feat: catalog shared contracts (Tasks 1-4)

Task 1: Establish catalog identity and atomic file ownership
- CatalogService, CatalogServiceInterface, ModelMetadata DTO
- PdoCatalogRepository with atomic operations, savepoint support
- FOR UPDATE file locks in ascending order, FOR NO KEY UPDATE for metadata
- FK-compatible lock composition documented (spec §6)

Task 2: M3 import & indexing engine delivery layer
- ImportService, ImportWorker with durable collision handling
- SpaceChecker, StreamingHasher, VerifiedCopier, StlValidator
- FileEnumerator, LocalFilesystem, PathSafety
- Console command wiring via ContainerWiringTest

Task 3: Importer integration and repair
- LegacyGuessedImportGroupingAdapter isolates first-directory guessing
- ImportService uses adapter; ImportWorker uses CatalogServiceInterface
- Null owners remain null; DomainException mapped to CATALOG_CONFLICT
- SQL failure after registration/path update with caller sentinel rollback

Task 4: PostgreSQL contention evidence and bounded acceptance
- 6 new PostgresCatalogContractsTest cases with independent sessions
- Generated schemas, 250ms lock timeouts, 5s statement timeouts
- Ascending file locks, FOR SHARE contention, stale-owner rejection
- Rollback/retry, nested transaction survival, root creation contention
- Negative controls: replay FOR UPDATE removal (11 failures), metadata FOR UPDATE (4 FK KEY SHARE timeouts)
- Fresh Docker PostgreSQL 17 / PHP 8.4.26: 473 tests, 2718 assertions, 10 skips, exit 0

All local: 474 tests, 2347 assertions, 54 skips, foundation/lint clean.
This commit is contained in:
Cline committed 2026-10-13 19:41:03 -05:00
1 parent 6b9fb54b4f
commit a829d6e8c6
21 files changed
+1903 -61

No files matched your search

+43 -1
View File
@@ -6,12 +6,54 @@ namespace MeshVault\Catalog;
use MeshVault\Repository\CatalogRepositoryInterface;
final class CatalogService
final class CatalogService implements CatalogServiceInterface
{
public function __construct(private readonly CatalogRepositoryInterface $catalog)
{
}
/** @see CatalogServiceInterface::atomic() for supported lock-order compositions. */
public function atomic(callable $operation): mixed
{
return $this->catalog->atomic($operation);
}
public function modelMetadata(int $modelId): ?array
{
return $this->catalog->modelMetadata($modelId);
}
public function updateModelMetadata(int $modelId, array $expected, array $replacement): array
{
return $this->catalog->updateModelMetadata($modelId, $expected, $replacement);
}
public function changeFileStatus(int $fileId, string $expectedStatus, string $targetStatus): array
{
return $this->catalog->changeFileStatus($fileId, $expectedStatus, $targetStatus);
}
public function file(int $fileId): ?array
{
return $this->catalog->file($fileId);
}
public function fileByManagedPath(string $managedPath): ?array
{
return $this->catalog->fileByManagedPath($managedPath);
}
public function registerFile(
?int $modelId,
string $managedPath,
string $hash,
int $size,
?string $mime = null,
string $status = 'Present'
): int {
return $this->catalog->recordFile($modelId, $managedPath, $hash, $size, $mime, $status);
}
public function createModel(string $folderPath, string $modelName): int
{
return $this->catalog->atomic(fn (): int => $this->catalog->createModel($folderPath, $modelName));
+55
View File
@@ -0,0 +1,55 @@
<?php
declare(strict_types=1);
namespace MeshVault\Catalog;
/**
* Shared public boundary for catalog operations used by importers and consumers.
*/
interface CatalogServiceInterface
{
/**
* Compose on the shared connection, using savepoints inside caller transactions.
* Success never commits caller work; failure rolls back only this operation.
* Supported: creation, then metadata, then ownership/status; or one registration,
* then item/job progress. Acquire multiple file locks in globally ascending ID order.
* Never enter catalog mutation holding job/item locks, acquire root/folder or
* metadata locks after file locks, or follow registration with metadata mutation.
* Arbitrary registration loops/batches are not supported: this is not a lock scheduler.
* Metadata uses FOR NO KEY UPDATE, compatible with ownership's implicit FK key-share.
*/
public function atomic(callable $operation): mixed;
public function modelMetadata(int $modelId): ?array;
public function updateModelMetadata(int $modelId, array $expected, array $replacement): array;
public function changeFileStatus(int $fileId, string $expectedStatus, string $targetStatus): array;
public function file(int $fileId): ?array;
public function fileByManagedPath(string $managedPath): ?array;
public function registerFile(
?int $modelId,
string $managedPath,
string $hash,
int $size,
?string $mime = null,
string $status = 'Present'
): int;
public function createModel(string $folderPath, string $modelName): int;
/**
* @return array{model: array, folder: array, files: array}|null
*/
public function aggregate(int $modelId): ?array;
public function assignFiles(int $modelId, array $fileIds): void;
public function reassignFiles(int $expectedModelId, int $targetModelId, array $fileIds): void;
public function detachFiles(int $expectedModelId, array $fileIds): void;
}
+75
View File
@@ -0,0 +1,75 @@
<?php
declare(strict_types=1);
namespace MeshVault\Catalog;
use InvalidArgumentException;
final class ModelMetadata
{
public static function validateTitle(string $title): void
{
$length = preg_match_all('/./us', $title);
if (
$length === false || $length < 1 || $length > 255 || preg_match('/^\s*$/u', $title)
|| str_contains($title, "\0")
) {
throw new InvalidArgumentException('Model name must contain 1 to 255 characters and not be blank or NUL.');
}
}
public static function validateShape(array $metadata): void
{
if (count($metadata) !== 3 || !isset($metadata['title']) || !is_string($metadata['title'])) {
throw new InvalidArgumentException('Metadata requires exactly title, notes and sourceURL.');
}
foreach (['notes', 'sourceURL'] as $field) {
if (
!array_key_exists($field, $metadata)
|| ($metadata[$field] !== null && !is_string($metadata[$field]))
) {
throw new InvalidArgumentException('Metadata requires nullable string notes and sourceURL.');
}
}
}
/** Validate/normalize changed fields only; expected is always a raw snapshot. */
public static function replacement(array $expected, array $replacement): array
{
self::validateShape($expected);
self::validateShape($replacement);
$result = [];
foreach (['title', 'notes', 'sourceURL'] as $field) {
$value = $replacement[$field];
if ($value !== $expected[$field]) {
if ($field === 'title') {
self::validateTitle($value);
} elseif ($value === '' || $value === null) {
$value = null;
} elseif ($field === 'notes') {
if (preg_match('//u', $value) !== 1 || str_contains($value, "\0")) {
throw new InvalidArgumentException('Notes must be valid UTF-8 without NUL.');
}
} else {
self::validateUrl($value);
}
}
$result[$field] = $value;
}
return $result;
}
private static function validateUrl(string $url): void
{
$parts = parse_url($url);
if (
strlen($url) > 2048 || preg_match('/[^\x21-\x7e]|\\\\|%(?![0-9a-fA-F]{2})/', $url)
|| $parts === false || !in_array(strtolower($parts['scheme'] ?? ''), ['http', 'https'], true)
|| empty($parts['host']) || isset($parts['user']) || isset($parts['pass'])
|| filter_var($url, FILTER_VALIDATE_URL) === false
) {
throw new InvalidArgumentException('Source URL must be an absolute ASCII HTTP(S) URL without userinfo.');
}
}
}
+3 -16
View File
@@ -5,14 +5,13 @@ declare(strict_types=1);
namespace MeshVault\Indexer;
use MeshVault\Indexer\Contracts\FileEnumeratorInterface;
use MeshVault\Repository\CatalogRepositoryInterface;
use MeshVault\Repository\ImportJobRepositoryInterface;
use Throwable;
/**
* Turns a source tree into a durable import job manifest.
*
* Grouping policy (explicit, never inferred from file contents): each file is
* Legacy guessed grouping (not user-confirmed associations): each file is
* assigned to the model named after its first directory component below the
* source root; files sitting directly in the source root join a model named
* after the source directory itself. Every model lives in a folder of the same
@@ -24,7 +23,7 @@ final readonly class ImportService
public function __construct(
private FileEnumeratorInterface $enumerator,
private ImportJobRepositoryInterface $jobs,
private CatalogRepositoryInterface $catalog,
private LegacyGuessedImportGroupingAdapter $groupingAdapter,
private SpaceChecker $space,
private string $libraryRoot,
) {
@@ -59,15 +58,12 @@ final readonly class ImportService
/** @return list<array<string, mixed>> */
private function manifest(string $sourcePath, ScanResult $scan): array
{
$models = [];
$items = [];
foreach ($scan->files as $file) {
$group = $this->group($sourcePath, $file);
$models[$group] ??= $this->catalog->ensureModel('/' . $group, $group);
$items[] = [
'source_path' => $file->sourcePath,
'size' => $file->size,
'model_id' => $models[$group],
'model_id' => $this->groupingAdapter->modelFor($sourcePath, $file),
];
}
// Scan failures become terminal items so the job reports them durably.
@@ -82,13 +78,4 @@ final readonly class ImportService
}
return $items;
}
private function group(string $sourcePath, ScannedFile $file): string
{
$parent = trim(str_replace('\\', '/', dirname($file->relativePath)), '/');
if ($parent === '' || $parent === '.') {
return basename(rtrim(str_replace('\\', '/', $sourcePath), '/'));
}
return explode('/', $parent)[0];
}
}
+11 -9
View File
@@ -7,10 +7,11 @@ namespace MeshVault\Indexer;
use MeshVault\Exception\PathSafetyException;
use MeshVault\Indexer\Filesystem\PathSafety;
use MeshVault\Indexer\SpaceChecker;
use MeshVault\Repository\CatalogRepositoryInterface;
use MeshVault\Catalog\CatalogServiceInterface;
use MeshVault\Repository\ImportJobRepositoryInterface;
use Psr\Log\LoggerInterface;
use Throwable;
use DomainException;
/**
* Serializes every managed-library mutation behind one advisory lock file and
@@ -32,7 +33,7 @@ final readonly class ImportWorker
/** @param null|callable(): bool $shouldStop Consulted before each item. */
public function __construct(
private ImportJobRepositoryInterface $jobs,
private CatalogRepositoryInterface $catalog,
private CatalogServiceInterface $catalog,
private StreamingHasher $hasher,
private VerifiedCopier $copier,
private StlValidator $validator,
@@ -112,20 +113,21 @@ final readonly class ImportWorker
$published = $this->publish($plan, $destination);
$relative = $this->relative($published);
// Catalog and progress commit together, so a crash here simply replays.
$this->jobs->atomic(function () use ($item, $itemId, $plan, $published, $relative): void {
// Update managed_path to the actual published path (with collision suffix if any)
// so import_items.managed_path always points at the published file.
$this->jobs->updateManagedPath($itemId, $relative);
if ($item['model_id'] !== null) {
$this->catalog->recordFile(
(int) $item['model_id'],
$this->catalog->atomic(function () use ($item, $itemId, $plan, $relative): void {
// Catalog locks precede item/job locks on the shared PDO connection.
try {
$this->catalog->registerFile(
$item['model_id'] === null ? null : (int) $item['model_id'],
$relative,
$plan['hash'],
$plan['size'],
null,
'Present'
);
} catch (DomainException $error) {
throw new ImportException('CATALOG_CONFLICT', $error->getMessage(), $error);
}
$this->jobs->updateManagedPath($itemId, $relative);
$this->jobs->finishItem($itemId, 'Completed');
});
} catch (Throwable $error) {
@@ -0,0 +1,27 @@
<?php
declare(strict_types=1);
namespace MeshVault\Indexer;
/**
* Stateless adapter for legacy guessed import grouping.
* Mimics the original grouping logic: top-level directory under source root,
* or source directory name for files in the source root.
*/
final class LegacyGuessedImportGroupingAdapter
{
public function __construct(private readonly \MeshVault\Repository\CatalogRepositoryInterface $catalog)
{
}
public function modelFor(string $sourcePath, ScannedFile $file): int
{
$parent = trim(dirname(str_replace('\\', '/', $file->relativePath)), '/');
$group = $parent === '' || $parent === '.'
? basename(rtrim(str_replace('\\', '/', $sourcePath), '/'))
: explode('/', $parent)[0];
// No retained IDs: rollback and later ambiguity must be visible on every call.
return $this->catalog->ensureModel('/' . $group, $group);
}
}
+13 -1
View File
@@ -15,6 +15,11 @@ interface CatalogRepositoryInterface
/** Legacy lookup: create on zero matches, reuse one, reject ambiguity. */
public function ensureModel(string $folderPath, string $modelName): int;
/** Read a normalized stored row without acquiring locks or creating rows. */
public function file(int $fileId): ?array;
public function fileByManagedPath(string $managedPath): ?array;
public function recordFile(
?int $modelId,
string $managedPath,
@@ -35,5 +40,12 @@ interface CatalogRepositoryInterface
*/
public function changeFileOwner(array $fileIds, ?int $expectedOwnerId, ?int $targetOwnerId): void;
public function setFileStatus(int $fileId, string $status): void;
/** Raw observation; absent models return null. */
public function modelMetadata(int $modelId): ?array;
/** Full raw named-tuple CAS; validate changed content only after locked comparison. */
public function updateModelMetadata(int $modelId, array $expected, array $replacement): array;
/** Locked status CAS, including no-ops; returns the normalized complete file row. */
public function changeFileStatus(int $fileId, string $expectedStatus, string $targetStatus): array;
}
+128 -15
View File
@@ -7,6 +7,7 @@ namespace MeshVault\Repository;
use DomainException;
use InvalidArgumentException;
use MeshVault\Catalog\ManagedPath;
use MeshVault\Catalog\ModelMetadata;
use PDO;
final class PdoCatalogRepository extends PdoRepository implements CatalogRepositoryInterface
@@ -28,10 +29,7 @@ final class PdoCatalogRepository extends PdoRepository implements CatalogReposit
private function createOrFindModel(string $folderPath, string $modelName, bool $legacy): int
{
$length = preg_match_all('/./us', $modelName);
if ($length === false || $length > 255 || preg_match('/^\s*$/u', $modelName)) {
throw new InvalidArgumentException('Model name must contain 1 to 255 characters and not be blank.');
}
ModelMetadata::validateTitle($modelName);
if ($folderPath === '') {
throw new InvalidArgumentException('A home folder is required.');
}
@@ -63,6 +61,40 @@ final class PdoCatalogRepository extends PdoRepository implements CatalogReposit
$this->execute("UPDATE folders SET name = name WHERE path = '/'");
}
public function file(int $fileId): ?array
{
if ($fileId <= 0) {
throw new InvalidArgumentException('File ID must be positive.');
}
return $this->normalizeFile($this->row('SELECT * FROM files WHERE id = ?', [$fileId]));
}
public function fileByManagedPath(string $managedPath): ?array
{
ManagedPath::validate($managedPath);
$this->validateText($managedPath);
return $this->normalizeFile($this->row('SELECT * FROM files WHERE managed_path = ?', [$managedPath]));
}
private function normalizeFile(?array $file): ?array
{
if ($file === null) {
return null;
}
foreach (['id', 'model_id', 'folder_id', 'size'] as $field) {
$file[$field] = $file[$field] === null ? null : (int) $file[$field];
}
return $file;
}
private function validateText(string $value, ?int $maximum = null): void
{
$length = preg_match_all('/./us', $value);
if ($length === false || str_contains($value, "\0") || ($maximum !== null && $length > $maximum)) {
throw new InvalidArgumentException('File text must be valid UTF-8 without NUL and within length limits.');
}
}
public function recordFile(
?int $modelId,
string $managedPath,
@@ -73,6 +105,11 @@ final class PdoCatalogRepository extends PdoRepository implements CatalogReposit
): int {
$this->validateStatus($status);
ManagedPath::validate($managedPath);
$this->validateText($managedPath);
$this->validateText($hash, 64);
if ($mime !== null) {
$this->validateText($mime, 128);
}
if ($modelId !== null && $modelId <= 0) {
throw new InvalidArgumentException('Model ID must be positive or null.');
}
@@ -82,18 +119,22 @@ final class PdoCatalogRepository extends PdoRepository implements CatalogReposit
return $this->transaction(function () use ($modelId, $managedPath, $hash, $size, $mime, $status): int {
// Serialize catalog writers on the root row, including paths assigned to different models.
$this->lockRoot();
$existing = $this->row('SELECT * FROM files WHERE managed_path = ?', [$managedPath]);
$lock = $this->pdo->getAttribute(PDO::ATTR_DRIVER_NAME) === 'pgsql' ? ' FOR UPDATE' : '';
$existing = $this->row('SELECT * FROM files WHERE managed_path = ?' . $lock, [$managedPath]);
if ($existing !== null) {
$owner = $existing['model_id'] === null ? null : (int) $existing['model_id'];
if (
$owner !== $modelId || $existing['hash'] !== $hash
|| (int) $existing['size'] !== $size
|| $existing['size'] === null || (int) $existing['size'] !== $size
) {
throw new DomainException('Managed path already records a different file.');
}
return (int) $existing['id'];
}
$folder = $this->ensureFolder(dirname(str_replace('\\', '/', $managedPath)));
if ($modelId !== null && $this->row('SELECT id FROM models WHERE id = ?', [$modelId]) === null) {
throw new DomainException('Unknown model.');
}
$folder = $this->ensureFolder(dirname($managedPath));
$file = $this->row(
'INSERT INTO files (model_id, folder_id, name, managed_path, hash, size, mime_type, status)'
. ' VALUES (?, ?, ?, ?, ?, ?, ?, ?) RETURNING id',
@@ -164,17 +205,89 @@ final class PdoCatalogRepository extends PdoRepository implements CatalogReposit
});
}
public function setFileStatus(int $fileId, string $status): void
public function modelMetadata(int $modelId): ?array
{
$this->validateStatus($status);
if (
$this->validateModelId($modelId);
return $this->readMetadata($modelId);
}
private function validateModelId(int $modelId): void
{
if ($modelId <= 0) {
throw new InvalidArgumentException('Model ID must be positive.');
}
}
private function readMetadata(int $modelId, string $lock = ''): ?array
{
$row = $this->row('SELECT name, notes, source_url FROM models WHERE id = ?' . $lock, [$modelId]);
return $row === null ? null : [
'title' => $row['name'], 'notes' => $row['notes'], 'sourceURL' => $row['source_url'],
];
}
public function updateModelMetadata(int $modelId, array $expected, array $replacement): array
{
$this->validateModelId($modelId);
ModelMetadata::validateShape($expected);
ModelMetadata::validateShape($replacement);
return $this->transaction(function () use ($modelId, $expected, $replacement): array {
// Compatible with ownership's implicit FK KEY SHARE; never strengthen to FOR UPDATE.
$lock = $this->pdo->getAttribute(PDO::ATTR_DRIVER_NAME) === 'pgsql' ? ' FOR NO KEY UPDATE' : '';
$current = $this->readMetadata($modelId, $lock);
if ($current === null) {
throw new DomainException('Unknown model.');
}
foreach ($current as $field => $value) {
if ($expected[$field] !== $value) {
throw new DomainException('Model metadata has changed.');
}
}
$next = ModelMetadata::replacement($expected, $replacement);
$columns = [];
$values = [];
foreach (['title' => 'name', 'notes' => 'notes', 'sourceURL' => 'source_url'] as $field => $column) {
if ($next[$field] !== $current[$field]) {
$columns[] = $column . ' = ?';
$values[] = $next[$field];
}
}
if ($columns !== []) {
$values[] = $modelId;
$this->execute(
'UPDATE models SET ' . implode(', ', $columns) . ', updated_at = CURRENT_TIMESTAMP WHERE id = ?',
$values
);
}
return $next;
});
}
public function changeFileStatus(int $fileId, string $expectedStatus, string $targetStatus): array
{
if ($fileId <= 0) {
throw new InvalidArgumentException('File ID must be positive.');
}
$this->validateStatus($expectedStatus);
$this->validateStatus($targetStatus);
return $this->transaction(function () use ($fileId, $expectedStatus, $targetStatus): array {
$lock = $this->pdo->getAttribute(PDO::ATTR_DRIVER_NAME) === 'pgsql' ? ' FOR UPDATE' : '';
$file = $this->normalizeFile($this->row('SELECT * FROM files WHERE id = ?' . $lock, [$fileId]));
if ($file === null) {
throw new DomainException('Unknown file.');
}
if ($file['status'] !== $expectedStatus) {
throw new DomainException('File status has changed.');
}
if ($expectedStatus === $targetStatus) {
return $file;
}
$this->execute(
'UPDATE files SET status = ?, updated_at = CURRENT_TIMESTAMP WHERE id = ?',
[$status, $fileId]
)->rowCount() === 0
) {
throw new DomainException('Unknown file.');
}
[$targetStatus, $fileId]
);
return $this->file($fileId);
});
}
private function validateStatus(string $status): void