modal 1.1.5.dev83__py3-none-any.whl → 1.3.1.dev8__py3-none-any.whl

modal_proto/api.proto

@@ -4,6 +4,7 @@ option go_package = "github.com/modal-labs/modal/go/proto";
 
 package modal.client;
 
+import "google/protobuf/any.proto";
 import "google/protobuf/empty.proto";
 import "google/protobuf/struct.proto";
 import "google/protobuf/timestamp.proto";
@@ -301,6 +302,7 @@ message AppCreateRequest {
   string description = 2;    // Human readable label for the app
   string environment_name = 5;
   AppState app_state = 6;
+  map<string, string> tags = 7;  // Additional metadata to attach to the App
 }
 
 message AppCreateResponse {
@@ -366,6 +368,7 @@ message AppGetLogsRequest {
   float timeout = 2;
   string last_entry_id = 4;
   string function_id = 5;
+  string parametrized_function_id = 11;
   string input_id = 6;
   string task_id = 7;
   string function_call_id = 9;
@@ -728,6 +731,7 @@ message BuildFunction {
 message CancelInputEvent {
   repeated string input_ids = 1;
   bool terminate_containers = 2;
+  string cancellation_reason = 3;
 }
 
 message CheckpointInfo {
@@ -848,6 +852,12 @@ message CloudBucketMount {
     GCP = 3;
   }
 
+  enum MetadataTTLType {
+    METADATA_TTL_TYPE_UNSPECIFIED = 0;
+    METADATA_TTL_TYPE_MINIMAL = 1;
+    METADATA_TTL_TYPE_INDEFINITE = 2;
+  }
+
   string bucket_name = 1;
   string mount_path = 2;
   string credentials_secret_id = 3;
@@ -857,6 +867,11 @@ message CloudBucketMount {
   optional string bucket_endpoint_url = 7;
   optional string key_prefix = 8;
   optional string oidc_auth_role_arn = 9;
+  bool force_path_style = 10;
+  oneof metadata_ttl_oneof {
+    MetadataTTLType metadata_ttl_type = 11;
+    uint64 metadata_ttl_seconds = 12;
+  }
 }
 
 message ClusterGetRequest {
@@ -1277,6 +1292,10 @@ message EnvironmentListItem {
   bool default = 4;
   bool is_managed = 5;
   string environment_id = 6;
+  optional int32 max_concurrent_tasks = 7;
+  optional int32 max_concurrent_gpus = 8;
+  int32 current_concurrent_tasks = 9;
+  int32 current_concurrent_gpus = 10;
 }
 
 message EnvironmentListResponse {
@@ -1300,6 +1319,8 @@ message EnvironmentUpdateRequest {
   string current_name = 1;
   google.protobuf.StringValue name = 2;
   google.protobuf.StringValue web_suffix = 3;
+  optional int32 max_concurrent_tasks = 4;
+  optional int32 max_concurrent_gpus = 5;
 }
 
 // A file entry when listing files in a volume or network file system.
@@ -1448,6 +1469,8 @@ message Function {
 
   bool block_network = 44;
 
+  // In the SDK, we've deprecated `max_inputs` (which only every implemented `max_inputs=1`)
+  // in favor of a boolean `single_use_containers` parameter.
   uint32 max_inputs = 46;
 
   repeated S3Mount s3_mounts = 47;
@@ -1530,6 +1553,14 @@ message Function {
   uint32 startup_timeout_secs = 86;
   repeated DataFormat supported_input_formats = 87; // can be used as inputs
   repeated DataFormat supported_output_formats = 88;
+  optional HTTPConfig http_config = 89;
+
+  // Attribute on the module with the implementation, which may differ from function_name
+  // when the user provided a custom name= for the Function inside the Application namespace
+  string implementation_name = 90;
+
+  bool single_use_containers = 91; // When True, containers will shut down after handling a single input
+
 }
 
 message FunctionAsyncInvokeRequest {
@@ -1585,6 +1616,7 @@ message FunctionCallGetDataRequest {
     string attempt_token = 3;
   }
   uint64 last_index = 2;
+  bool use_gapless_read = 4;
 }
 
 message FunctionCallInfo {
@@ -1707,6 +1739,11 @@ message FunctionData {
   uint32 startup_timeout_secs = 36;
   repeated DataFormat supported_input_formats = 37;
   repeated DataFormat supported_output_formats = 38;
+  optional HTTPConfig http_config = 39;
+
+  // Attribute on the module with the implementation, which may differ from function_name
+  // when the user provided a custom name= for the Function inside the Application namespace
+  string implementation_name = 40;
 }
 
 message FunctionExtended {
@@ -2025,6 +2062,7 @@ message GPUConfig {
   string gpu_type = 4;
 }
 
+
 message GeneratorDone {  // Sent as the output when a generator finishes running.
   uint64 items_total = 1;
 }
@@ -2066,6 +2104,14 @@ message GenericResult {  // Used for both tasks and function outputs
   string propagation_reason = 13; // (?)
 }
 
+message HTTPConfig {
+  uint32 port = 1;
+  repeated string proxy_regions = 2;
+  uint32 startup_timeout = 3;
+  uint32 exit_grace_period = 4;
+  bool h2_enabled = 5;
+}
+
 message Image {
   repeated BaseImage base_images = 5;
   repeated string dockerfile_commands = 6;
@@ -2446,8 +2492,9 @@ message Proxy {
   string name = 1;
   double created_at = 2;
   string environment_name = 3;
-  string proxy_id = 5;
   repeated ProxyIp proxy_ips = 4;
+  string proxy_id = 5;
+  string region = 6;
 }
 
 message ProxyAddIpRequest {
@@ -2461,6 +2508,7 @@ message ProxyAddIpResponse {
 message ProxyCreateRequest {
   string name = 1;
   string environment_name = 2;
+  string region = 3;
 }
 
 message ProxyCreateResponse {
@@ -2609,6 +2657,22 @@ message QueuePutRequest {
   int32 partition_ttl_seconds = 6;
 }
 
+// Retry repolicy used by GRPCError.details for the server to give instructions
+// for the client to retry.
+message RPCRetryPolicy {
+  float retry_after_secs = 1;
+}
+
+// A copy google.rpc.Status for GRPCError.details:
+// https://github.com/googleapis/googleapis/blob/master/google/rpc/status.proto
+// RPCStatus is compatible with google.rpc.Status, so one can encode messages using
+// google.rpc.Status. The `details` field can be a list of any message, but for client
+// to decode it, the messages should be defined here (`modal_proto`).
+message RPCStatus {
+  int32 code = 1;
+  string message = 2;
+  repeated google.protobuf.Any details = 3;
+}
 
 message RateLimit {
   int32 limit = 1;
@@ -2735,6 +2799,11 @@ message Sandbox {
   // Exec commands for the sandbox will be issued directly to the sandbox
   // command router running on the Modal worker.
   bool direct_sandbox_commands_enabled = 34;
+
+  // Internal: restricts sandbox to run on this specific instance type.
+  // Set by server during SandboxRestore to ensure the restored sandbox runs
+  // on the same instance type as the original snapshot.
+  string _restore_instance_type = 35;
 }
 
 message SandboxCreateConnectTokenRequest {
@@ -2757,18 +2826,6 @@ message SandboxCreateResponse {
   string sandbox_id = 1;
 }
 
-// Used to get a JWT and URL for direct access to a sandbox router server
-// running on the modal-worker, so the client can issue exec commands (and other
-// operations as they become available) directly to the worker.
-message SandboxGetCommandRouterAccessRequest {
-  string sandbox_id = 1;
-}
-
-message SandboxGetCommandRouterAccessResponse {
-  string jwt = 1;
-  string url = 2;
-}
-
 message SandboxGetFromNameRequest {
   string sandbox_name = 1;
   string environment_name = 2;
@@ -2829,7 +2886,7 @@ message SandboxInfo {
   string app_id = 5;
   repeated SandboxTag tags = 6; // TODO: Not yet exposed in client library.
   string name = 7;
-
+  string image_id = 8;
   reserved 2; // modal.client.Sandbox definition
 }
 
@@ -2977,17 +3034,13 @@ message Schedule {
   }
 }
 
+// Scheduling constraints for Functions and Sandboxes.
 message SchedulerPlacement {
-  // TODO(irfansharif):
-  // - Fold in cloud, resource needs here too.
-  // - Allow specifying list of zones, cloud, fallback and alternative
-  //   GPU types.
-
   repeated string regions = 4;
-  // TODO(irfansharif): Make these two repeated.
-  optional string _zone = 2; // admin-only
-  optional string _lifecycle = 3; // admin-only, "on-demand" or "spot", else ignored
-  repeated string _instance_types = 5; // admin-only
+  optional string _zone = 2 [deprecated = true];
+  optional string _lifecycle = 3 [deprecated = true];
+  repeated string _instance_types = 5 [deprecated = true];
+  bool nonpreemptible = 6;  // Functions only
 
   reserved 1;
 }
@@ -3155,12 +3208,16 @@ message TaskCurrentInputsResponse {
   repeated string input_ids = 1;
 }
 
-message TaskGetAutoscalingMetricsRequest {
+// Used to get a JWT and URL for direct access to a task command router
+// running on the modal-worker, so the client can issue exec commands (and other
+// operations as they become available) directly to the worker.
+message TaskGetCommandRouterAccessRequest {
   string task_id = 1;
 }
 
-message TaskGetAutoscalingMetricsResponse {
-  AutoscalingMetrics metrics = 1;
+message TaskGetCommandRouterAccessResponse {
+  string jwt = 1;
+  string url = 2;
 }
 
 message TaskInfo {
@@ -3551,16 +3608,16 @@ message WebUrlInfo {
 message WebhookConfig {
   WebhookType type = 1;
   string method = 2;
-  string requested_suffix = 4;
+  string requested_suffix = 4;  // User-supplied "label" component of URL
   WebhookAsyncMode async_mode = 5;
   repeated CustomDomainConfig custom_domains = 6;
   uint32 web_server_port = 7;
   float web_server_startup_timeout = 8;
   bool web_endpoint_docs = 9;
   bool requires_proxy_auth = 10;
+  string ephemeral_suffix = 11;  // Additional URL suffix added for ephemeral Apps
 }
 
-
 message WorkspaceBillingReportItem {
   string object_id = 1;
   string description = 2;
@@ -3740,7 +3797,6 @@ service ModalClient {
   // Sandboxes
   rpc SandboxCreate(SandboxCreateRequest) returns (SandboxCreateResponse);
   rpc SandboxCreateConnectToken(SandboxCreateConnectTokenRequest) returns (SandboxCreateConnectTokenResponse);
-  rpc SandboxGetCommandRouterAccess(SandboxGetCommandRouterAccessRequest) returns (SandboxGetCommandRouterAccessResponse);
   rpc SandboxGetFromName(SandboxGetFromNameRequest) returns (SandboxGetFromNameResponse);
   rpc SandboxGetLogs(SandboxGetLogsRequest) returns (stream TaskLogsBatch);
   rpc SandboxGetResourceUsage(SandboxGetResourceUsageRequest) returns (SandboxGetResourceUsageResponse);
@@ -3779,7 +3835,7 @@ service ModalClient {
   // Tasks
   rpc TaskClusterHello(TaskClusterHelloRequest) returns (TaskClusterHelloResponse);
   rpc TaskCurrentInputs(google.protobuf.Empty) returns (TaskCurrentInputsResponse);
-  rpc TaskGetAutoscalingMetrics(TaskGetAutoscalingMetricsRequest) returns (TaskGetAutoscalingMetricsResponse); // Used for flash autoscaling
+  rpc TaskGetCommandRouterAccess(TaskGetCommandRouterAccessRequest) returns (TaskGetCommandRouterAccessResponse);
   rpc TaskList(TaskListRequest) returns (TaskListResponse);
   rpc TaskResult(TaskResultRequest) returns (google.protobuf.Empty);
 

modal_proto/api_grpc.py

@@ -9,6 +9,7 @@ import grpclib.client
 if typing.TYPE_CHECKING:
     import grpclib.server
 
+import google.protobuf.any_pb2
 import google.protobuf.empty_pb2
 import google.protobuf.struct_pb2
 import google.protobuf.timestamp_pb2
@@ -482,10 +483,6 @@ class ModalClientBase(abc.ABC):
     async def SandboxCreateConnectToken(self, stream: 'grpclib.server.Stream[modal_proto.api_pb2.SandboxCreateConnectTokenRequest, modal_proto.api_pb2.SandboxCreateConnectTokenResponse]') -> None:
         pass
 
-    @abc.abstractmethod
-    async def SandboxGetCommandRouterAccess(self, stream: 'grpclib.server.Stream[modal_proto.api_pb2.SandboxGetCommandRouterAccessRequest, modal_proto.api_pb2.SandboxGetCommandRouterAccessResponse]') -> None:
-        pass
-
     @abc.abstractmethod
     async def SandboxGetFromName(self, stream: 'grpclib.server.Stream[modal_proto.api_pb2.SandboxGetFromNameRequest, modal_proto.api_pb2.SandboxGetFromNameResponse]') -> None:
         pass
@@ -615,7 +612,7 @@ class ModalClientBase(abc.ABC):
         pass
 
     @abc.abstractmethod
-    async def TaskGetAutoscalingMetrics(self, stream: 'grpclib.server.Stream[modal_proto.api_pb2.TaskGetAutoscalingMetricsRequest, modal_proto.api_pb2.TaskGetAutoscalingMetricsResponse]') -> None:
+    async def TaskGetCommandRouterAccess(self, stream: 'grpclib.server.Stream[modal_proto.api_pb2.TaskGetCommandRouterAccessRequest, modal_proto.api_pb2.TaskGetCommandRouterAccessResponse]') -> None:
         pass
 
     @abc.abstractmethod
@@ -1416,12 +1413,6 @@ class ModalClientBase(abc.ABC):
                 modal_proto.api_pb2.SandboxCreateConnectTokenRequest,
                 modal_proto.api_pb2.SandboxCreateConnectTokenResponse,
             ),
-            '/modal.client.ModalClient/SandboxGetCommandRouterAccess': grpclib.const.Handler(
-                self.SandboxGetCommandRouterAccess,
-                grpclib.const.Cardinality.UNARY_UNARY,
-                modal_proto.api_pb2.SandboxGetCommandRouterAccessRequest,
-                modal_proto.api_pb2.SandboxGetCommandRouterAccessResponse,
-            ),
             '/modal.client.ModalClient/SandboxGetFromName': grpclib.const.Handler(
                 self.SandboxGetFromName,
                 grpclib.const.Cardinality.UNARY_UNARY,
@@ -1614,11 +1605,11 @@ class ModalClientBase(abc.ABC):
                 google.protobuf.empty_pb2.Empty,
                 modal_proto.api_pb2.TaskCurrentInputsResponse,
             ),
-            '/modal.client.ModalClient/TaskGetAutoscalingMetrics': grpclib.const.Handler(
-                self.TaskGetAutoscalingMetrics,
+            '/modal.client.ModalClient/TaskGetCommandRouterAccess': grpclib.const.Handler(
+                self.TaskGetCommandRouterAccess,
                 grpclib.const.Cardinality.UNARY_UNARY,
-                modal_proto.api_pb2.TaskGetAutoscalingMetricsRequest,
-                modal_proto.api_pb2.TaskGetAutoscalingMetricsResponse,
+                modal_proto.api_pb2.TaskGetCommandRouterAccessRequest,
+                modal_proto.api_pb2.TaskGetCommandRouterAccessResponse,
             ),
             '/modal.client.ModalClient/TaskList': grpclib.const.Handler(
                 self.TaskList,
@@ -2472,12 +2463,6 @@ class ModalClientStub:
             modal_proto.api_pb2.SandboxCreateConnectTokenRequest,
             modal_proto.api_pb2.SandboxCreateConnectTokenResponse,
         )
-        self.SandboxGetCommandRouterAccess = grpclib.client.UnaryUnaryMethod(
-            channel,
-            '/modal.client.ModalClient/SandboxGetCommandRouterAccess',
-            modal_proto.api_pb2.SandboxGetCommandRouterAccessRequest,
-            modal_proto.api_pb2.SandboxGetCommandRouterAccessResponse,
-        )
         self.SandboxGetFromName = grpclib.client.UnaryUnaryMethod(
             channel,
             '/modal.client.ModalClient/SandboxGetFromName',
@@ -2670,11 +2655,11 @@ class ModalClientStub:
             google.protobuf.empty_pb2.Empty,
             modal_proto.api_pb2.TaskCurrentInputsResponse,
         )
-        self.TaskGetAutoscalingMetrics = grpclib.client.UnaryUnaryMethod(
+        self.TaskGetCommandRouterAccess = grpclib.client.UnaryUnaryMethod(
             channel,
-            '/modal.client.ModalClient/TaskGetAutoscalingMetrics',
-            modal_proto.api_pb2.TaskGetAutoscalingMetricsRequest,
-            modal_proto.api_pb2.TaskGetAutoscalingMetricsResponse,
+            '/modal.client.ModalClient/TaskGetCommandRouterAccess',
+            modal_proto.api_pb2.TaskGetCommandRouterAccessRequest,
+            modal_proto.api_pb2.TaskGetCommandRouterAccessResponse,
         )
         self.TaskList = grpclib.client.UnaryUnaryMethod(
             channel,