ArchiveBox/archivebox/BACKGROUND_HOOKS_IMPLEMENTATION_PLAN.md

Background Hooks Implementation Plan

Overview

This plan implements support for long-running background hooks that run concurrently with other extractors, while maintaining proper result collection, cleanup, and state management.

Key Changes:

• Background hooks use .bg.js/.bg.py/.bg.sh suffix
• Hooks output JSONL (any line with {type: 'ModelName', ...})
• run_hook() is generic - just parses JSONL, doesn't know about specific models
• Each Model.run() extends records of its own type with computed fields
• ArchiveResult.run() extends ArchiveResult records with output_files, output_size, etc.
• No HookResult TypedDict - just list of dicts with 'type' field
• Binary FK is optional and only set when hook reports cmd
• Split output field into output_str (human-readable) and output_json (structured)
• Add fields: output_files (dict), output_size (bytes), output_mimetypes (CSV)
• External tools (fdupes, ZFS, Btrfs) handle deduplication via filesystem

New ArchiveResult Fields:

# Output fields (replace old 'output' field)
output_str = TextField()           # Human-readable summary: "Downloaded 5 files"
output_json = JSONField()          # Structured metadata (headers, redirects, etc.)
output_files = JSONField()         # Dict: {'index.html': {}, 'style.css': {}}
output_size = BigIntegerField()    # Total bytes across all files
output_mimetypes = CharField()     # CSV sorted by size: "text/html,text/css,image/png"

output_files Structure:

• Dict keyed by relative path (not a list!)
• Values are empty dicts {} for now, extensible for future metadata
• Preserves insertion order (Python 3.7+)
• Easy to query: ArchiveResult.objects.filter(output_files__has_key='index.html')
• Easy to extend: Add size, hash, mime_type to values later without migration
• Why not derive size/mimetypes from output_files? Performance. Total size and mimetype summary are accessed frequently (admin views, sorting, filtering). Aggregating on every access would be slow. We keep summary fields (output_size, output_mimetypes) as denormalized cache for fast reads.

---

Phase 1: Database Migration

Add new fields to ArchiveResult

# archivebox/core/migrations/00XX_archiveresult_background_hooks.py

from django.db import migrations, models

class Migration(migrations.Migration):
    dependencies = [
        ('core', 'XXXX_previous_migration'),
        ('machine', 'XXXX_latest_machine_migration'),
    ]

    operations = [
        # Add new fields (keep old 'output' temporarily for migration)
        migrations.AddField(
            model_name='archiveresult',
            name='output_str',
            field=models.TextField(
                blank=True,
                help_text='Human-readable output summary (e.g., "Downloaded 5 files")'
            ),
        ),

        migrations.AddField(
            model_name='archiveresult',
            name='output_json',
            field=models.JSONField(
                null=True,
                blank=True,
                help_text='Structured metadata (headers, redirects, etc.) - should NOT duplicate ArchiveResult fields'
            ),
        ),

        migrations.AddField(
            model_name='archiveresult',
            name='output_files',
            field=models.JSONField(
                default=dict,
                help_text='Dict of {relative_path: {metadata}} - values are empty dicts for now, extensible for future metadata'
            ),
        ),

        migrations.AddField(
            model_name='archiveresult',
            name='output_size',
            field=models.BigIntegerField(
                default=0,
                help_text='Total recursive size in bytes of all output files'
            ),
        ),

        migrations.AddField(
            model_name='archiveresult',
            name='output_mimetypes',
            field=models.CharField(
                max_length=512,
                blank=True,
                help_text='CSV of mimetypes sorted by size descending'
            ),
        ),

        # Add binary FK (optional)
        migrations.AddField(
            model_name='archiveresult',
            name='binary',
            field=models.ForeignKey(
                'machine.InstalledBinary',
                on_delete=models.SET_NULL,
                null=True,
                blank=True,
                help_text='Primary binary used by this hook (optional)'
            ),
        ),
    ]

Data Migration for Existing .output Field

# archivebox/core/migrations/00XX_migrate_output_field.py

from django.db import migrations
import json

def migrate_output_field(apps, schema_editor):
    """
    Migrate existing 'output' field to new split fields.

    Logic:
    - If output contains JSON {...}, move to output_json
    - If output is a file path and exists in output_files, ensure it's first
    - Otherwise, move to output_str
    """
    ArchiveResult = apps.get_model('core', 'ArchiveResult')

    for ar in ArchiveResult.objects.all():
        old_output = ar.output or ''

        # Case 1: JSON output
        if old_output.strip().startswith('{'):
            try:
                parsed = json.loads(old_output)
                ar.output_json = parsed
                ar.output_str = ''
            except json.JSONDecodeError:
                # Not valid JSON, treat as string
                ar.output_str = old_output

        # Case 2: File path (check if it looks like a relative path)
        elif '/' in old_output or '.' in old_output:
            # Might be a file path - if it's in output_files, it's already there
            # output_files is now a dict, so no reordering needed
            ar.output_str = old_output  # Keep as string for display

        # Case 3: Plain string summary
        else:
            ar.output_str = old_output

        ar.save(update_fields=['output_str', 'output_json', 'output_files'])

def reverse_migrate(apps, schema_editor):
    """Reverse migration - copy output_str back to output."""
    ArchiveResult = apps.get_model('core', 'ArchiveResult')

    for ar in ArchiveResult.objects.all():
        ar.output = ar.output_str or ''
        ar.save(update_fields=['output'])

class Migration(migrations.Migration):
    dependencies = [
        ('core', '00XX_archiveresult_background_hooks'),
    ]

    operations = [
        migrations.RunPython(migrate_output_field, reverse_migrate),

        # Now safe to remove old 'output' field
        migrations.RemoveField(
            model_name='archiveresult',
            name='output',
        ),
    ]

---

Phase 2: Hook Output Format

Hooks emit single JSON object to stdout

Contract:

• Hook emits ONE JSON object with type: 'ArchiveResult'
• Hook can provide: status, output_str, output_json, cmd (optional)
• Hook should NOT set: output_files, output_size, output_mimetypes (runner calculates these)
• output_json should NOT duplicate ArchiveResult fields (no status, start_ts, etc. in output_json)
• Runner calculates: output_files, output_size, output_mimetypes, start_ts, end_ts, binary FK

Example outputs:

// Simple string output
console.log(JSON.stringify({
    type: 'ArchiveResult',
    status: 'succeeded',
    output_str: 'Downloaded index.html (4.2 KB)'
}));

// With structured metadata (headers, redirects, etc.)
console.log(JSON.stringify({
    type: 'ArchiveResult',
    status: 'succeeded',
    output_str: 'Archived https://example.com',
    output_json: {
        headers: {'content-type': 'text/html', 'server': 'nginx'},
        redirects: [{from: 'http://example.com', to: 'https://example.com'}]
    }
}));

// With explicit cmd (for binary FK)
console.log(JSON.stringify({
    type: 'ArchiveResult',
    status: 'succeeded',
    output_str: 'Archived with wget',
    cmd: ['wget', '-p', '-k', 'https://example.com']
}));

// Just structured data (no human-readable string)
console.log(JSON.stringify({
    type: 'ArchiveResult',
    status: 'succeeded',
    output_json: {
        title: 'My Page Title',
        charset: 'UTF-8'
    }
}));

// BAD: Don't duplicate ArchiveResult fields in output_json
console.log(JSON.stringify({
    type: 'ArchiveResult',
    status: 'succeeded',
    output_json: {
        status: 'succeeded',  // ❌ BAD - duplicates ArchiveResult.status
        output_files: ['index.html'],  // ❌ BAD - runner calculates this
        custom_data: 'ok'  // ✅ GOOD - custom fields only
    }
}));

---

Phase 3: run_hook() is Generic (No HookResult TypedDict)

run_hook() is a generic JSONL parser - it doesn't know about ArchiveResult, InstalledBinary, or any specific model. It just:

1. Executes the hook script
2. Parses JSONL output (any line starting with { that has a type field)
3. Adds metadata about plugin and hook path
4. Returns list of dicts

# archivebox/hooks.py

def run_hook(
    script: Path,
    output_dir: Path,
    timeout: int = 300,
    config_objects: Optional[List[Any]] = None,
    **kwargs: Any
) -> Optional[List[dict]]:
    """
    Execute a hook script and parse JSONL output.

    This function is generic and doesn't know about specific model types.
    It just executes the script and parses any JSONL lines with 'type' field.

    Each Model.run() method handles its own record types differently:
    - ArchiveResult.run() extends ArchiveResult records with computed fields
    - Machine.run() creates InstalledBinary records from hook output
    - etc.

    Returns:
        List of dicts with 'type' field, each extended with metadata:
        [
            {
                'type': 'ArchiveResult',
                'status': 'succeeded',
                'output_str': '...',
                'plugin': 'wget',
                'plugin_hook': 'archivebox/plugins/wget/on_Snapshot__21_wget.py',
                # ... other hook-reported fields
            },
            {
                'type': 'InstalledBinary',
                'name': 'wget',
                'plugin': 'wget',
                'plugin_hook': 'archivebox/plugins/wget/on_Snapshot__21_wget.py',
                # ... other hook-reported fields
            }
        ]

        None if background hook (still running)
    """

Key Insight: Hooks output JSONL. Any line with {type: 'ModelName', ...} creates/updates that model. The type field determines what gets created. Each Model.run() method decides how to handle records of its own type.

Helper: create_model_record()

# archivebox/hooks.py

def create_model_record(record: dict) -> Any:
    """
    Generic helper to create/update model instances from hook output.

    Args:
        record: Dict with 'type' field and model data

    Returns:
        Created/updated model instance
    """
    from machine.models import InstalledBinary, Dependency

    model_type = record.pop('type')

    if model_type == 'InstalledBinary':
        obj, created = InstalledBinary.objects.get_or_create(**record)
        return obj
    elif model_type == 'Dependency':
        obj, created = Dependency.objects.get_or_create(**record)
        return obj
    # Add more types as needed
    else:
        raise ValueError(f"Unknown record type: {model_type}")

---

Phase 4: Update run_hook() Implementation

Location: archivebox/hooks.py

def find_binary_for_cmd(cmd: List[str], machine_id: str) -> Optional[str]:
    """
    Find InstalledBinary for a command, trying abspath first then name.
    Only matches binaries on the current machine.

    Args:
        cmd: Command list (e.g., ['/usr/bin/wget', '-p', 'url'])
        machine_id: Current machine ID

    Returns:
        Binary ID if found, None otherwise
    """
    if not cmd:
        return None

    from machine.models import InstalledBinary

    bin_path_or_name = cmd[0]

    # Try matching by absolute path first
    binary = InstalledBinary.objects.filter(
        abspath=bin_path_or_name,
        machine_id=machine_id
    ).first()

    if binary:
        return str(binary.id)

    # Fallback: match by binary name
    bin_name = Path(bin_path_or_name).name
    binary = InstalledBinary.objects.filter(
        name=bin_name,
        machine_id=machine_id
    ).first()

    return str(binary.id) if binary else None


def run_hook(
    script: Path,
    output_dir: Path,
    timeout: int = 300,
    config_objects: Optional[List[Any]] = None,
    **kwargs: Any
) -> Optional[List[dict]]:
    """
    Execute a hook script and parse JSONL output.

    This is a GENERIC function that doesn't know about specific model types.
    It just executes and parses JSONL (any line with {type: 'ModelName', ...}).

    Runner responsibilities:
    - Detect background hooks (.bg. in filename)
    - Capture stdout/stderr to log files
    - Parse JSONL output and add plugin metadata
    - Clean up log files and PID files

    Hook responsibilities:
    - Emit JSONL: {type: 'ArchiveResult', status, output_str, output_json, cmd}
    - Can emit multiple types: {type: 'InstalledBinary', ...}
    - Write actual output files

    Args:
        script: Path to hook script
        output_dir: Working directory (where output files go)
        timeout: Max execution time in seconds
        config_objects: Config override objects (Machine, Crawl, Snapshot)
        **kwargs: CLI arguments passed to script

    Returns:
        List of dicts with 'type' field for foreground hooks
        None for background hooks (still running)
    """
    import time
    from datetime import datetime, timezone
    from machine.models import Machine

    start_time = time.time()

    # 1. SETUP
    is_background = '.bg.' in script.name  # Detect .bg.js/.bg.py/.bg.sh
    effective_timeout = timeout * 10 if is_background else timeout

    # Infrastructure files (ALL hooks)
    stdout_file = output_dir / 'stdout.log'
    stderr_file = output_dir / 'stderr.log'
    pid_file = output_dir / 'hook.pid'

    # Capture files before execution
    files_before = set(output_dir.rglob('*')) if output_dir.exists() else set()
    start_ts = datetime.now(timezone.utc)

    # 2. BUILD COMMAND
    ext = script.suffix.lower()
    if ext == '.sh':
        interpreter_cmd = ['bash', str(script)]
    elif ext == '.py':
        interpreter_cmd = ['python3', str(script)]
    elif ext == '.js':
        interpreter_cmd = ['node', str(script)]
    else:
        interpreter_cmd = [str(script)]

    # Build CLI arguments from kwargs
    cli_args = []
    for key, value in kwargs.items():
        if key.startswith('_'):
            continue

        arg_key = f'--{key.replace("_", "-")}'
        if isinstance(value, bool):
            if value:
                cli_args.append(arg_key)
        elif value is not None and value != '':
            if isinstance(value, (dict, list)):
                cli_args.append(f'{arg_key}={json.dumps(value)}')
            else:
                str_value = str(value).strip()
                if str_value:
                    cli_args.append(f'{arg_key}={str_value}')

    full_cmd = interpreter_cmd + cli_args

    # 3. SET UP ENVIRONMENT
    env = os.environ.copy()
    # ... (existing env setup from current run_hook implementation)

    # 4. CREATE OUTPUT DIRECTORY
    output_dir.mkdir(parents=True, exist_ok=True)

    # 5. EXECUTE PROCESS
    try:
        with open(stdout_file, 'w') as out, open(stderr_file, 'w') as err:
            process = subprocess.Popen(
                full_cmd,
                cwd=str(output_dir),
                stdout=out,
                stderr=err,
                env=env,
            )

            # Write PID for all hooks
            pid_file.write_text(str(process.pid))

            if is_background:
                # Background hook - return immediately, don't wait
                return None

            # Foreground hook - wait for completion
            try:
                returncode = process.wait(timeout=effective_timeout)
            except subprocess.TimeoutExpired:
                process.kill()
                process.wait()
                returncode = -1
                with open(stderr_file, 'a') as err:
                    err.write(f'\nHook timed out after {effective_timeout}s')

        # 6. COLLECT RESULTS (foreground only)
        end_ts = datetime.now(timezone.utc)

        stdout = stdout_file.read_text() if stdout_file.exists() else ''
        stderr = stderr_file.read_text() if stderr_file.exists() else ''

        # Parse ALL JSONL output (any line with {type: 'ModelName', ...})
        records = []
        for line in stdout.splitlines():
            line = line.strip()
            if not line or not line.startswith('{'):
                continue
            try:
                data = json.loads(line)
                if 'type' in data:
                    # Add plugin metadata to every record
                    plugin_name = script.parent.name  # Directory name (e.g., 'wget')
                    data['plugin'] = plugin_name
                    data['plugin_hook'] = str(script.relative_to(Path.cwd()))
                    records.append(data)
            except json.JSONDecodeError:
                continue

        # 7. CLEANUP
        # Delete empty logs (keep non-empty for debugging)
        if stdout_file.exists() and stdout_file.stat().st_size == 0:
            stdout_file.unlink()
        if stderr_file.exists() and stderr_file.stat().st_size == 0:
            stderr_file.unlink()

        # Delete ALL .pid files on success
        if returncode == 0:
            for pf in output_dir.glob('*.pid'):
                pf.unlink(missing_ok=True)

        # 8. RETURN RECORDS
        # Returns list of dicts, each with 'type' field and plugin metadata
        return records

    except Exception as e:
        # On error, return empty list (hook failed, no records created)
        return []

---

Phase 5: Update ArchiveResult.run()

Location: archivebox/core/models.py

def run(self):
    """
    Execute this ArchiveResult's extractor and update status.

    For foreground hooks: Waits for completion and updates immediately
    For background hooks: Returns immediately, leaves status='started'

    This method extends any ArchiveResult records from hook output with
    computed fields (output_files, output_size, binary FK, etc.).
    """
    from django.utils import timezone
    from archivebox.hooks import BUILTIN_PLUGINS_DIR, USER_PLUGINS_DIR, run_hook
    from machine.models import Machine

    config_objects = [self.snapshot.crawl, self.snapshot] if self.snapshot.crawl else [self.snapshot]

    # Find hook for this extractor
    hook = None
    for base_dir in (BUILTIN_PLUGINS_DIR, USER_PLUGINS_DIR):
        if not base_dir.exists():
            continue
        matches = list(base_dir.glob(f'*/on_Snapshot__{self.extractor}.*'))
        if matches:
            hook = matches[0]
            break

    if not hook:
        self.status = self.StatusChoices.FAILED
        self.output_str = f'No hook found for: {self.extractor}'
        self.retry_at = None
        self.save()
        return

    # Use plugin directory name instead of extractor name
    plugin_name = hook.parent.name
    extractor_dir = Path(self.snapshot.output_dir) / plugin_name

    start_ts = timezone.now()

    # Run the hook (returns list of JSONL records)
    records = run_hook(
        hook,
        output_dir=extractor_dir,
        config_objects=config_objects,
        url=self.snapshot.url,
        snapshot_id=str(self.snapshot.id),
    )

    # BACKGROUND HOOK - still running
    if records is None:
        self.status = self.StatusChoices.STARTED
        self.start_ts = start_ts
        self.pwd = str(extractor_dir)
        self.save()
        return

    # FOREGROUND HOOK - process records
    end_ts = timezone.now()

    # Find the ArchiveResult record (enforce single output)
    ar_records = [r for r in records if r.get('type') == 'ArchiveResult']
    assert len(ar_records) <= 1, f"Hook {hook} output {len(ar_records)} ArchiveResults, expected 0-1"

    if ar_records:
        hook_data = ar_records[0]

        # Apply hook's data
        status_str = hook_data.get('status', 'failed')
        status_map = {
            'succeeded': self.StatusChoices.SUCCEEDED,
            'failed': self.StatusChoices.FAILED,
            'skipped': self.StatusChoices.SKIPPED,
        }
        self.status = status_map.get(status_str, self.StatusChoices.FAILED)

        self.output_str = hook_data.get('output_str', '')
        self.output_json = hook_data.get('output_json')

        # Set extractor from plugin metadata
        self.extractor = hook_data['plugin']

        # Determine binary FK from cmd (ArchiveResult-specific logic)
        if 'cmd' in hook_data:
            self.cmd = json.dumps(hook_data['cmd'])
            machine = Machine.current()
            binary_id = find_binary_for_cmd(hook_data['cmd'], machine.id)
            if binary_id:
                self.binary_id = binary_id
    else:
        # No ArchiveResult output - hook didn't report, treat as failed
        self.status = self.StatusChoices.FAILED
        self.output_str = 'Hook did not output ArchiveResult'

    # Set timestamps and metadata
    self.start_ts = start_ts
    self.end_ts = end_ts
    self.pwd = str(extractor_dir)
    self.retry_at = None

    # POPULATE OUTPUT FIELDS FROM FILESYSTEM (ArchiveResult-specific)
    if extractor_dir.exists():
        self._populate_output_fields(extractor_dir)

    self.save()

    # Create any side-effect records (InstalledBinary, Dependency, etc.)
    for record in records:
        if record['type'] != 'ArchiveResult':
            create_model_record(record)  # Generic helper that dispatches by type

    # Clean up empty output directory (no real files after excluding logs/pids)
    if extractor_dir.exists():
        try:
            # Check if only infrastructure files remain
            remaining_files = [
                f for f in extractor_dir.rglob('*')
                if f.is_file() and f.name not in ('stdout.log', 'stderr.log', 'hook.pid', 'listener.pid')
            ]
            if not remaining_files:
                # Remove infrastructure files
                for pf in extractor_dir.glob('*.log'):
                    pf.unlink(missing_ok=True)
                for pf in extractor_dir.glob('*.pid'):
                    pf.unlink(missing_ok=True)
                # Try to remove directory if empty
                if not any(extractor_dir.iterdir()):
                    extractor_dir.rmdir()
        except (OSError, RuntimeError):
            pass

    # Queue discovered URLs, trigger indexing, etc.
    self._queue_urls_for_crawl(extractor_dir)

    if self.status == self.StatusChoices.SUCCEEDED:
        # Update snapshot title if this is title extractor
        extractor_name = get_extractor_name(self.extractor)
        if extractor_name == 'title':
            self._update_snapshot_title(extractor_dir)

        # Trigger search indexing
        self.trigger_search_indexing()


def _populate_output_fields(self, output_dir: Path) -> None:
    """
    Walk output directory and populate output_files, output_size, output_mimetypes fields.

    Args:
        output_dir: Directory containing output files
    """
    import mimetypes
    from collections import defaultdict

    exclude_names = {'stdout.log', 'stderr.log', 'hook.pid', 'listener.pid'}

    # Track mimetypes and sizes for aggregation
    mime_sizes = defaultdict(int)
    total_size = 0
    output_files = {}  # Dict keyed by relative path

    for file_path in output_dir.rglob('*'):
        # Skip non-files and infrastructure files
        if not file_path.is_file():
            continue
        if file_path.name in exclude_names:
            continue

        # Get file stats
        stat = file_path.stat()
        mime_type, _ = mimetypes.guess_type(str(file_path))
        mime_type = mime_type or 'application/octet-stream'

        # Track for ArchiveResult fields
        relative_path = str(file_path.relative_to(output_dir))
        output_files[relative_path] = {}  # Empty dict, extensible for future metadata
        mime_sizes[mime_type] += stat.st_size
        total_size += stat.st_size

    # Populate ArchiveResult fields
    self.output_files = output_files  # Dict preserves insertion order (Python 3.7+)
    self.output_size = total_size

    # Build output_mimetypes CSV (sorted by size descending)
    sorted_mimes = sorted(mime_sizes.items(), key=lambda x: x[1], reverse=True)
    self.output_mimetypes = ','.join(mime for mime, _ in sorted_mimes)

Querying output_files with Django

Since output_files is a dict keyed by relative path, you can use Django's JSON field lookups:

# Check if a specific file exists
ArchiveResult.objects.filter(output_files__has_key='index.html')

# Check if any of multiple files exist (OR)
from django.db.models import Q
ArchiveResult.objects.filter(
    Q(output_files__has_key='index.html') |
    Q(output_files__has_key='index.htm')
)

# Get all results that have favicon
ArchiveResult.objects.filter(output_files__has_key='favicon.ico')

# Check in Python (after fetching)
if 'index.html' in archiveresult.output_files:
    print("Found index.html")

# Get list of all paths
paths = list(archiveresult.output_files.keys())

# Count files
file_count = len(archiveresult.output_files)

# Future: When we add metadata, query still works
# output_files = {'index.html': {'size': 4096, 'hash': 'abc...'}}
ArchiveResult.objects.filter(output_files__index_html__size__gt=1000)  # size > 1KB

Structure for Future Extension:

Current (empty metadata):

{
    'index.html': {},
    'style.css': {},
    'images/logo.png': {}
}

Future (with optional metadata):

{
    'index.html': {
        'size': 4096,
        'hash': 'abc123...',
        'mime_type': 'text/html'
    },
    'style.css': {
        'size': 2048,
        'hash': 'def456...',
        'mime_type': 'text/css'
    }
}

All existing queries continue to work unchanged - the dict structure is backward compatible.

---

Phase 6: Background Hook Finalization

Helper Functions

Location: archivebox/core/models.py or new archivebox/core/background_hooks.py

def find_background_hooks(snapshot) -> List['ArchiveResult']:
    """
    Find all ArchiveResults that are background hooks still running.

    Args:
        snapshot: Snapshot instance

    Returns:
        List of ArchiveResults with status='started'
    """
    return list(snapshot.archiveresult_set.filter(
        status=ArchiveResult.StatusChoices.STARTED
    ))


def check_background_hook_completed(archiveresult: 'ArchiveResult') -> bool:
    """
    Check if background hook process has exited.

    Args:
        archiveresult: ArchiveResult instance

    Returns:
        True if completed (process exited), False if still running
    """
    extractor_dir = Path(archiveresult.pwd)
    pid_file = extractor_dir / 'hook.pid'

    if not pid_file.exists():
        return True  # No PID file = completed or failed to start

    try:
        pid = int(pid_file.read_text().strip())
        os.kill(pid, 0)  # Signal 0 = check if process exists
        return False  # Still running
    except (OSError, ValueError):
        return True  # Process exited or invalid PID


def finalize_background_hook(archiveresult: 'ArchiveResult') -> None:
    """
    Collect final results from completed background hook.

    Same logic as ArchiveResult.run() but for background hooks that already started.

    Args:
        archiveresult: ArchiveResult instance to finalize
    """
    from django.utils import timezone
    from machine.models import Machine

    extractor_dir = Path(archiveresult.pwd)
    stdout_file = extractor_dir / 'stdout.log'
    stderr_file = extractor_dir / 'stderr.log'

    # Read logs
    stdout = stdout_file.read_text() if stdout_file.exists() else ''

    # Parse JSONL output (same as run_hook)
    records = []
    for line in stdout.splitlines():
        line = line.strip()
        if not line or not line.startswith('{'):
            continue
        try:
            data = json.loads(line)
            if 'type' in data:
                records.append(data)
        except json.JSONDecodeError:
            continue

    # Find the ArchiveResult record
    ar_records = [r for r in records if r.get('type') == 'ArchiveResult']
    assert len(ar_records) <= 1, f"Background hook output {len(ar_records)} ArchiveResults, expected 0-1"

    if ar_records:
        hook_data = ar_records[0]

        # Apply hook's data
        status_str = hook_data.get('status', 'failed')
        status_map = {
            'succeeded': ArchiveResult.StatusChoices.SUCCEEDED,
            'failed': ArchiveResult.StatusChoices.FAILED,
            'skipped': ArchiveResult.StatusChoices.SKIPPED,
        }
        archiveresult.status = status_map.get(status_str, ArchiveResult.StatusChoices.FAILED)

        archiveresult.output_str = hook_data.get('output_str', '')
        archiveresult.output_json = hook_data.get('output_json')

        # Determine binary FK from cmd
        if 'cmd' in hook_data:
            archiveresult.cmd = json.dumps(hook_data['cmd'])
            machine = Machine.current()
            binary_id = find_binary_for_cmd(hook_data['cmd'], machine.id)
            if binary_id:
                archiveresult.binary_id = binary_id
    else:
        # No output = failed
        archiveresult.status = ArchiveResult.StatusChoices.FAILED
        archiveresult.output_str = 'Background hook did not output ArchiveResult'

    archiveresult.end_ts = timezone.now()
    archiveresult.retry_at = None

    # POPULATE OUTPUT FIELDS FROM FILESYSTEM
    if extractor_dir.exists():
        archiveresult._populate_output_fields(extractor_dir)

    archiveresult.save()

    # Create any side-effect records
    for record in records:
        if record['type'] != 'ArchiveResult':
            create_model_record(record)

    # Cleanup
    for pf in extractor_dir.glob('*.pid'):
        pf.unlink(missing_ok=True)
    if stdout_file.exists() and stdout_file.stat().st_size == 0:
        stdout_file.unlink()
    if stderr_file.exists() and stderr_file.stat().st_size == 0:
        stderr_file.unlink()

Update SnapshotMachine

Location: archivebox/core/statemachines.py

class SnapshotMachine(StateMachine, strict_states=True):
    # ... existing states ...

    def is_finished(self) -> bool:
        """
        Check if snapshot archiving is complete.

        A snapshot is finished when:
        1. No pending archiveresults remain (queued/started foreground hooks)
        2. All background hooks have completed
        """
        # Check if any pending archiveresults exist
        if self.snapshot.pending_archiveresults().exists():
            return False

        # Check and finalize background hooks
        background_hooks = find_background_hooks(self.snapshot)
        for bg_hook in background_hooks:
            if not check_background_hook_completed(bg_hook):
                return False  # Still running

            # Completed - finalize it
            finalize_background_hook(bg_hook)

        # All done
        return True

---

Phase 6b: Deduplication

Deduplication is handled by external filesystem tools like fdupes (hardlinks), ZFS dedup, Btrfs duperemove, or rdfind. Users can run these tools periodically on the archive directory to identify and link duplicate files. ArchiveBox doesn't need to track hashes or manage deduplication itself - the filesystem layer handles it transparently.

---

Phase 7: Rename Background Hooks

Files to rename:

# Use .bg. suffix (not __background)
mv archivebox/plugins/consolelog/on_Snapshot__21_consolelog.js \
   archivebox/plugins/consolelog/on_Snapshot__21_consolelog.bg.js

mv archivebox/plugins/ssl/on_Snapshot__23_ssl.js \
   archivebox/plugins/ssl/on_Snapshot__23_ssl.bg.js

mv archivebox/plugins/responses/on_Snapshot__24_responses.js \
   archivebox/plugins/responses/on_Snapshot__24_responses.bg.js

Update hook content to emit proper JSON:

Each hook should emit:

console.log(JSON.stringify({
    type: 'ArchiveResult',
    status: 'succeeded',  // or 'failed' or 'skipped'
    output_str: 'Captured 15 console messages',  // human-readable summary
    output_json: {  // optional structured metadata
        // ... specific to each hook
    }
}));

---

Phase 8: Update Existing Hooks

Update all hooks to emit proper JSON format

Example: favicon hook

# Before
print(f'Favicon saved ({size} bytes)')
print(f'OUTPUT={OUTPUT_FILE}')
print(f'STATUS=succeeded')

# After
result = {
    'type': 'ArchiveResult',
    'status': 'succeeded',
    'output_str': f'Favicon saved ({size} bytes)',
    'output_json': {
        'size': size,
        'format': 'ico'
    }
}
print(json.dumps(result))

Example: wget hook with explicit cmd

# After wget completes
cat <<EOF
{"type": "ArchiveResult", "status": "succeeded", "output_str": "Downloaded index.html", "cmd": ["wget", "-p", "-k", "$URL"]}
EOF

---

Testing Strategy

1. Unit Tests

# tests/test_background_hooks.py

def test_background_hook_detection():
    """Test .bg. suffix detection"""
    assert is_background_hook(Path('on_Snapshot__21_test.bg.js'))
    assert not is_background_hook(Path('on_Snapshot__21_test.js'))

def test_find_binary_by_abspath():
    """Test binary matching by absolute path"""
    machine = Machine.current()
    binary = InstalledBinary.objects.create(
        name='wget',
        abspath='/usr/bin/wget',
        machine=machine
    )

    cmd = ['/usr/bin/wget', '-p', 'url']
    assert find_binary_for_cmd(cmd, machine.id) == str(binary.id)

def test_find_binary_by_name():
    """Test binary matching by name fallback"""
    machine = Machine.current()
    binary = InstalledBinary.objects.create(
        name='wget',
        abspath='/usr/local/bin/wget',
        machine=machine
    )

    cmd = ['wget', '-p', 'url']
    assert find_binary_for_cmd(cmd, machine.id) == str(binary.id)

def test_parse_hook_json():
    """Test JSON parsing from stdout"""
    stdout = '''
    Some log output
    {"type": "ArchiveResult", "status": "succeeded", "output_str": "test"}
    More output
    '''
    result = parse_hook_output_json(stdout)
    assert result['status'] == 'succeeded'
    assert result['output_str'] == 'test'

2. Integration Tests

def test_foreground_hook_execution(snapshot):
    """Test foreground hook runs and returns results"""
    ar = ArchiveResult.objects.create(
        snapshot=snapshot,
        extractor='11_favicon',
        status=ArchiveResult.StatusChoices.QUEUED
    )

    ar.run()
    ar.refresh_from_db()

    assert ar.status in [
        ArchiveResult.StatusChoices.SUCCEEDED,
        ArchiveResult.StatusChoices.FAILED
    ]
    assert ar.start_ts is not None
    assert ar.end_ts is not None
    assert ar.output_size >= 0

def test_background_hook_execution(snapshot):
    """Test background hook starts but doesn't block"""
    ar = ArchiveResult.objects.create(
        snapshot=snapshot,
        extractor='21_consolelog',
        status=ArchiveResult.StatusChoices.QUEUED
    )

    start = time.time()
    ar.run()
    duration = time.time() - start

    ar.refresh_from_db()

    # Should return quickly (< 5 seconds)
    assert duration < 5
    # Should be in 'started' state
    assert ar.status == ArchiveResult.StatusChoices.STARTED
    # PID file should exist
    assert (Path(ar.pwd) / 'hook.pid').exists()

def test_background_hook_finalization(snapshot):
    """Test background hook finalization after completion"""
    # Start background hook
    ar = ArchiveResult.objects.create(
        snapshot=snapshot,
        extractor='21_consolelog',
        status=ArchiveResult.StatusChoices.STARTED,
        pwd='/path/to/output'
    )

    # Simulate completion (hook writes output and exits)
    # ...

    # Finalize
    finalize_background_hook(ar)
    ar.refresh_from_db()

    assert ar.status == ArchiveResult.StatusChoices.SUCCEEDED
    assert ar.end_ts is not None
    assert ar.output_size > 0

---

Migration Path

Step 1: Create migration

cd archivebox
python manage.py makemigrations core --name archiveresult_background_hooks

Step 2: Update run_hook()

• Add background hook detection
• Add log file capture
• Parse JSONL output (any line with {type: 'ModelName', ...})
• Add plugin and plugin_hook metadata to each record

Step 3: Update ArchiveResult.run()

• Handle None result for background hooks (return immediately)
• Parse records list from run_hook()
• Assert only one ArchiveResult record per hook
• Extend ArchiveResult record with computed fields (output_files, output_size, binary FK)
• Call _populate_output_fields() to walk directory and populate summary fields
• Call create_model_record() for any side-effect records (InstalledBinary, etc.)

Step 4: Add finalization helpers

• find_background_hooks()
• check_background_hook_completed()
• finalize_background_hook()

Step 5: Update SnapshotMachine.is_finished()

• Check for background hooks
• Finalize completed ones

Step 6: Rename hooks

• Rename 3 background hooks with .bg. suffix

Step 7: Update hook outputs

• Update all hooks to emit JSON format
• Remove manual timestamp/status calculation

Step 8: Test

• Unit tests
• Integration tests
• Manual testing with real snapshots

---

Success Criteria

• ✅ Background hooks start immediately without blocking other extractors
• ✅ Background hooks are finalized after completion with full results
• ✅ All output stats calculated by runner, not hooks
• ✅ Binary FK optional and only set when determinable
• ✅ Clean separation between output_str (human) and output_json (structured)
• ✅ output_files stored as dict for easy querying and future extensibility
• ✅ Log files cleaned up on success, kept on failure
• ✅ PID files cleaned up after completion
• ✅ No plugin-specific code in core (generic polling mechanism)

---

Future Enhancements

1. Timeout for orphaned background hooks

If a background hook runs longer than MAX_LIFETIME after all foreground hooks complete, force kill it.

2. Progress reporting

Background hooks could write progress to a file that gets polled:

fs.writeFileSync('progress.txt', '50%');

3. Multiple results per hook

If needed in future, extend to support multiple JSON outputs by collecting all {type: 'ArchiveResult'} lines.

4. Dependency tracking

Store all binaries used by a hook (not just primary), useful for hooks that chain multiple tools.

5. Per-file metadata in output_files

If needed, extend output_files values to include per-file metadata:

output_files = {
    'index.html': {
        'size': 4096,
        'hash': 'abc123...',
        'mime_type': 'text/html',
        'modified_at': '2025-01-15T10:30:00Z'
    }
}

Can query with custom SQL for complex per-file queries (e.g., "find all results with any file > 50KB"). Summary fields (output_size, output_mimetypes) remain as denormalized cache for performance.