harden struct

This commit is contained in:
team 1
2026-02-16 12:34:22 +01:00
parent c046a83767
commit 57a35af6c8
6 changed files with 308 additions and 252 deletions

View File

@@ -1,6 +1,5 @@
<?php
declare(strict_types=1);
namespace App\Index;
@@ -8,14 +7,14 @@ namespace App\Index;
final class IndexMetaManager
{
private string $metaPath;
private IndexConfiguration $config;
public function __construct(
string $projectDir,
private readonly IndexConfiguration $config,
string $relativeMetaPath = '/var/knowledge/index_meta.json'
)
{
$this->metaPath = rtrim($projectDir, '/') . $relativeMetaPath;
string $metaPath,
IndexConfiguration $config
) {
$this->metaPath = $metaPath;
$this->config = $config;
}
public function getMetaPath(): string
@@ -24,8 +23,6 @@ final class IndexMetaManager
}
/**
* Gibt null zurück, wenn noch kein Meta existiert (frisches System).
*
* @return array<string,mixed>|null
*/
public function readMeta(): ?array
@@ -48,43 +45,19 @@ final class IndexMetaManager
}
/**
* Erstellt Meta, falls nicht vorhanden (z. B. nach erstem Global Reindex).
* Überschreibt NICHT automatisch, wenn vorhanden.
*
* @return array<string,mixed>
*/
public function createInitialMetaIfMissing(): array
{
$existing = $this->readMeta();
if ($existing !== null) {
return $existing;
}
$meta = $this->buildMetaPayload(indexVersion: 1);
$this->atomicWriteJson($meta);
return $meta;
}
/**
* Guardrail: Prüft, ob die aktuelle Config kompatibel zur gespeicherten Meta ist.
* Wenn nicht: IndexStructureChangedException -> Global Reindex erzwingen.
* Guardrail:
* - Wenn Meta fehlt → initialisieren
* - Wenn Struktur driftet → Exception
*/
public function validateAgainstCurrent(): void
{
$meta = $this->readMeta();
// Wenn noch kein Meta existiert, lassen wir lokale Ingests NICHT einfach laufen.
// Governance: Erst Global Reindex erzeugt Meta sauber.
if ($meta === null) {
throw new IndexStructureChangedException(
'index_meta.json missing. Please run a Global Reindex to initialize index structure metadata.',
['reason' => 'missing_meta']
);
$meta = $this->createInitialMeta();
}
$expected = $this->config->toStructureArray();
$diff = $this->diffStructure($meta, $expected);
if ($diff !== []) {
@@ -96,11 +69,7 @@ final class IndexMetaManager
}
/**
* Wird beim Global Reindex verwendet:
* - index_version++ (oder initialisieren)
* - Meta atomar schreiben
*
* @return array<string,mixed> new meta
* Wird beim Global Reindex aufgerufen
*/
public function writeMetaForGlobalReindex(): array
{
@@ -122,35 +91,34 @@ final class IndexMetaManager
return $this->config;
}
// -------------------------
// ---------------------------------------------------------
// Internals
// -------------------------
// ---------------------------------------------------------
private function createInitialMeta(): array
{
$meta = $this->buildMetaPayload(1);
$this->atomicWriteJson($meta);
return $meta;
}
/**
* @return array<string,mixed>
*/
private function buildMetaPayload(int $indexVersion): array
{
$structure = $this->config->toStructureArray();
return [
'index_version' => $indexVersion,
'created_at' => (new \DateTimeImmutable())->format(DATE_ATOM),
'embedding_model' => $structure['embedding_model'],
'index_version' => $indexVersion,
'created_at' => (new \DateTimeImmutable())->format(DATE_ATOM),
'embedding_model' => $structure['embedding_model'],
'embedding_dimension' => $structure['embedding_dimension'],
'chunk_size' => $structure['chunk_size'],
'chunk_overlap' => $structure['chunk_overlap'],
'scoring_version' => $structure['scoring_version'],
'index_format' => $structure['index_format'],
'vector_backend' => $structure['vector_backend'],
'chunk_size' => $structure['chunk_size'],
'chunk_overlap' => $structure['chunk_overlap'],
'scoring_version' => $structure['scoring_version'],
'index_format' => $structure['index_format'],
'vector_backend' => $structure['vector_backend'],
];
}
/**
* @param array<string,mixed> $meta
* @param array<string,mixed> $expected
* @return array<string,mixed> diff
*/
private function diffStructure(array $meta, array $expected): array
{
$diff = [];
@@ -160,28 +128,18 @@ final class IndexMetaManager
if ($actual !== $value) {
$diff[$key] = [
'expected' => $value,
'actual' => $actual,
'actual' => $actual,
];
}
}
// index_format ist zwingend
if (($meta['index_format'] ?? null) !== 'ndjson') {
$diff['index_format'] = [
'expected' => 'ndjson',
'actual' => $meta['index_format'] ?? null,
];
}
return $diff;
}
/**
* @param array<string,mixed> $payload
*/
private function atomicWriteJson(array $payload): void
{
$dir = \dirname($this->metaPath);
$dir = dirname($this->metaPath);
if (!is_dir($dir) && !mkdir($dir, 0777, true) && !is_dir($dir)) {
throw new \RuntimeException('Unable to create directory: ' . $dir);
}
@@ -197,7 +155,6 @@ final class IndexMetaManager
throw new \RuntimeException('Unable to write temp meta file');
}
// atomarer Switch
if (!rename($tmp, $this->metaPath)) {
@unlink($tmp);
throw new \RuntimeException('Unable to switch meta file atomically');