fix AWS S3 startup for production deployments

Pass S3_REGION and S3_PUBLIC_USE_SSL through compose, treat blank public
SSL as unset, and skip CreateBucket when IAM only allows access to an
existing bucket.

deploy/docker-compose.yml

@@ -24,7 +24,9 @@ services:
       S3_ENDPOINT: ${S3_ENDPOINT:-seaweedfs:8333}
       S3_PUBLIC_ENDPOINT: ${S3_PUBLIC_ENDPOINT:-localhost:8333}
       S3_BUCKET: ${S3_BUCKET:-officeconvert}
+      S3_REGION: ${S3_REGION:-}
       S3_USE_SSL: ${S3_USE_SSL:-false}
+      S3_PUBLIC_USE_SSL: ${S3_PUBLIC_USE_SSL:-}
       S3_ACCESS_KEY: ${S3_ACCESS_KEY:-minioadmin}
       S3_SECRET_KEY: ${S3_SECRET_KEY:-minioadmin}
       S3_SESSION_TTL_SECONDS: ${S3_SESSION_TTL_SECONDS:-3600}

docker-compose.yml

@@ -23,7 +23,9 @@ services:
       S3_ENDPOINT: ${S3_ENDPOINT:-seaweedfs:8333}
       S3_PUBLIC_ENDPOINT: ${S3_PUBLIC_ENDPOINT:-localhost:8333}
       S3_BUCKET: ${S3_BUCKET:-officeconvert}
+      S3_REGION: ${S3_REGION:-}
       S3_USE_SSL: ${S3_USE_SSL:-false}
+      S3_PUBLIC_USE_SSL: ${S3_PUBLIC_USE_SSL:-}
       S3_ACCESS_KEY: ${S3_ACCESS_KEY:-minioadmin}
       S3_SECRET_KEY: ${S3_SECRET_KEY:-minioadmin}
       S3_SESSION_TTL_SECONDS: ${S3_SESSION_TTL_SECONDS:-3600}

python/packages/server/src/officeconvert_server/config.py

@@ -31,10 +31,10 @@ class ServerConfig:
 def load_server_config() -> ServerConfig:
     """Load server configuration from environment variables."""
     s3_secure = os.getenv("S3_USE_SSL", "false").lower() == "true"
-    public_ssl_env = os.getenv("S3_PUBLIC_USE_SSL")
+    public_ssl_env = os.getenv("S3_PUBLIC_USE_SSL", "").strip()
     s3_public_secure = (
         public_ssl_env.lower() == "true"
-        if public_ssl_env is not None
+        if public_ssl_env
         else s3_secure
     )
     region_env = os.getenv("S3_REGION", "").strip()

python/packages/server/src/officeconvert_server/storage.py

@@ -86,16 +86,19 @@ class S3Store:
     def ensure_bucket(self, bucket_name: str) -> None:
         """Create a bucket if it does not already exist.
 
-        Uses CreateBucket only, not HeadBucket. Some S3-compatible stores
+        Tries CreateBucket first (idempotent on SeaweedFS and when the caller
-        (including SeaweedFS) mishandle or over-restrict HeadBucket; the MinIO
+        owns the bucket). AWS production IAM often grants object access only on
-        client's bucket_exists() maps non-NoSuchBucket errors to failures.
+        a pre-provisioned bucket; in that case CreateBucket returns
-        Idempotent create covers the same contract with fewer round trips.
+        AccessDenied even though HeadBucket succeeds.
         """
         try:
             self._client.make_bucket(bucket_name)
         except S3Error as exc:
             if exc.code in ("BucketAlreadyOwnedByYou", "BucketAlreadyExists"):
                 return
+            if exc.code in ("AccessDenied", "Forbidden"):
+                if self._client.bucket_exists(bucket_name):
+                    return
             raise
 
     def presigned_put_url(self, bucket_name: str, object_key: str, *, ttl_seconds: int) -> str: