes go to beets DB)
kid3 "/mnt/music/Metallica/72 Seasons (2023) [FLAC]/"
# Unmount
fusermount -u /mnt/music
```
## Architecture
```
┌─────────────────────────────────────────────────────────────┐
│ User Applications │
│ (mpv, Rhythmbox, Kid3, etc.) │
└─────────────────────────┬───────────────────────────────────┘
│ POSIX calls (open, read, write)
┌─────────────────────────────────────────────────────────────┐
│ Linux Kernel │
│ FUSE module │
└─────────────────────────┬───────────────────────────────────┘
│ /dev/fuse
┌─────────────────────────────────────────────────────────────┐
│ beetfs │
│ ┌─────────────┐ ┌──────────────┐ ┌───────────────────┐ │
│ │ FSNode Tree │ │ FileHandler │ │ InterpolatedFLAC │ │
│ │ (in-memory) │ │ (cache) │ │ (header synth) │ │
│ └─────────────┘ └──────────────┘ └───────────────────┘ │
└────────┬────────────────┬───────────────────┬───────────────┘
│ │ │
▼ ▼ ▼
┌─────────────┐ ┌─────────────────┐ ┌─────────────────┐
│ Beets DB │ │ Real Files │ │ Mutagen │
│ (SQLite) │ │ (on disk) │ │ (parsing) │
└─────────────┘ └─────────────────┘ └─────────────────┘
```
## Limitations
### Current Bugs (Non-Functional)
1. **Nested Methods Bug**: Lines 758-1144 are indented inside `access()`, making FUSE operations unreachable
2. **Directory Tree Bug**: `FSNode.adddir()` crashes when building tree for non-empty library
### Design Limitations
1. **Memory Usage**: Entire file loaded into RAM on open
2. **Mount Time**: O(N) - loads all library items at mount
3. **No Lazy Loading**: Full directory tree built upfront
4. **Single Format**: Only FLAC has full metadata overlay
5. **No Real File Modification**: Writes only update DB, not actual files
6. **Python 2.7 GIL**: Single-threaded performance
### Not Supported
- Creating/deleting files or directories
- Moving/renaming files
- Modifying audio content
- Album art / embedded images
- Multi-value tags
- Non-ASCII in some edge cases
## Configuration
Currently hardcoded. Potential configuration points:
| Setting | Current Value | Description |
|---------|---------------|-------------|
| `PATH_FORMAT` | `$artist/$album ($year)...` | Directory structure template |
| `METADATA_RW_FIELDS` | 17 fields | Fields available for read/write |
| Caching | Always on | FileHandler caching behavior |
| Threading | Disabled | `multithreaded = 0` |
## Dependencies
- Python 2.7
- fuse-python
- beets 1.4.x
- mutagen (FLAC/MP3 parsing)
## See Also
- [e2e-test-plan.md](e2e-test-plan.md) - Test strategy and bug documentation
- [benchmark-plan.md](benchmark-plan.md) - Performance measurement methodology
- [benchmark-results.md](benchmark-results.md) - Current benchmark status
+451
View File
@@ -0,0 +1,451 @@
# Rust Migration Analysis for beetfs
## Executive Summary
Migrating beetfs from Python to Rust is **strongly recommended** based on research findings. Expected improvements:
| Metric | Python (Current) | Rust (Expected) | Improvement |
|--------|------------------|-----------------|-------------|
| **Memory per file** | ~280 bytes overhead | ~60 bytes | **4-5x reduction** |
| **File open latency** | 200-500ms | 20-50ms | **10x faster** |
| **Read latency** | 5-10ms | 0.5-2ms | **5-10x faster** |
| **Concurrent opens** | ~1,000 (threading) | ~100,000+ (Tokio) | **100x more** |
| **GC pauses** | 50-2200ms | 0ms | **Eliminated** |
---
## 1. Rust FUSE Ecosystem
### Recommended: **fuser**
| Attribute | Value |
|-----------|-------|
| **Downloads** | 3.2M+ |
| **Maturity** | Production-ready |
| **Platforms** | Linux, macOS, FreeBSD |
| **Async** | Experimental (stable sync API) |
| **Used by** | AWS Mountpoint for S3 |
**API Example:**
```rust
use fuser::{Filesystem, Request, ReplyData};
impl Filesystem for BeetFS {
fn read(&self, _req: &Request, ino: u64, _fh: u64,
offset: i64, size: u32, _flags: i32,
_lock: Option<u64>, reply: ReplyData) {
let file = self.get_file(ino);
if offset < file.header_len {
// Return metadata from database (interpolated)
reply.data(&file.header[offset as usize..]);
} else {
// Return audio from original file (zero-copy via mmap)
let audio_offset = offset - file.header_len;
reply.data(&file.mmap[audio_offset as usize..]);
}
}
}
```
### Alternatives
| Library | Async | Maturity | Best For |
|---------|-------|----------|----------|
| **fuser** | Experimental | ⭐⭐⭐⭐⭐ | General purpose |
| **fuse3** | Native | ⭐⭐⭐⭐ | Async-heavy, Linux-only |
| **polyfuse** | Native | ⭐⭐⭐ | Custom control flow |
---
## 2. Rust Audio Metadata: **lofty**
Full feature parity with Python's mutagen:
| Feature | mutagen (Python) | lofty (Rust) |
|---------|------------------|--------------|
| FLAC Vorbis Comments | ✅ | ✅ |
| MP3 ID3v2 (all versions) | ✅ | ✅ |
| OGG Vorbis Comments | ✅ | ✅ |
| Opus metadata | ✅ | ✅ |
| In-memory manipulation | ✅ | ✅ |
| Header generation | ✅ | ✅ `dump_to()` |
| Picture/artwork | ✅ | ✅ |
**API Comparison:**
```python
# Python mutagen
audio = mutagen.File("song.flac")
audio['artist'] = 'New Artist'
audio['title'] = 'New Title'
audio.save()
```
```rust
// Rust lofty
let mut file = lofty::read_from_path("song.flac")?;
let tag = file.primary_tag_mut().unwrap();
tag.set_artist("New Artist".to_string());
tag.set_title("New Title".to_string());
tag.save_to_path("song.flac", WriteOptions::default())?;
```
**Header Generation (Critical for beetfs):**
```rust
// Generate FLAC header with modified tags WITHOUT writing to file
let mut buffer = Vec::new();
tag.dump_to(&mut buffer, WriteOptions::default())?;
// `buffer` contains serialized metadata header
```
---
## 3. Memory Benefits
### Python Object Overhead
| Python Type | Size | Notes |
|-------------|------|-------|
| Empty dict | 232 bytes | Base overhead |
| Dict entry | +184 bytes | Per key-value |
| Empty string | 49 bytes | Base overhead |
| Empty list | 56 bytes | Base overhead |
| Small int | 28 bytes | Even for `0` |
**Current beetfs FileHandler (Python):**
```
self.path → str → 49 + len(path) bytes
self.real_path → str → 49 + len(path) bytes
self.item → dict → 232 + entries
self.header → bytes → 33 + len(header)
self.music_data → bytes → 33 + len(audio) ← CRITICAL: full file!
self.inf → object → 100+ bytes
─────────────────────────────────────────
TOTAL: ~500 bytes + entire file in RAM
```
### Rust Struct Efficiency
```rust
struct FileHandler {
path: PathBuf, // 24 bytes (ptr+len+cap)
real_path: PathBuf, // 24 bytes
item_id: u64, // 8 bytes
header: Vec<u8>, // 24 bytes (ptr+len+cap) + header data
mmap: Mmap, // 24 bytes (NO file data in RAM!)
header_len: u64, // 8 bytes
audio_offset: u64, // 8 bytes
}
// TOTAL: ~120 bytes + header only (audio via mmap)
```
### Memory Comparison
| Scenario | Python | Rust | Savings |
|----------|--------|------|---------|
| 1 file (50MB) | ~50 MB | ~64 KB | **780x** |
| 10 files (50MB each) | ~500 MB | ~640 KB | **780x** |
| 100 files (50MB each) | ~5 GB | ~6.4 MB | **780x** |
| Library scan (1000 files) | **OOM** | ~64 MB | ∞ |
**Key insight**: Rust can use memory-mapped files (`mmap`) to serve audio data with zero copies, eliminating the need to load files into RAM.
---
## 4. Latency Benefits
### Python FUSE Bottlenecks
1. **Dict-to-struct conversion**: Every FUSE callback requires converting Python dicts to C structs
2. **GIL contention**: Single-threaded execution despite multi-core CPUs
3. **GC pauses**: Stop-the-world pauses of 50-2200ms under load
4. **Object allocation**: Creating Python objects for every I/O operation
### Rust FUSE Advantages
1. **Zero-cost abstractions**: No runtime overhead for type conversions
2. **No GIL**: True parallelism across all cores
3. **No GC**: Deterministic memory management, no pauses
4. **Stack allocation**: Small objects allocated on stack, not heap
### Benchmark Data
| Operation | Python FUSE | Rust FUSE | Improvement |
|-----------|-------------|-----------|-------------|
| File stat | 5-10ms | 0.5-1ms | **10x** |
| Small read | 5-10ms | 0.5-2ms | **5-10x** |
| Large read | 115 MB/s | 260+ MB/s | **2-3x** |
| Metadata lookup | 10ms | <1ms | **10x** |
### GC Pause Elimination
```
Python GC Pauses (measured):
├── P50: ~10ms
├── P95: ~50ms
├── P99: ~320ms
└── Max: ~2200ms (!)
Rust (no GC):
├── P50: ~0.5ms
├── P95: ~1ms
├── P99: ~2ms
└── Max: ~5ms (deterministic)
```
---
## 5. Concurrency Benefits
### Python Threading Limitations
```python
# Python (current beetfs)
server.multithreaded = 0 # Single-threaded!
# Even with threading enabled:
# - GIL prevents true parallelism
# - ~8MB per thread
# - OS limits: ~1000-2000 threads max
# - Context switch: 1-10μs (kernel)
```
### Rust Async (Tokio)
```rust
// Rust with Tokio
#[tokio::main]
async fn main() {
// Can handle 100K+ concurrent operations
// - ~2KB per task (4000x less than thread)
// - Work-stealing scheduler
// - Context switch: ~10ns (userspace)
}
```
| Metric | Python Threading | Rust Tokio |
|--------|------------------|------------|
| Memory per task | 8 MB | 2 KB |
| Max concurrent | ~1,000 | ~100,000+ |
| Context switch | 1-10μs | ~10ns |
| Parallelism | Blocked by GIL | True multi-core |
---
## 6. Zero-Copy I/O
### Python (Current)
```python
# Every read copies data through Python:
self.file_object.read() # syscall → kernel buffer
# kernel buffer → Python bytes object
# Python bytes → FUSE reply buffer
# = 2-3 copies per read
```
### Rust (Proposed)
```rust
// Memory-mapped file + zero-copy reply:
let mmap = unsafe { MmapOptions::new().map(&file)? };
fn read(&self, ..., reply: ReplyData) {
// Direct slice from mmap → FUSE kernel
reply.data(&self.mmap[offset..offset+size]);
// = 0 copies (kernel reads directly from mapped pages)
}
```
### I/O Comparison
| Scenario | Python | Rust | Benefit |
|----------|--------|------|---------|
| Serve 50MB file | 50MB copied to RAM | 0 bytes copied | **50MB saved** |
| 100 concurrent reads | 5GB buffers | ~0 (shared mmap) | **5GB saved** |
| Throughput | 115 MB/s | 260+ MB/s | **2.3x faster** |
---
## 7. Real-World Migration Results
### Case Studies
| Project | Metric | Python | Rust | Improvement |
|---------|--------|--------|------|-------------|
| API Service | Response time | 200ms | 8ms | **96% faster** |
| Data Pipeline | Processing | 3 hours | 4.5 min | **40x faster** |
| Web Backend | Memory | 1.2 GB | 180 MB | **85% less** |
| Trajectory Lib | Compute | baseline | 10x faster | **10x** |
### AWS Mountpoint for S3
- Built on **fuser** (Rust FUSE)
- Handles **terabits/sec** aggregate throughput
- Production-ready since 2024
- Validates Rust FUSE at scale
---
## 8. Migration Architecture
### Proposed Rust beetfs Structure
```
beetfs-rs/
├── Cargo.toml
├── src/
│ ├── main.rs # Entry point, mount logic
│ ├── lib.rs # Library root
│ ├── fs/
│ │ ├── mod.rs # FUSE filesystem impl
│ │ ├── tree.rs # Virtual directory tree (FSNode equivalent)
│ │ ├── file.rs # File handler with mmap
│ │ └── stat.rs # File attributes
│ ├── metadata/
│ │ ├── mod.rs # Metadata overlay logic
│ │ ├── flac.rs # FLAC header generation (using lofty)
│ │ ├── mp3.rs # MP3 ID3 header generation
│ │ └── db.rs # Database interface (SQLite or custom)
│ └── config.rs # Configuration (path templates, etc.)
└── tests/
├── fs_tests.rs
└── metadata_tests.rs
```
### Key Components
```rust
// Virtual directory tree (equivalent to FSNode)
pub struct VirtualTree {
root: Arc<RwLock<DirNode>>,
}
pub struct DirNode {
dirs: HashMap<OsString, Arc<RwLock<DirNode>>>,
files: HashMap<OsString, FileEntry>,
}
pub struct FileEntry {
inode: u64,
real_path: PathBuf,
metadata_id: i64, // Database reference
}
// File handler with memory-mapped audio
pub struct OpenFile {
header: Vec<u8>, // Generated header with DB metadata
header_len: usize,
mmap: Mmap, // Memory-mapped original file
audio_offset: usize, // Where audio starts in original
}
impl OpenFile {
pub fn read(&self, offset: usize, size: usize) -> &[u8] {
if offset < self.header_len {
// Return from generated header (DB metadata)
&self.header[offset..min(offset + size, self.header_len)]
} else {
// Return from mmap (original audio, zero-copy)
let audio_off = offset - self.header_len + self.audio_offset;
&self.mmap[audio_off..audio_off + size]
}
}
}
```
---
## 9. Migration Effort Estimate
### Timeline
| Phase | Duration | Deliverable |
|-------|----------|-------------|
| **1. Prototype** | 1-2 weeks | Basic FUSE mount, read-only |
| **2. Core features** | 2-3 weeks | Metadata overlay, FLAC support |
| **3. Full parity** | 2-3 weeks | MP3, write support, all fields |
| **4. Testing** | 1-2 weeks | Unit tests, integration tests |
| **5. Optimization** | 1-2 weeks | mmap, async, benchmarking |
**Total: 7-12 weeks**
### Skill Requirements
- Rust fundamentals (ownership, borrowing, lifetimes)
- FUSE protocol knowledge (from Python experience)
- Audio metadata formats (FLAC, ID3)
- Async Rust (Tokio) - optional for Phase 5
---
## 10. Risk Assessment
### Low Risk ✅
| Factor | Why Low Risk |
|--------|--------------|
| FUSE library | fuser is production-proven (AWS) |
| Metadata library | lofty has full mutagen parity |
| Core algorithm | Same logic, different language |
| File format support | FLAC/MP3/OGG all supported |
### Medium Risk ⚠️
| Factor | Mitigation |
|--------|------------|
| Learning curve | Existing Rust experience helps |
| Edge cases | Port Python tests to Rust |
| Async complexity | Start with sync API, add async later |
### Benefits vs Effort
```
Current Python Issues:
├── Memory: OOM on library scan → Fixed by mmap
├── Latency: 200-500ms file open → Fixed by zero-copy
├── GC pauses: 50-2200ms → Eliminated
├── Concurrency: single-threaded → Fixed by async
└── MP3 support: disabled → Implemented properly
Migration Effort: 7-12 weeks
Expected Lifetime: 5+ years
ROI: Highly positive
```
---
## 11. Recommendation
### ✅ **Proceed with Rust Migration**
**Justification:**
1. **10x memory reduction** via mmap (eliminates OOM)
2. **5-10x latency improvement** (eliminates blocking reads)
3. **GC pauses eliminated** (deterministic performance)
4. **100x concurrency** improvement (Tokio async)
5. **Production-proven** ecosystem (fuser + lofty)
6. **Reasonable effort** (7-12 weeks)
### Next Steps
1. **Set up Rust project** with fuser and lofty dependencies
2. **Port FSNode** to Rust VirtualTree
3. **Implement basic FUSE** operations (read, getattr, readdir)
4. **Add metadata overlay** with lofty for FLAC
5. **Add mmap** for zero-copy audio serving
6. **Benchmark** against Python implementation
7. **Add MP3/OGG** support
8. **Add async** with Tokio (optional)
### Dependencies
```toml
[dependencies]
fuser = "0.17"
lofty = "0.21"
memmap2 = "0.9"
tokio = { version = "1", features = ["full"], optional = true }
rusqlite = "0.31" # For beets DB compatibility
```
+899
View File
@@ -0,0 +1,899 @@
# MusicFS: Design Doc
**Authors:** [TBD]
**Status:** Draft
**Last Updated:** 2026-05-12
**Reviewers:** [TBD]
**Approvers:** [TBD]
**Requirements:** [requirements.md](requirements.md)
---
[TOC]
---
## 1. Abstract
MusicFS is a read-only FUSE filesystem that presents music libraries organized
by metadata (artist/album/track) rather than physical file paths. It supports
multiple origin storage backends (local, NFS, S3, SFTP), provides intelligent
caching with delta synchronization, and exposes a plugin architecture for
extensibility.
The system addresses limitations of the existing beetfs implementation:
- O(N) mount time → O(1) lazy loading
- Full file in RAM → streaming with content-addressable chunks
- Single origin → federated multi-origin with failover
- No offline support → cache-first with graceful degradation
Target users are media enthusiasts with large music collections (100K-10M+
tracks) distributed across multiple storage systems who want a unified,
metadata-organized view without modifying original files.
---
## 2. Background
### 2.1 Current State
The existing beetfs implementation is a Python 2.7 FUSE plugin for beets that:
- Presents a virtual filesystem organized by metadata templates
- Overlays metadata from beets database onto file headers
- Supports metadata writes back to the beets database
### 2.2 Pain Points
| Problem | Impact |
|---------|--------|
| O(N) mount time (5-120s for large libraries) | Unusable for large collections |
| Loads entire file into RAM on open | OOM risk, 50-100MB per file |
| Python GIL limits concurrency | Poor performance under load |
| No caching between sessions | Repeated work on every mount |
| Single local origin only | Can't federate across storage |
| No offline support | Unusable without origin access |
| Critical bugs (nested methods, tree building) | Non-functional |
### 2.3 Related Systems
| System | Relationship |
|--------|--------------|
| [beets](https://beets.io/) | Source of inspiration; potential import source |
| [rclone mount](https://rclone.org/commands/rclone_mount/) | Similar FUSE + remote storage; no metadata organization |
| [Plex/Jellyfin](https://jellyfin.org/) | Media servers with metadata; not filesystem-based |
---
## 3. Goals & Non-Goals
### 3.1 Goals
| ID | Goal | Success Metric |
|----|------|----------------|
| G1 | O(1) mount time | <500ms regardless of library size |
| G2 | Minimal memory footprint | <50MB idle, <500MB peak |
| G3 | Support multiple origins | ≥2 origins with automatic failover |
| G4 | Offline-first operation | Serve cached data when origin unavailable |
| G5 | Delta synchronization | >90% bandwidth reduction vs full sync |
| G6 | Plugin extensibility | Support custom origins, formats, metadata sources |
| G7 | Full-text search | Sub-second search across 1M+ tracks |
### 3.2 Design Requirements
The following quantitative requirements drive architectural decisions. Full
specification in [requirements.md](requirements.md).
#### 3.2.1 Latency Requirements
| Operation | Target | Maximum | Requirement |
|-----------|--------|---------|-------------|
| `stat()` cached | <1ms | 5ms | NFR-1.1 |
| `readdir()` cached | <10ms | 50ms | NFR-1.2 |
| `open()` cached | <5ms | 20ms | NFR-1.3 |
| `read()` cached | <1ms | 5ms | NFR-1.4 |
| `read()` cache miss (local) | <50ms | 200ms | NFR-1.5 |
| `read()` cache miss (remote) | <200ms | 1000ms | NFR-1.6 |
| Mount completion | <100ms | 500ms | NFR-1.7 |
| Search query (1M files) | <500ms | 1000ms | FR-14 |
**Design Response:**
- Lazy loading eliminates mount-time I/O → O(1) mount
- In-memory LRU cache for hot metadata → <1ms stat
- SQLite with indexes → O(log n) lookups
- Async I/O via tokio → non-blocking operations
#### 3.2.2 Throughput Requirements
| Metric | Target | Requirement |
|--------|--------|-------------|
| Sequential read (cached) | >500 MB/s | NFR-2.1 |
| Sequential read (local origin) | >200 MB/s | NFR-2.2 |
| Metadata ops/sec | >1000 | NFR-2.3 |
| Concurrent file handles | >1000 | NFR-2.4 |
**Design Response:**
- Memory-mapped chunk files → kernel-optimized reads
- No GIL (Rust) → true parallelism
- Async FUSE ops → handle many concurrent requests
#### 3.2.3 Scalability Requirements
| Metric | Target | Stretch | Requirement |
|--------|--------|---------|-------------|
| Library size | 1M files | 10M files | NFR-3.1, NFR-3.5 |
| Directory entries | 100K | 1M | NFR-3.2 |
| Concurrent clients | 10 | 100+ | NFR-3.6 |
| Mount time scaling | O(1) | O(1) | NFR-3.3 |
**Design Response:**
- Lazy tree loading → mount time independent of size
- SQLite indexes → O(log n) regardless of scale
- Streaming readdir → handle large directories
- Connection pooling → support many clients
#### 3.2.4 Resource Requirements
| Resource | Idle | Active (1K files) | Peak | Requirement |
|----------|------|-------------------|------|-------------|
| Memory | <50 MB | <200 MB | <500 MB | NFR-4.1-4.3 |
| Per-file overhead | - | <1 KB | - | NFR-4.4 |
| Metadata cache | - | 100 MB default | configurable | NFR-5.1 |
| Content cache | - | 10 GB default | configurable | NFR-5.2 |
**Design Response:**
- Streaming reads → never load full file in memory
- Content-addressed chunks → bounded cache with LRU eviction
- Metadata in SQLite → minimal per-file RAM overhead
#### 3.2.5 Efficiency Requirements
| Metric | Target | Requirement |
|--------|--------|-------------|
| Delta sync bandwidth reduction | >90% | NFR-6.4 |
| Cache hit rate (warm) | >95% | Derived |
| Deduplication ratio | >10% typical | FR-20 |
**Design Response:**
- CDC chunking → stable boundaries, minimal re-transfer
- Content-addressable storage → automatic deduplication
- Prefetch engine → anticipate access patterns
#### 3.2.6 Reliability Requirements
| Scenario | Behavior | Requirement |
|----------|----------|-------------|
| Origin offline | Serve cached data | NFR-7.1 |
| Network failure | Graceful degradation, no crash | NFR-7.2 |
| Failed operation | Retry with backoff (100ms, 500ms, 2s) | NFR-7.3 |
| Malformed audio | Skip file, log error, don't crash | NFR-7.4 |
| Chunk corruption | Detect via checksum, re-fetch | NFR-8.1, NFR-8.4 |
| Interrupted sync | Resume from last good state | NFR-8.3 |
| Unclean unmount | Recover on next mount | NFR-8.2 |
**Design Response:**
- Cache-first architecture → offline operation by default
- Origin federation with health checks → survive single origin failure
- xxHash checksums on all chunks → detect corruption
- WAL mode SQLite → ACID transactions, crash recovery
#### 3.2.7 Concurrent Access Requirements
| Scenario | Limit | Latency Impact | Requirement |
|----------|-------|----------------|-------------|
| Simultaneous open files | >1000 handles | None | NFR-2.4 |
| Parallel read ops | >100 concurrent | <2x p99 latency | Derived |
| Multiple clients | >10 (target 100+) | Linear degradation | NFR-3.6 |
| Readdir during sync | No blocking | Serve stale if needed | FR-9.2 |
**Design Response:**
- Async I/O (tokio) → non-blocking operations
- No GIL → true parallelism across cores
- Read-write locks on cache → readers don't block readers
- Stale-while-revalidate → serve cached during refresh
### 3.3 Non-Goals
| ID | Non-Goal | Rationale |
|----|----------|-----------|
| NG1 | Write to origin files | Read-only by design; preserves originals |
| NG2 | Transcoding | Out of scope for MVP; plugin possible later |
| NG3 | Video file support | Focus on audio; deferred to future |
| NG4 | Distributed/clustered mode | Single-node for MVP; architecture supports later |
| NG5 | Mobile app | CLI/daemon only; filesystem interface |
---
## 4. Proposed Design
### 4.1 High-Level Architecture
```plantuml
@startuml
!theme plain
skinparam componentStyle rectangle
package "User Space" {
[Media Players\n(mpv, VLC, Plex)] as Apps
package "MusicFS Daemon" {
[FUSE Interface] as FUSE
[Control API] as Control
[Metrics] as Metrics
package "Core Services" {
[Virtual Path\nResolver] as VPR
[Event Bus] as Events
[Search Engine\n(tantivy)] as Search
}
package "Plugin Host" {
[Origin\nPlugins] as OriginPlugins
[Metadata\nPlugins] as MetaPlugins
[Format\nPlugins] as FormatPlugins
}
package "Storage Layer" {
[Content-Addressable\nStore (CAS)] as CAS
database "SQLite\n(metadata)" as SQLite
database "sled\n(chunks)" as Sled
}
[Origin\nFederation] as Federation
}
}
package "Origins (Read-Only)" {
[Local FS] as Local
[NFS] as NFS
[S3] as S3
[SFTP] as SFTP
}
Apps --> FUSE : POSIX
FUSE --> VPR
VPR --> Events
VPR --> Search
VPR --> CAS
CAS --> SQLite
CAS --> Sled
VPR --> Federation
Federation --> OriginPlugins
OriginPlugins --> Local
OriginPlugins --> NFS
OriginPlugins --> S3
OriginPlugins --> SFTP
Control --> Events
Metrics --> Events
@enduml
```
### 4.2 Component Overview
| Component | Responsibility | Technology |
|-----------|---------------|------------|
| FUSE Interface | Translate POSIX ops to internal calls | fuser (Rust) |
| Virtual Path Resolver | Map virtual ↔ real paths | Custom |
| Event Bus | Decouple components, enable observability | tokio broadcast |
| Search Engine | Full-text metadata search | tantivy |
| Plugin Host | Load/manage plugins | Native + WASM |
| CAS | Content-addressed chunk storage | Custom + sled |
| Origin Federation | Multi-origin routing with failover | Custom |
### 4.3 Detailed Design
#### 4.3.1 Virtual Path Resolution
The resolver maps metadata-based virtual paths to real origin paths.
```plantuml
@startuml
!theme plain
participant "FUSE" as F
participant "VirtualPathResolver" as VPR
participant "MetadataIndex" as MI
participant "TreeCache" as TC
participant "OriginFederation" as OF
F -> VPR : lookup("/Metallica/72 Seasons/01.flac")
VPR -> TC : get_cached(path)
alt cache hit
TC --> VPR : CachedEntry
else cache miss
VPR -> MI : query(artist="Metallica", album="72 Seasons", track=1)
MI --> VPR : FileRecord { origin_id, real_path, metadata }
VPR -> TC : store(path, entry)
end
VPR -> OF : resolve_origin(origin_id)
OF --> VPR : OriginHandle
VPR --> F : ResolvedPath { origin, real_path, inode }
@enduml
```
**Path Template Grammar:**
```
template = segment ("/" segment)*
segment = (literal | variable)+
variable = "$" identifier
identifier = "artist" | "album" | "title" | "track" | "year" | "genre"
| "format" | "format_upper" | "disc"
```
**Default Template:**
```
$artist/$album ($year) [$format_upper]/$track - $title.$format
```
#### 4.3.2 Content-Addressable Store (CAS)
All file content is stored as content-addressed chunks, enabling deduplication
and efficient delta sync.
```plantuml
@startuml
!theme plain
package "Content-Addressable Store" {
component "Chunk Manager" as CM
component "CDC Chunker\n(FastCDC)" as CDC
component "Hash Index\n(xxHash64)" as Hash
database "Chunk Files\n~/.cache/musicfs/chunks/" as Chunks
database "Index DB\n(sled)" as Index
CM --> CDC : chunk data
CDC --> Hash : compute hash
Hash --> Index : store hash → location
CM --> Chunks : write chunk file
}
note right of CDC
Avg chunk: 64KB
Min: 16KB, Max: 256KB
Stable boundaries for delta sync
end note
@enduml
```
**Chunk Storage Layout:**
```
~/.cache/musicfs/
├── chunks/
│ ├── aa/
│ │ ├── aa1b2c3d4e5f6789... (64KB chunk)
│ │ └── aa9f8e7d6c5b4a32...
│ ├── ab/
│ └── ... (256 subdirs for distribution)
├── metadata.db (SQLite: file metadata, tree cache)
├── search.idx/ (tantivy: full-text index)
└── chunks.sled/ (sled: hash → chunk location)
```
#### 4.3.3 Origin Federation
Multiple origins are managed with priority-based routing and health tracking.
```plantuml
@startuml
!theme plain
participant "VirtualPathResolver" as VPR
participant "OriginFederation" as OF
participant "HealthChecker" as HC
participant "Origin[Local]" as O1
participant "Origin[NFS]" as O2
participant "Origin[S3]" as O3
VPR -> OF : read(real_path, offset, size)
OF -> OF : select_origin(priority, health)
alt Origin[Local] healthy (pri=1)
OF -> O1 : read()
O1 --> OF : data
else Origin[Local] unhealthy, try NFS (pri=2)
OF -> O2 : read()
alt success
O2 --> OF : data
else failure
OF -> O3 : read()
O3 --> OF : data
end
end
OF --> VPR : data
note over HC
Background health checks
every 30s per origin
end note
@enduml
```
**Origin Configuration:**
```toml
[[origins]]
id = "local"
type = "local"
path = "/mnt/nas/music"
priority = 1
[[origins]]
id = "backup"
type = "s3"
bucket = "music-backup"
priority = 2
```
#### 4.3.4 Plugin System
Plugins extend functionality without modifying core code.
```plantuml
@startuml
!theme plain
interface "Plugin" {
+name(): String
+version(): Version
+init(config)
+shutdown()
}
interface "OriginPlugin" {
+list_dir(path): Vec<DirEntry>
+read(path, offset, size): Vec<u8>
+stat(path): FileStat
+watch(path, callback): WatchHandle
}
interface "MetadataPlugin" {
+extract(data, format): Metadata
+can_handle(format): bool
}
interface "FormatPlugin" {
+extensions(): Vec<String>
+parse_header(data): AudioHeader
+synthesize_header(metadata): Vec<u8>
}
Plugin <|-- OriginPlugin
Plugin <|-- MetadataPlugin
Plugin <|-- FormatPlugin
class "LocalFSPlugin" implements OriginPlugin
class "S3Plugin" implements OriginPlugin
class "SymphoniaPlugin" implements MetadataPlugin
class "FlacPlugin" implements FormatPlugin
class "Mp3Plugin" implements FormatPlugin
@enduml
```
**Plugin Loading:**
1. **Built-in:** Compiled into binary (Local, S3, SFTP, symphonia)
2. **Native:** Dynamic libraries (`.so`/`.dylib`) loaded at runtime
3. **WASM:** Sandboxed plugins via wasmtime (future)
#### 4.3.5 Data Flow: Read Operation
```plantuml
@startuml
!theme plain
|FUSE|
start
:receive read(path, offset, size);
|VirtualPathResolver|
:resolve virtual path to real path;
:lookup file metadata;
|CAS|
:compute chunk range for [offset, offset+size];
if (all chunks cached?) then (yes)
:read from local chunk files;
else (no)
|OriginFederation|
:select healthy origin by priority;
:fetch missing byte range;
|CAS|
:chunk fetched data (CDC);
:store chunks by hash;
:update chunk manifest;
endif
|EventBus|
:emit FileAccessed event;
|FUSE|
:return data to application;
stop
@enduml
```
#### 4.3.6 Data Schema
**Metadata Index (SQLite):**
```sql
CREATE TABLE files (
id INTEGER PRIMARY KEY,
origin_id TEXT NOT NULL,
real_path TEXT NOT NULL,
virtual_path TEXT NOT NULL,
-- Metadata (see FR-6 in requirements.md)
title TEXT,
artist TEXT,
album TEXT,
album_artist TEXT,
genre TEXT,
year INTEGER,
track INTEGER,
disc INTEGER,
duration_ms INTEGER,
bitrate INTEGER,
sample_rate INTEGER,
format TEXT,
-- Sync state
origin_mtime INTEGER,
origin_size INTEGER,
content_hash TEXT,
chunk_manifest BLOB, -- msgpack: [(chunk_hash, offset, size)]
last_sync INTEGER,
UNIQUE(origin_id, real_path)
);
CREATE INDEX idx_virtual ON files(virtual_path);
CREATE INDEX idx_artist_album ON files(artist, album);
CREATE INDEX idx_content_hash ON files(content_hash);
CREATE TABLE artwork (
id INTEGER PRIMARY KEY,
file_id INTEGER REFERENCES files(id),
art_type TEXT, -- 'front', 'back'
chunk_hash TEXT, -- reference to CAS
width INTEGER,
height INTEGER,
UNIQUE(file_id, art_type)
);
CREATE TABLE collections (
id INTEGER PRIMARY KEY,
name TEXT UNIQUE,
query_json TEXT, -- smart collection query
created_at INTEGER
);
```
#### 4.3.7 Control API
**Unix Socket Protocol (JSON-RPC 2.0):**
```json
// Request: Get cache statistics
{"jsonrpc": "2.0", "method": "cache.stats", "id": 1}
// Response
{
"jsonrpc": "2.0",
"id": 1,
"result": {
"hits": 15234,
"misses": 421,
"hit_rate": 0.973,
"chunks_stored": 84521,
"chunks_unique": 71203,
"dedup_ratio": 0.157,
"size_bytes": 5368709120
}
}
// Request: Search
{"jsonrpc": "2.0", "method": "search", "params": {"query": "metallica"}, "id": 2}
// Request: Refresh origin
{"jsonrpc": "2.0", "method": "origin.rescan", "params": {"id": "local"}, "id": 3}
```
**CLI Interface:**
```bash
musicfs mount /mnt/music # Mount filesystem
musicfs status # Show daemon status
musicfs cache stats # Cache statistics
musicfs cache clear --origin=local # Clear cache for origin
musicfs search "metallica heavy" # Search library
musicfs origin list # List origins and health
musicfs origin rescan local # Force rescan
```
---
## 5. Cross-Cutting Concerns
### 5.1 Security & Privacy
| Concern | Mitigation |
|---------|------------|
| Credential storage | Use system keyring (secret-service) or env vars; never in config file |
| Credential exposure | Redact from logs; exclude from `/proc/cmdline` |
| Cache at rest | Optional encryption via age/libsodium (P3 requirement) |
| Plugin sandboxing | WASM plugins run in wasmtime sandbox; native plugins require trust |
| Access control | Respect origin permissions; run as unprivileged user |
| No PII handling | Filesystem metadata only; no user data collected |
### 5.2 Observability
**Metrics (Prometheus format):**
```
musicfs_fuse_ops_total{op="read"} 152341
musicfs_fuse_ops_total{op="readdir"} 8234
musicfs_fuse_latency_seconds{op="read",quantile="0.99"} 0.004
musicfs_cache_hits_total 142107
musicfs_cache_misses_total 10234
musicfs_cache_size_bytes 5368709120
musicfs_origin_health{origin="local"} 1
musicfs_origin_health{origin="s3"} 0
musicfs_sync_files_changed{origin="local"} 15
```
**Logging Levels:**
| Level | Content |
|-------|---------|
| ERROR | Unrecoverable failures, data corruption |
| WARN | Recoverable failures, origin timeouts |
| INFO | Mount/unmount, sync completion, config reload |
| DEBUG | Cache hits/misses, origin selection |
| TRACE | Individual FUSE operations, chunk I/O |
**Golden Signals Dashboard:**
1. **Latency:** p50/p95/p99 for read, stat, readdir
2. **Traffic:** FUSE ops/sec, bytes read/sec
3. **Errors:** Origin failures, cache corruption
4. **Saturation:** Cache fullness, open file handles
### 5.3 Scalability & Performance
**Expected Load:**
| Metric | Target | Maximum |
|--------|--------|---------|
| Library size | 1M files | 10M files |
| Concurrent clients | 10 | 100+ |
| FUSE ops/sec | 1,000 | 10,000 |
| Read throughput | 500 MB/s | 1 GB/s |
**Scaling Strategy:**
- **Horizontal:** Not supported (single daemon per mountpoint)
- **Vertical:** Increase cache size, add origins
**Resource Requirements:**
| Resource | Minimum | Recommended |
|----------|---------|-------------|
| CPU | 1 core | 4 cores |
| RAM | 256 MB | 2 GB |
| Disk (cache) | 1 GB | 50 GB |
| Network | 10 Mbps | 1 Gbps |
### 5.4 Testing Plan
| Test Type | Scope | Tools |
|-----------|-------|-------|
| Unit | Individual components | cargo test |
| Integration | Component interaction | cargo test --features integration |
| E2E | Full FUSE operations | pytest + real mount |
| Performance | Latency, throughput | criterion.rs, custom benchmarks |
| Stress | High load, large libraries | locust, custom generators |
| Chaos | Origin failures, network issues | toxiproxy |
**Test Matrix:**
```
Origins: [local, s3, sftp] × [healthy, degraded, offline]
Cache: [cold, warm, full]
Library: [100, 10K, 1M, 10M] files
Operations: [mount, readdir, stat, read, search]
```
---
## 6. Alternatives Considered
### 6.1 Alternative A: Extend beetfs (Python)
**Description:** Fix bugs in existing beetfs, add features incrementally.
**Rejected Because:**
- Python GIL fundamentally limits concurrency
- Python 2.7 EOL; migration to Python 3 substantial
- Architecture (full file in RAM) requires rewrite anyway
- No async I/O support in fuse-python
### 6.2 Alternative B: Use rclone mount
**Description:** Use rclone's FUSE mount with VFS caching.
**Rejected Because:**
- No metadata-based virtual path organization
- No metadata overlay functionality
- Limited plugin extensibility
- Would require forking and heavy modification
### 6.3 Alternative C: Build as Plex/Jellyfin Plugin
**Description:** Extend existing media server with virtual filesystem view.
**Rejected Because:**
- Tied to specific media server
- Not a true filesystem (no POSIX interface)
- Heavy runtime dependency
- Different use case (streaming vs filesystem)
### 6.4 Alternative D: Go Implementation
**Description:** Implement in Go using go-fuse.
**Considered Trade-offs:**
| Aspect | Rust | Go |
|--------|------|-----|
| Memory safety | Compile-time | GC pauses |
| Concurrency | async/await, no GC | goroutines, GC |
| FUSE library | fuser (mature) | go-fuse (mature) |
| Learning curve | Steeper | Gentler |
| Binary size | Smaller | Larger |
**Decision:** Rust chosen for zero-cost abstractions, no GC pauses during I/O,
and better fit for systems programming.
---
## 7. Implementation Plan
### 7.1 Phase 1: MVP (4 weeks)
**Goal:** Basic functional filesystem with single origin.
| Week | Deliverables |
|------|--------------|
| 1 | Project setup, FUSE skeleton, local origin plugin |
| 2 | Metadata extraction (symphonia), SQLite schema |
| 3 | Virtual path resolver, tree cache, basic readdir/stat/read |
| 4 | CAS implementation, chunk caching, integration tests |
**Exit Criteria:**
- Mount and browse local music library
- Play audio files through mounted filesystem
- Cache persists across restarts
### 7.2 Phase 2: Delta Sync & Multi-Origin (3 weeks)
**Goal:** Efficient synchronization and origin federation.
| Week | Deliverables |
|------|--------------|
| 5 | CDC chunking (FastCDC), delta detection |
| 6 | Origin federation, priority routing, health checks |
| 7 | S3 origin plugin, SFTP origin plugin |
**Exit Criteria:**
- Delta sync achieves >90% bandwidth reduction
- Automatic failover between origins
- Remote origins functional
### 7.3 Phase 3: Search & Smart Features (2 weeks)
**Goal:** Full-text search and intelligent caching.
| Week | Deliverables |
|------|--------------|
| 8 | tantivy integration, search indexing, `/.search/` virtual dir |
| 9 | Smart collections, prefetch engine, access pattern learning |
**Exit Criteria:**
- Search returns results in <1s for 1M tracks
- Prefetch reduces cache misses by >50%
### 7.4 Phase 4: Plugin System & Polish (2 weeks)
**Goal:** Extensibility and production readiness.
| Week | Deliverables |
|------|--------------|
| 10 | Plugin host, plugin API stabilization, example plugins |
| 11 | Control API, metrics, documentation, packaging |
**Exit Criteria:**
- Custom origin plugin loadable at runtime
- Prometheus metrics exported
- systemd service functional
### 7.5 Rollout Strategy
```plantuml
@startuml
!theme plain
[*] --> Alpha
Alpha --> Beta : Internal testing complete
Beta --> GA : Community testing complete
state Alpha {
[*] --> DevTesting
DevTesting --> DogFood : Core features work
}
state Beta {
[*] --> LimitedRelease
LimitedRelease --> PublicBeta : No critical bugs
}
state GA {
[*] --> Stable
}
note right of Alpha : 2-4 weeks\nDevelopers only
note right of Beta : 4-8 weeks\nEarly adopters
note right of GA : Stable releases
@enduml
```
**Feature Flags:**
```toml
[features]
search_enabled = true
smart_collections = false # Beta
wasm_plugins = false # Experimental
```
**Rollback:** Binary replacement + cache clear; no data migration needed.
---
## 8. Glossary & References
### 8.1 Glossary
| Term | Definition |
|------|------------|
| **CAS** | Content-Addressable Store; data stored/retrieved by hash |
| **CDC** | Content-Defined Chunking; chunking with stable boundaries |
| **FUSE** | Filesystem in Userspace; kernel interface for user-space filesystems |
| **Origin** | Source storage backend (local, S3, NFS, etc.) |
| **Virtual Path** | Metadata-derived path shown to users |
| **Real Path** | Actual path on origin storage |
### 8.2 References
| Document | Link |
|----------|------|
| Requirements Specification | [requirements.md](requirements.md) |
| beetfs (Original) | [beetsplug/beetFs.py](../../beetsplug/beetFs.py) |
| beetfs Features | [v1/features.md](../v1/features.md) |
| fuser (Rust FUSE) | https://github.com/cberner/fuser |
| tantivy (Search) | https://github.com/quickwit-oss/tantivy |
| symphonia (Audio) | https://github.com/pdrat/symphonia |
| FastCDC | https://github.com/nlfiedler/fastcdc-rs |
| wasmtime | https://wasmtime.dev/ |
### 8.3 Dependencies
| Crate | Version | Purpose |
|-------|---------|---------|
| fuser | 0.14+ | FUSE interface |
| tokio | 1.x | Async runtime |
| rusqlite | 0.31+ | SQLite bindings |
| sled | 0.34+ | Embedded key-value store |
| tantivy | 0.21+ | Full-text search |
| symphonia | 0.5+ | Audio metadata extraction |
| fastcdc | 3.x | Content-defined chunking |
| xxhash-rust | 0.8+ | Fast hashing |
| serde | 1.x | Serialization |
| toml | 0.8+ | Configuration |
| tracing | 0.1+ | Logging/instrumentation |
| metrics | 0.22+ | Prometheus metrics |
+649
View File
@@ -0,0 +1,649 @@
# Music Library FUSE Filesystem - Requirements Specification
**Version**: 1.0
**Date**: 2026-05-12
**Status**: Draft
## 1. Introduction
### 1.1 Purpose
This document specifies the requirements for a FUSE-based virtual filesystem that presents a music library organized by metadata. The system overlays metadata onto audio files without modifying originals and operates as a read-only client against the origin storage.
### 1.2 Scope
The system provides:
- Virtual filesystem accessible via standard POSIX operations
- Metadata-based directory structure (artist/album/track)
- Local caching with delta synchronization
- Support for local and remote origin storage
### 1.3 Definitions
| Term | Definition |
|------|------------|
| **Origin** | The source storage containing original audio files (local FS, NFS, S3, etc.) |
| **Virtual path** | The metadata-derived path shown to users (e.g., `/Artist/Album/Track.flac`) |
| **Real path** | The actual path on origin storage |
| **Metadata overlay** | Serving synthesized file headers from cached metadata |
| **CDC** | Content-Defined Chunking - algorithm for stable file segmentation |
---
## 2. System Overview
```
┌─────────────────────────────────────────────────────────────────┐
│ User Applications │
│ (mpv, Rhythmbox, Plex, etc.) │
└─────────────────────────────┬───────────────────────────────────┘
│ POSIX (read-only)
┌─────────────────────────────────────────────────────────────────┐
│ FUSE Interface │
├─────────────────────────────────────────────────────────────────┤
│ Plugin Host │
│ ┌─────────────┐ ┌─────────────┐ ┌─────────────┐ │
│ │ Origin │ │ Metadata │ │ Format │ │
│ │ Plugins │ │ Plugins │ │ Plugins │ │
│ └─────────────┘ └─────────────┘ └─────────────┘ │
├─────────────────────────────────────────────────────────────────┤
│ Core Services │
│ ┌───────────┐ ┌───────────┐ ┌───────────┐ ┌───────────┐ │
│ │ Virtual │ │ Event │ │ Search │ │ Control │ │
│ │ Path │ │ Bus │ │ Index │ │ API │ │
│ │ Resolver │ │ │ │ │ │ │ │
│ └───────────┘ └───────────┘ └───────────┘ └───────────┘ │
├─────────────────────────────────────────────────────────────────┤
│ Storage Layer │
│ ┌─────────────────────────────────────────────────────────┐ │
│ │ Content-Addressable Chunk Store │ │
│ │ ┌──────────┐ ┌──────────┐ ┌──────────┐ │ │
│ │ │ Metadata │ │ Content │ │ Tree │ │ │
│ │ │ Cache │ │ Chunks │ │ Cache │ │ │
│ │ │ (SQLite) │ │ (CAS) │ │ │ │ │
│ │ └──────────┘ └──────────┘ └──────────┘ │ │
│ └─────────────────────────────────────────────────────────┘ │
├─────────────────────────────────────────────────────────────────┤
│ Origin Federation │
│ ┌─────────┐ ┌─────────┐ ┌─────────┐ ┌─────────┐ │
│ │ Local │ │ NFS │ │ S3 │ │ SFTP │ │
│ │ FS │ │ │ │ │ │ │ │
│ └─────────┘ └─────────┘ └─────────┘ └─────────┘ │
└─────────────────────────────────────────────────────────────────┘
│ read-only
┌─────────────────────────────────────────────────────────────────┐
│ Origin Storage(s) │
│ (original audio files) │
└─────────────────────────────────────────────────────────────────┘
```
---
## 3. Functional Requirements
### 3.1 Filesystem Operations
#### FR-1: Mount/Unmount
| ID | Requirement |
|----|-------------|
| FR-1.1 | The system SHALL mount as a FUSE filesystem at a user-specified mountpoint |
| FR-1.2 | The system SHALL return control to the caller within 500ms of mount initiation |
| FR-1.3 | The system SHALL unmount cleanly via `fusermount -u` |
| FR-1.4 | The system SHALL release all resources (file handles, connections) on unmount |
#### FR-2: Directory Operations
| ID | Requirement |
|----|-------------|
| FR-2.1 | The system SHALL present files organized by metadata path format |
| FR-2.2 | The system SHALL support configurable path templates (e.g., `$artist/$album/$track - $title.$format`) |
| FR-2.3 | The system SHALL return directory listings via `readdir()` |
| FR-2.4 | The system SHALL support nested directory traversal to arbitrary depth |
| FR-2.5 | The system SHALL handle directories with 100,000+ entries |
#### FR-3: File Operations (Read)
| ID | Requirement |
|----|-------------|
| FR-3.1 | The system SHALL support `open()` for reading |
| FR-3.2 | The system SHALL support `read()` with arbitrary offset and size |
| FR-3.3 | The system SHALL support `seek()` operations for random access |
| FR-3.4 | The system SHALL return file attributes via `stat()` / `fstat()` |
| FR-3.5 | The system SHALL support concurrent reads from multiple processes |
#### FR-4: Read-Only Constraint
| ID | Requirement |
|----|-------------|
| FR-4.1 | The system SHALL NOT modify original files on the origin storage |
| FR-4.2 | The system SHALL NOT push any changes to the origin server |
| FR-4.3 | The system SHALL return `EROFS` (Read-only filesystem) for write operations |
| FR-4.4 | The system SHALL return `EROFS` for `create()`, `mkdir()`, `unlink()`, `rmdir()` |
| FR-4.5 | The system SHALL return `EROFS` for `rename()`, `chmod()`, `chown()`, `truncate()` |
### 3.2 Metadata Handling
#### FR-5: Metadata Overlay
| ID | Requirement |
|----|-------------|
| FR-5.1 | The system SHALL extract metadata from audio files on first access |
| FR-5.2 | The system SHALL cache extracted metadata in a local database |
| FR-5.3 | The system SHALL serve file headers with metadata from cache |
| FR-5.4 | The system SHALL support FLAC Vorbis comments |
| FR-5.5 | The system SHALL support MP3 ID3v2 tags |
| FR-5.6 | The system SHOULD support additional formats (OGG, M4A, OPUS) |
#### FR-6: Metadata Fields
| ID | Requirement |
|----|-------------|
| FR-6.1 | The system SHALL extract and cache: title, artist, album, genre |
| FR-6.2 | The system SHALL extract and cache: year, track number, disc number |
| FR-6.3 | The system SHALL extract and cache: duration, bitrate, sample rate |
| FR-6.4 | The system SHOULD extract: composer, album artist, lyrics |
| FR-6.5 | The system SHALL handle missing metadata gracefully with defaults |
### 3.3 Caching
#### FR-7: Metadata Cache
| ID | Requirement |
|----|-------------|
| FR-7.1 | The system SHALL persist metadata cache across restarts |
| FR-7.2 | The system SHALL store metadata in SQLite database |
| FR-7.3 | The system SHALL index by both virtual path and real path |
| FR-7.4 | The system SHALL invalidate cache entries when origin file changes |
#### FR-8: Content Cache
| ID | Requirement |
|----|-------------|
| FR-8.1 | The system SHALL cache file content in fixed-size chunks |
| FR-8.2 | The system SHALL use content-defined chunking for cache efficiency |
| FR-8.3 | The system SHALL store chunk hashes for delta detection |
| FR-8.4 | The system SHALL evict chunks under memory/disk pressure |
#### FR-9: Directory Tree Cache
| ID | Requirement |
|----|-------------|
| FR-9.1 | The system SHALL cache directory listings locally |
| FR-9.2 | The system SHALL serve `readdir()` from cache without origin access |
| FR-9.3 | The system SHALL refresh tree cache based on configurable policy |
| FR-9.4 | The system SHALL support forced refresh via signal or special file |
### 3.4 Synchronization
#### FR-10: Change Detection
| ID | Requirement |
|----|-------------|
| FR-10.1 | The system SHALL detect changes to origin files |
| FR-10.2 | The system SHALL use inotify for local filesystem origins |
| FR-10.3 | The system SHALL use polling for remote origins without push support |
| FR-10.4 | The system SHALL compare mtime and size for change detection |
| FR-10.5 | The system SHALL support content-hash verification on demand |
#### FR-11: Delta Sync
| ID | Requirement |
|----|-------------|
| FR-11.1 | The system SHALL download only changed portions of files |
| FR-11.2 | The system SHALL use CDC to identify changed chunks |
| FR-11.3 | The system SHALL preserve unchanged chunks in cache |
| FR-11.4 | The system SHALL handle file additions and deletions |
### 3.5 Origin Support
#### FR-12: Origin Types
| ID | Requirement |
|----|-------------|
| FR-12.1 | The system SHALL support local filesystem as origin |
| FR-12.2 | The system SHOULD support NFS mounted filesystems |
| FR-12.3 | The system SHOULD support SMB/CIFS shares |
| FR-12.4 | The system SHOULD support S3-compatible object storage |
| FR-12.5 | The system SHOULD support SFTP servers |
| FR-12.6 | The system SHALL provide pluggable origin interface |
#### FR-13: Multiple Origins [P0]
| ID | Requirement |
|----|-------------|
| FR-13.1 | The system SHALL support multiple simultaneous origins |
| FR-13.2 | The system SHALL present unified virtual tree across origins |
| FR-13.3 | The system SHALL support origin priority/preference ordering |
| FR-13.4 | The system SHALL handle duplicate files across origins |
| FR-13.5 | The system SHALL support per-origin configuration |
### 3.6 Search & Discovery
#### FR-14: Full-Text Search [P1]
| ID | Requirement |
|----|-------------|
| FR-14.1 | The system SHALL index metadata for full-text search |
| FR-14.2 | The system SHALL expose search via virtual directory (`/.search/query/`) |
| FR-14.3 | The system SHALL support fuzzy matching |
| FR-14.4 | The system SHOULD support search by audio fingerprint |
#### FR-15: Smart Collections [P1]
| ID | Requirement |
|----|-------------|
| FR-15.1 | The system SHALL support query-based virtual folders |
| FR-15.2 | The system SHALL support saved searches as directories |
| FR-15.3 | The system SHALL support dynamic playlists (recently played, most played) |
| FR-15.4 | The system SHOULD support user-defined metadata fields |
### 3.7 Album Art
#### FR-16: Cover Art Handling [P1]
| ID | Requirement |
|----|-------------|
| FR-16.1 | The system SHALL extract embedded album art |
| FR-16.2 | The system SHALL expose art as virtual files (`/Artist/Album/cover.jpg`) |
| FR-16.3 | The system SHALL cache artwork separately from audio |
| FR-16.4 | The system SHALL support multiple art sizes (thumbnail, medium, full) |
| FR-16.5 | The system SHOULD fetch missing art from online sources |
### 3.8 Control & API
#### FR-17: Control Interface [P0]
| ID | Requirement |
|----|-------------|
| FR-17.1 | The system SHALL expose control via Unix socket |
| FR-17.2 | The system SHOULD expose REST/gRPC API |
| FR-17.3 | The system SHALL support cache management commands (clear, refresh, stats) |
| FR-17.4 | The system SHALL support runtime configuration changes |
| FR-17.5 | The system SHALL support graceful shutdown with drain |
#### FR-18: Event System [P0]
| ID | Requirement |
|----|-------------|
| FR-18.1 | The system SHALL emit events for file access |
| FR-18.2 | The system SHALL support webhook notifications |
| FR-18.3 | The system SHOULD support event streaming (SSE/WebSocket) |
| FR-18.4 | The system SHALL log access patterns for analysis |
### 3.9 Caching Enhancements
#### FR-19: Intelligent Prefetching [P1]
| ID | Requirement |
|----|-------------|
| FR-19.1 | The system SHALL learn access patterns |
| FR-19.2 | The system SHALL support playlist-aware prefetching |
| FR-19.3 | The system SHOULD support time-based prefetching |
| FR-19.4 | The system SHALL support manual prefetch hints (`/.prefetch/path/`) |
#### FR-20: Content-Addressable Storage [P0]
| ID | Requirement |
|----|-------------|
| FR-20.1 | The system SHALL store chunks by content hash |
| FR-20.2 | The system SHALL detect identical files across library |
| FR-20.3 | The system SHALL report deduplication statistics |
| FR-20.4 | The system SHALL enable cache sharing via content addressing |
### 3.10 Integration
#### FR-21: Metadata Sources [P1]
| ID | Requirement |
|----|-------------|
| FR-21.1 | The system SHOULD integrate with MusicBrainz |
| FR-21.2 | The system SHOULD integrate with Discogs |
| FR-21.3 | The system SHOULD integrate with Last.fm |
| FR-21.4 | The system SHOULD support AcoustID fingerprinting |
| FR-21.5 | The system SHALL support custom metadata plugins |
#### FR-22: Import & Migration [P1]
| ID | Requirement |
|----|-------------|
| FR-22.1 | The system SHALL import from beets database |
| FR-22.2 | The system SHOULD import from iTunes/Apple Music library |
| FR-22.3 | The system SHALL export library metadata |
### 3.11 Extensibility
#### FR-23: Plugin System [P0]
| ID | Requirement |
|----|-------------|
| FR-23.1 | The system SHALL support loadable plugins |
| FR-23.2 | The system SHALL define stable plugin API |
| FR-23.3 | The system SHALL support plugins for: origins, metadata extractors, formats |
| FR-23.4 | The system SHOULD support WASM plugins for sandboxed execution |
| FR-23.5 | The system SHALL provide plugin lifecycle management (load, unload, reload) |
#### FR-24: Format Extensibility [P1]
| ID | Requirement |
|----|-------------|
| FR-24.1 | The system SHALL support pluggable codec modules |
| FR-24.2 | The system SHOULD support audiobook formats (M4B, chapters) |
| FR-24.3 | The system SHALL allow format plugins to register file extensions |
### 3.12 High Availability [P3]
#### FR-25: Resilience
| ID | Requirement |
|----|-------------|
| FR-25.1 | The system SHOULD support active-passive failover |
| FR-25.2 | The system SHOULD support read replicas |
| FR-25.3 | The system SHALL support zero-downtime upgrades |
| FR-25.4 | The system SHALL support cache backup/restore |
| FR-25.5 | The system SHALL validate cache integrity on startup |
---
## 4. Non-Functional Requirements
### 4.1 Performance
#### NFR-1: Latency
| ID | Requirement | Target | Maximum |
|----|-------------|--------|---------|
| NFR-1.1 | `stat()` on cached file | <1ms | 5ms |
| NFR-1.2 | `readdir()` on cached directory | <10ms | 50ms |
| NFR-1.3 | `open()` on cached file | <5ms | 20ms |
| NFR-1.4 | `read()` from cache | <1ms | 5ms |
| NFR-1.5 | `read()` cache miss (local origin) | <50ms | 200ms |
| NFR-1.6 | `read()` cache miss (remote origin) | <200ms | 1000ms |
| NFR-1.7 | Mount completion | <100ms | 500ms |
#### NFR-2: Throughput
| ID | Requirement | Target |
|----|-------------|--------|
| NFR-2.1 | Sequential read throughput (cached) | >500 MB/s |
| NFR-2.2 | Sequential read throughput (local origin) | >200 MB/s |
| NFR-2.3 | Metadata operations per second | >1000 ops/s |
| NFR-2.4 | Concurrent file handles | >1000 |
#### NFR-3: Scalability
| ID | Requirement |
|----|-------------|
| NFR-3.1 | The system SHALL handle libraries with 1,000,000+ files |
| NFR-3.2 | The system SHALL handle directories with 100,000+ entries |
| NFR-3.3 | The system SHALL maintain O(1) mount time regardless of library size |
| NFR-3.4 | The system SHALL maintain O(log n) lookup time for paths |
| NFR-3.5 | The system SHOULD handle libraries with 10,000,000+ files [P3] |
| NFR-3.6 | The system SHOULD support 100+ concurrent clients [P3] |
| NFR-3.7 | The system SHOULD achieve <100μs cached stat for high-performance use [P3] |
### 4.2 Resource Usage
#### NFR-4: Memory
| ID | Requirement | Limit |
|----|-------------|-------|
| NFR-4.1 | Idle memory usage | <50 MB |
| NFR-4.2 | Active usage (1000 files accessed) | <200 MB |
| NFR-4.3 | Peak usage under load | <500 MB |
| NFR-4.4 | Per-file metadata overhead | <1 KB |
| NFR-4.5 | The system SHALL NOT load entire files into memory |
#### NFR-5: Disk
| ID | Requirement |
|----|-------------|
| NFR-5.1 | Metadata cache size SHALL be configurable (default: 100 MB) |
| NFR-5.2 | Content cache size SHALL be configurable (default: 10 GB) |
| NFR-5.3 | The system SHALL evict cache entries under disk pressure |
| NFR-5.4 | The system SHALL function with cache disabled (passthrough mode) |
#### NFR-6: Network
| ID | Requirement |
|----|-------------|
| NFR-6.1 | The system SHALL minimize network round-trips via batching |
| NFR-6.2 | The system SHALL use connection pooling for remote origins |
| NFR-6.3 | The system SHALL support bandwidth limiting (configurable) |
| NFR-6.4 | Delta sync SHALL achieve >90% bandwidth reduction vs full copy |
### 4.3 Reliability
#### NFR-7: Availability
| ID | Requirement |
|----|-------------|
| NFR-7.1 | The system SHALL serve cached data when origin is unavailable |
| NFR-7.2 | The system SHALL gracefully degrade with network failures |
| NFR-7.3 | The system SHALL retry failed operations with exponential backoff |
| NFR-7.4 | The system SHALL not crash on malformed audio files |
#### NFR-8: Data Integrity
| ID | Requirement |
|----|-------------|
| NFR-8.1 | The system SHALL verify chunk integrity via checksums |
| NFR-8.2 | The system SHALL use ACID transactions for cache database |
| NFR-8.3 | The system SHALL recover from interrupted synchronization |
| NFR-8.4 | The system SHALL detect and report cache corruption |
### 4.4 Usability
#### NFR-9: Configuration
| ID | Requirement |
|----|-------------|
| NFR-9.1 | The system SHALL support configuration via file (TOML/YAML) |
| NFR-9.2 | The system SHALL support configuration via command-line arguments |
| NFR-9.3 | The system SHALL support configuration via environment variables |
| NFR-9.4 | The system SHALL provide sensible defaults for all options |
#### NFR-10: Observability
| ID | Requirement |
|----|-------------|
| NFR-10.1 | The system SHALL log operations at configurable verbosity |
| NFR-10.2 | The system SHALL expose metrics (cache hit rate, latency, etc.) |
| NFR-10.3 | The system SHALL support health check endpoint/signal |
| NFR-10.4 | The system SHOULD support integration with Prometheus/StatsD |
### 4.5 Compatibility
#### NFR-11: Platform Support
| ID | Requirement |
|----|-------------|
| NFR-11.1 | The system SHALL run on Linux (kernel 4.x+) |
| NFR-11.2 | The system SHOULD run on macOS (via macFUSE) |
| NFR-11.3 | The system SHALL require FUSE kernel module |
| NFR-11.4 | The system SHALL run without root privileges (user-space FUSE) |
#### NFR-12: Application Compatibility
| ID | Requirement |
|----|-------------|
| NFR-12.1 | The system SHALL work with standard media players (mpv, VLC, etc.) |
| NFR-12.2 | The system SHALL work with media servers (Plex, Jellyfin) |
| NFR-12.3 | The system SHALL work with file managers (Nautilus, Dolphin) |
| NFR-12.4 | The system SHALL correctly report file sizes and timestamps |
### 4.6 Security
#### NFR-13: Access Control
| ID | Requirement |
|----|-------------|
| NFR-13.1 | The system SHALL respect origin file permissions |
| NFR-13.2 | The system SHALL run as unprivileged user |
| NFR-13.3 | The system SHALL support credential storage for remote origins |
| NFR-13.4 | The system SHALL NOT expose credentials in logs or process list |
### 4.7 Maintainability
#### NFR-14: Code Quality
| ID | Requirement |
|----|-------------|
| NFR-14.1 | The system SHALL be implemented in a memory-safe language |
| NFR-14.2 | The system SHALL have no global interpreter lock (no Python/Ruby) |
| NFR-14.3 | The system SHALL use async I/O for concurrent operations |
| NFR-14.4 | The system SHALL have modular architecture with pluggable components |
---
## 5. Constraints
### 5.1 Technical Constraints
| ID | Constraint |
|----|------------|
| C-1 | Must use FUSE for filesystem interface |
| C-2 | Must not require kernel module development |
| C-3 | Must work with existing audio file formats (no transcoding) |
| C-4 | Cache database must be portable (no external database server) |
### 5.2 Operational Constraints
| ID | Constraint |
|----|------------|
| C-5 | Client is read-only; no writes propagate to origin |
| C-6 | Must function offline with cached data |
| C-7 | Must not corrupt origin files under any circumstances |
---
## 6. Assumptions
| ID | Assumption |
|----|------------|
| A-1 | Origin storage is accessible via supported protocol |
| A-2 | Audio files contain valid metadata headers |
| A-3 | Sufficient local disk space for caching is available |
| A-4 | FUSE kernel module is installed and accessible |
| A-5 | Network connectivity is intermittent but generally available |
---
## 7. Dependencies
| ID | Dependency | Purpose |
|----|------------|---------|
| D-1 | FUSE library (fuser/libfuse) | Filesystem interface |
| D-2 | SQLite | Metadata and tree cache |
| D-3 | Audio parsing library (symphonia) | Metadata extraction |
| D-4 | Async runtime (tokio) | Concurrent I/O |
| D-5 | CDC library (fastcdc) | Content chunking |
| D-6 | Full-text search (tantivy) | Search index [P1] |
| D-7 | Image processing (image) | Album art thumbnails [P1] |
| D-8 | HTTP client (reqwest) | Remote origins, metadata APIs |
| D-9 | WASM runtime (wasmtime) | Plugin sandboxing [P0] |
| D-10 | Hash library (xxhash/blake3) | Content addressing [P0] |
---
## 8. Acceptance Criteria
### 8.1 Functional Acceptance
| ID | Criterion |
|----|-----------|
| AC-1 | Mount filesystem and browse directories via `ls` |
| AC-2 | Play audio file through mounted filesystem with media player |
| AC-3 | Seek within audio file without full download |
| AC-4 | Directory listing completes without network access (when cached) |
| AC-5 | Confirm write operations return EROFS |
| AC-6 | Detect and sync changes from origin within configured interval |
### 8.2 Performance Acceptance
| ID | Criterion |
|----|-----------|
| AC-7 | Mount completes in <500ms for library of any size |
| AC-8 | Cached stat() completes in <5ms (p99) |
| AC-9 | Memory stays under 500MB with 10,000 files accessed |
| AC-10 | Tag-only change syncs <10KB of data |
### 8.3 Reliability Acceptance
| ID | Criterion |
|----|-----------|
| AC-11 | Filesystem remains accessible when origin is offline |
| AC-12 | No data corruption after unclean unmount |
| AC-13 | Recovers automatically when origin comes back online |
### 8.4 Multi-Origin Acceptance [P0]
| ID | Criterion |
|----|-----------|
| AC-14 | Configure and mount multiple origins simultaneously |
| AC-15 | Browse unified tree showing content from all origins |
| AC-16 | Access same file from preferred origin when duplicated |
### 8.5 Search & Discovery Acceptance [P1]
| ID | Criterion |
|----|-----------|
| AC-17 | Search for tracks by partial artist/album/title match |
| AC-18 | Browse smart collection (e.g., "Jazz from 1960s") |
| AC-19 | View album art via virtual cover.jpg file |
### 8.6 Plugin Acceptance [P0]
| ID | Criterion |
|----|-----------|
| AC-20 | Load custom origin plugin at runtime |
| AC-21 | Control daemon via Unix socket (cache stats, refresh) |
| AC-22 | Receive webhook on file access event |
### 8.7 Deduplication Acceptance [P0]
| ID | Criterion |
|----|-----------|
| AC-23 | Identical chunks stored once regardless of file count |
| AC-24 | Deduplication stats visible via control API |
---
## 9. Appendix
### 9.1 Comparison with beetfs
| Requirement Area | beetfs | This Specification |
|------------------|--------|-------------------|
| Mount time | O(N), 5-120s | O(1), <500ms (NFR-1.7) |
| Memory per file | Full file size | <1KB (NFR-4.4) |
| Write to origin | Yes (DB updates) | No (FR-4.1, FR-4.2) |
| Delta sync | None | Required (FR-11) |
| Remote origins | None | Required (FR-12) |
| Offline access | No | Required (NFR-7.1) |
| Cache persistence | No | Required (FR-7.1) |
### 9.2 Path Template Variables
| Variable | Description | Example |
|----------|-------------|---------|
| `$artist` | Track artist | "Metallica" |
| `$album` | Album name | "72 Seasons" |
| `$title` | Track title | "Lux Æterna" |
| `$track` | Track number (zero-padded) | "03" |
| `$disc` | Disc number | "1" |
| `$year` | Release year | "2023" |
| `$genre` | Genre | "Metal" |
| `$format` | File extension | "flac" |
| `$format_upper` | File extension (uppercase) | "FLAC" |
### 9.3 Error Codes
| Operation | Error | Code |
|-----------|-------|------|
| Any write operation | Read-only filesystem | EROFS (30) |
| File not found | No such file | ENOENT (2) |
| Origin unavailable | I/O error | EIO (5) |
| Permission denied | Access denied | EACCES (13) |