Increase request body size limit for services (#1934)

This commit increases the limit from 1 MiB to 64
MiB and makes it configurable with a new
dstack-server environment variable. The design
will also allow to make the limit configurable
per-gateway and/or per-service in the future.

Author: jvstme

docs/docs/reference/cli/index.md

@@ -421,6 +421,7 @@ $ dstack pool delete --help
  * `DSTACK_DATABASE_URL` – (Optional) The database URL to use instead of default SQLite. Currently `dstack` supports Postgres. Example: `postgresql+asyncpg://myuser:mypassword@localhost:5432/mydatabase`. Defaults to `None`.
  * `DSTACK_SERVER_CLOUDWATCH_LOG_GROUP` – (Optional) The CloudWatch Logs group for workloads logs. If not set, the default file-based log storage is used.
  * `DSTACK_SERVER_CLOUDWATCH_LOG_REGION` — (Optional) The CloudWatch Logs region. Defaults to `None`.
+ * `DSTACK_DEFAULT_SERVICE_CLIENT_MAX_BODY_SIZE` – (Optional) Request body size limit for services, in bytes. Defaults to 64 MiB.
  * `DSTACK_SERVER_DIR` – (Optional) Sets path to store data and server configs. Defaults to `~/.dstack/server`.
 
 ??? info "Internal environment variables"

gateway/src/dstack/gateway/core/nginx.py

@@ -37,6 +37,8 @@ class ServiceConfig(SiteConfig):
     project: str
     service_id: str
     auth: bool
+    # Default for compatibility with older state.json. TODO: remove after a few releases
+    client_max_body_size: int = 1024 * 1024
     servers: Dict[str, str] = {}
 
 
@@ -75,7 +77,13 @@ async def set_acme_settings(
             )
 
     async def register_service(
-        self, project: str, service_id: str, domain: str, https: bool, auth: bool
+        self,
+        project: str,
+        service_id: str,
+        domain: str,
+        https: bool,
+        auth: bool,
+        client_max_body_size: int,
     ):
         config_name = self.get_config_name(domain)
         conf = ServiceConfig(
@@ -84,6 +92,7 @@ async def register_service(
             domain=domain,
             https=https,
             auth=auth,
+            client_max_body_size=client_max_body_size,
         )
 
         async with self._lock:

gateway/src/dstack/gateway/core/store.py

@@ -38,6 +38,8 @@ class Service(BaseModel):
     domain: str
     https: bool = True
     auth: bool
+    # Default for compatibility with older state.json. TODO: remove after a few releases
+    client_max_body_size: int = 1024 * 1024
     options: dict
     replicas: List[Replica] = []
 
@@ -107,6 +109,7 @@ async def register_service(self, project: str, service: Service, ssh_private_key
                     service.domain,
                     service.https,
                     service.auth,
+                    service.client_max_body_size,
                 )
                 stack.push_async_callback(
                     supress_exc_async(self.nginx.unregister_domain, service.domain)

gateway/src/dstack/gateway/registry/routes.py

@@ -24,6 +24,7 @@ async def post_register_service(
             https=body.https,
             auth=body.auth,
             options=body.options,
+            client_max_body_size=body.client_max_body_size,
         ),
         body.ssh_private_key,
     )

gateway/src/dstack/gateway/registry/schemas.py

@@ -8,6 +8,7 @@ class RegisterServiceRequest(BaseModel):
     domain: str
     https: bool = True
     auth: bool = True
+    client_max_body_size: int
     options: dict = {}
     ssh_private_key: str
 

gateway/src/dstack/gateway/resources/nginx/service.jinja2

@@ -11,6 +11,7 @@ server {
     server_name {{ domain }};
 
     access_log /var/log/nginx/dstack.access.log dstack_stat;
+    client_max_body_size {{ client_max_body_size }};
 
     location / {
         {% if auth %}

src/dstack/_internal/server/services/gateways/__init__.py

@@ -412,6 +412,7 @@ async def _register_service_in_gateway(
                 service_https=service_https,
                 gateway_https=gateway_https,
                 auth=run_spec.configuration.auth,
+                client_max_body_size=settings.DEFAULT_SERVICE_CLIENT_MAX_BODY_SIZE,
                 options=service_spec.options,
                 ssh_private_key=run_model.project.ssh_private_key,
             )

src/dstack/_internal/server/services/gateways/client.py

@@ -47,6 +47,7 @@ async def register_service(
         service_https: bool,
         gateway_https: bool,
         auth: bool,
+        client_max_body_size: int,
         options: dict,
         ssh_private_key: str,
     ):
@@ -59,6 +60,7 @@ async def register_service(
             "domain": domain,
             "https": service_https,
             "auth": auth,
+            "client_max_body_size": client_max_body_size,
             "options": options,
             "ssh_private_key": ssh_private_key,
         }

src/dstack/_internal/server/settings.py

@@ -50,6 +50,9 @@
 ACME_SERVER = os.getenv("DSTACK_ACME_SERVER")
 ACME_EAB_KID = os.getenv("DSTACK_ACME_EAB_KID")
 ACME_EAB_HMAC_KEY = os.getenv("DSTACK_ACME_EAB_HMAC_KEY")
+DEFAULT_SERVICE_CLIENT_MAX_BODY_SIZE = int(
+    os.getenv("DSTACK_DEFAULT_SERVICE_CLIENT_MAX_BODY_SIZE", 64 * 1024 * 1024)
+)
 
 USER_PROJECT_DEFAULT_QUOTA = int(os.getenv("DSTACK_USER_PROJECT_DEFAULT_QUOTA", 10))