From 96b7f02b48a6a3231c1d781c2f50e884d688ada2 Mon Sep 17 00:00:00 2001
From: harsh mahajan <harshmahajan2345@gmail.com>
Date: Fri, 28 Aug 2026 13:29:32 +0530
Subject: [PATCH] fix(s3): retry refused connections and unresolvable hosts

The strategy returned null whenever there was no response, so it only
retried failures the service had answered. A refused connection or an
unresolvable host was passed straight through, even though the request never
reached the service and replaying it cannot duplicate an effect.

Seen in production: a Spaces endpoint refused connections for ~35 minutes and
every upload failed on the first attempt with CURLE_COULDNT_CONNECT.

Other transport failures stay non-retryable, since a reset or a timeout may
have been applied server-side.
---
 src/Storage/Device/S3/RetryStrategy.php       | 14 ++++++++--
 tests/Storage/Device/S3/RetryStrategyTest.php | 27 +++++++++++++++++++
 2 files changed, 39 insertions(+), 2 deletions(-)

diff --git a/src/Storage/Device/S3/RetryStrategy.php b/src/Storage/Device/S3/RetryStrategy.php
index fcd9309c..21aa0213 100644
--- a/src/Storage/Device/S3/RetryStrategy.php
+++ b/src/Storage/Device/S3/RetryStrategy.php
@@ -9,6 +9,8 @@
 use Psr\Http\Message\RequestInterface;
 use Psr\Http\Message\ResponseInterface;
 use Utopia\Client\Decorator\Retry\Strategy;
+use Utopia\Client\Exception\ConnectionException;
+use Utopia\Client\Exception\DnsException;
 
 /**
  * Retry strategy for transient S3 rate-limiting errors (e.g. SlowDown,
@@ -19,6 +21,10 @@
  * error code is not retried, while unparseable 429/503 responses fall back to
  * status-code detection.
  *
+ * A refused connection or an unresolvable host is retried too: the request
+ * never reached the service, so replaying it cannot duplicate an effect. Other
+ * transport failures are left alone, since the request may have been applied.
+ *
  * Waits use exponential backoff with full jitter so a fleet throttled at the
  * same moment does not retry in lockstep.
  * @see \Utopia\Tests\Storage\Device\S3\RetryStrategyTest
@@ -54,11 +60,15 @@ public function __construct(
 
     public function delay(RequestInterface $request, int $attempt, ?ResponseInterface $response, ?ClientExceptionInterface $error): ?float
     {
-        if ($attempt > $this->retries || ! $response instanceof ResponseInterface) {
+        if ($attempt > $this->retries) {
             return null;
         }
 
-        if (! $this->isTransient($response)) {
+        $retryable = $response instanceof ResponseInterface
+            ? $this->isTransient($response)
+            : $error instanceof ConnectionException || $error instanceof DnsException;
+
+        if (! $retryable) {
             return null;
         }
 
diff --git a/tests/Storage/Device/S3/RetryStrategyTest.php b/tests/Storage/Device/S3/RetryStrategyTest.php
index 3bd0ef2a..564aac25 100644
--- a/tests/Storage/Device/S3/RetryStrategyTest.php
+++ b/tests/Storage/Device/S3/RetryStrategyTest.php
@@ -5,6 +5,8 @@
 namespace Utopia\Tests\Storage\Device\S3;
 
 use PHPUnit\Framework\TestCase;
+use Utopia\Client\Exception\ConnectionException;
+use Utopia\Client\Exception\DnsException;
 use Utopia\Client\Exception\NetworkException;
 use Utopia\Psr7\Request;
 use Utopia\Psr7\Response;
@@ -79,6 +81,7 @@ public function testRetriesAreCapped(): void
         $this->assertNull($strategy->delay($this->request(), 3, $response, null));
     }
 
+    /** A reset connection may have been applied server-side, so it stays non-retryable. */
     public function testTransportErrorsAreNotRetried(): void
     {
         $error = new NetworkException($this->request(), 'Connection reset');
@@ -86,6 +89,30 @@ public function testTransportErrorsAreNotRetried(): void
         $this->assertNull(new RetryStrategy()->delay($this->request(), 1, null, $error));
     }
 
+    public function testRefusedConnectionIsRetried(): void
+    {
+        $error = new ConnectionException($this->request(), 'Could not connect to server');
+        $strategy = new RetryStrategy(delay: 0.5, randomizer: static fn(): float => 1.0);
+
+        $this->assertEqualsWithDelta(0.5, $strategy->delay($this->request(), 1, null, $error), PHP_FLOAT_EPSILON);
+    }
+
+    public function testUnresolvableHostIsRetried(): void
+    {
+        $error = new DnsException($this->request(), 'Could not resolve host');
+        $strategy = new RetryStrategy(delay: 0.5, randomizer: static fn(): float => 1.0);
+
+        $this->assertEqualsWithDelta(0.5, $strategy->delay($this->request(), 1, null, $error), PHP_FLOAT_EPSILON);
+    }
+
+    public function testConnectionRetriesAreCapped(): void
+    {
+        $error = new ConnectionException($this->request(), 'Could not connect to server');
+        $strategy = new RetryStrategy(retries: 2, delay: 0.5, randomizer: static fn(): float => 1.0);
+
+        $this->assertNull($strategy->delay($this->request(), 3, null, $error));
+    }
+
     public function testReadingTheBodyLeavesItReadable(): void
     {
         $body = '<?xml version="1.0" encoding="UTF-8"?><Error><Code>SlowDown</Code></Error>';