Merge prism-php#1008 — feat(gemini): extract rate limit violations and retry-after information

Author: wishborn

src/Providers/Gemini/Concerns/ProcessesRateLimits.php

@@ -0,0 +1,79 @@
+<?php
+
+declare(strict_types=1);
+
+namespace Prism\Prism\Providers\Gemini\Concerns;
+
+use Illuminate\Support\Carbon;
+use Prism\Prism\ValueObjects\ProviderRateLimit;
+
+trait ProcessesRateLimits
+{
+    /**
+     * @param  array<string, mixed>  $responseData
+     * @return ProviderRateLimit[]
+     */
+    protected function processRateLimits(array $responseData): array
+    {
+        $quotaViolations = data_get($this->responseDetail($responseData, 'QuotaFailure'), 'violations', []);
+
+        if (! is_array($quotaViolations)) {
+            return [];
+        }
+
+        $resetsAt = $this->buildResetsAtFromResponse($responseData);
+
+        return collect($quotaViolations)
+            ->filter(fn (mixed $violation): bool => is_array($violation))
+            ->map(fn (array $violation): ProviderRateLimit => new ProviderRateLimit(
+                name: (string) data_get($violation, 'quotaId', data_get($violation, 'quotaMetric', 'quota')),
+                limit: $this->toNullableInt(data_get($violation, 'quotaValue')),
+                remaining: null,
+                resetsAt: $resetsAt
+            ))
+            ->all();
+    }
+
+    /**
+     * @param  array<string, mixed>  $responseData
+     */
+    protected function extractRetryAfterSeconds(array $responseData): ?int
+    {
+        $retryDelay = data_get($this->responseDetail($responseData, 'RetryInfo'), 'retryDelay');
+
+        if (is_string($retryDelay) && preg_match('/^(?<seconds>\d+)(?:\.\d+)?s$/', $retryDelay, $matches) === 1) {
+            return (int) $matches['seconds'];
+        }
+
+        return null;
+    }
+
+    /**
+     * @param  array<string, mixed>  $responseData
+     */
+    private function buildResetsAtFromResponse(array $responseData): ?Carbon
+    {
+        $retryAfter = $this->extractRetryAfterSeconds($responseData);
+
+        return $retryAfter === null ? null : Carbon::now()->addSeconds($retryAfter);
+    }
+
+    private function toNullableInt(mixed $value): ?int
+    {
+        return is_int($value) || (is_string($value) && preg_match('/^\d+$/', $value) === 1)
+            ? (int) $value
+            : null;
+    }
+
+    /**
+     * @param  array<string, mixed>  $responseData
+     */
+    private function responseDetail(array $responseData, string $type): mixed
+    {
+        $details = data_get($responseData, 'error.details', []);
+
+        return is_array($details)
+            ? collect($details)->first(fn (mixed $detail): bool => data_get($detail, '@type') === "type.googleapis.com/google.rpc.$type")
+            : null;
+    }
+}

src/Providers/Gemini/Gemini.php

@@ -18,6 +18,7 @@
 use Prism\Prism\Exceptions\PrismRateLimitedException;
 use Prism\Prism\Images\Request as ImagesRequest;
 use Prism\Prism\Images\Response as ImagesResponse;
+use Prism\Prism\Providers\Gemini\Concerns\ProcessesRateLimits;
 use Prism\Prism\Providers\Gemini\Handlers\Audio;
 use Prism\Prism\Providers\Gemini\Handlers\Cache;
 use Prism\Prism\Providers\Gemini\Handlers\Embeddings;
@@ -36,6 +37,7 @@
 class Gemini extends Provider
 {
     use InitializesClient;
+    use ProcessesRateLimits;
 
     public function __construct(
         #[\SensitiveParameter] public readonly string $apiKey,
@@ -110,8 +112,13 @@ public function stream(TextRequest $request): Generator
 
     public function handleRequestException(string $model, RequestException $e): never
     {
+        $responseData = $e->response->json() ?? [];
+
         match ($e->response->getStatusCode()) {
-            429 => throw PrismRateLimitedException::make([]),
+            429 => throw PrismRateLimitedException::make(
+                rateLimits: $this->processRateLimits($responseData),
+                retryAfter: $this->extractRetryAfterSeconds($responseData)
+            ),
             503 => throw PrismProviderOverloadedException::make(class_basename($this)),
             default => $this->handleResponseErrors($e),
         };

tests/Providers/Gemini/ExceptionHandlingTest.php

@@ -38,6 +38,45 @@ function createGeminiMockResponse(int $statusCode, array $json = []): Response
         ->toThrow(PrismRateLimitedException::class);
 });
 
+it('extracts retry_after and rate limit details from quota violations', function (): void {
+    $mockResponse = createGeminiMockResponse(429, [
+        'error' => [
+            'code' => 429,
+            'message' => 'You exceeded your current quota. Please retry in 35.458759309s.',
+            'status' => 'RESOURCE_EXHAUSTED',
+            'details' => [
+                [
+                    '@type' => 'type.googleapis.com/google.rpc.QuotaFailure',
+                    'violations' => [
+                        [
+                            'quotaMetric' => 'generativelanguage.googleapis.com/generate_content_paid_tier_input_token_count',
+                            'quotaId' => 'GenerateContentPaidTierInputTokensPerModelPerMinute',
+                            'quotaValue' => '16000',
+                        ],
+                    ],
+                ],
+                [
+                    '@type' => 'type.googleapis.com/google.rpc.RetryInfo',
+                    'retryDelay' => '35s',
+                ],
+            ],
+        ],
+    ]);
+    $exception = new RequestException($mockResponse);
+
+    try {
+        $this->provider->handleRequestException('gemini-pro', $exception);
+        $this->fail('Expected PrismRateLimitedException to be thrown.');
+    } catch (PrismRateLimitedException $e) {
+        expect($e->retryAfter)->toBe(35)
+            ->and($e->rateLimits)->toHaveCount(1)
+            ->and($e->rateLimits[0]->name)->toBe('GenerateContentPaidTierInputTokensPerModelPerMinute')
+            ->and($e->rateLimits[0]->limit)->toBe(16000)
+            ->and($e->rateLimits[0]->remaining)->toBeNull()
+            ->and($e->rateLimits[0]->resetsAt)->not->toBeNull();
+    }
+});
+
 it('handles provider overloaded errors (503)', function (): void {
     $mockResponse = createGeminiMockResponse(503, []);
     $exception = new RequestException($mockResponse);