Merge branch 'master' into is1779/search-users

Author: pcrespov

packages/models-library/src/models_library/api_schemas_directorv2/services.py

@@ -103,3 +103,7 @@ class ServiceExtras(BaseModel):
 
 
 CHARS_IN_VOLUME_NAME_BEFORE_DIR_NAME: Final[NonNegativeInt] = 89
+
+
+DYNAMIC_SIDECAR_SERVICE_PREFIX: Final[str] = "dy-sidecar"
+DYNAMIC_PROXY_SERVICE_PREFIX: Final[str] = "dy-proxy"

packages/models-library/src/models_library/docker.py

@@ -37,7 +37,15 @@ def from_key(cls, key: str) -> "DockerLabelKey":
     str, StringConstraints(pattern=DOCKER_GENERIC_TAG_KEY_RE)
 ]
 
-DockerPlacementConstraint: TypeAlias = Annotated[str, StringConstraints(strip_whitespace = True, pattern = re.compile(r"^(?!-)(?![.])(?!.*--)(?!.*[.][.])[a-zA-Z0-9.-]*(?<!-)(?<![.])(!=|==)[a-zA-Z0-9_. -]*$"))]
+DockerPlacementConstraint: TypeAlias = Annotated[
+    str,
+    StringConstraints(
+        strip_whitespace=True,
+        pattern=re.compile(
+            r"^(?!-)(?![.])(?!.*--)(?!.*[.][.])[a-zA-Z0-9.-]*(?<!-)(?<![.])(!=|==)[a-zA-Z0-9_. -]*$"
+        ),
+    ),
+]
 
 _SIMCORE_RUNTIME_DOCKER_LABEL_PREFIX: Final[str] = "io.simcore.runtime."
 _BACKWARDS_COMPATIBILITY_SIMCORE_RUNTIME_DOCKER_LABELS_MAP: Final[dict[str, str]] = {
@@ -218,3 +226,8 @@ def from_docker_task(cls, docker_task: Task) -> "StandardSimcoreDockerLabels":
             ]
         },
     )
+
+
+DockerNodeID: TypeAlias = Annotated[
+    str, StringConstraints(strip_whitespace=True, pattern=re.compile(r"[a-zA-Z0-9]"))
+]

packages/service-library/src/servicelib/rabbitmq/rpc_interfaces/agent/containers.py

@@ -0,0 +1,37 @@
+import logging
+from datetime import timedelta
+from typing import Final
+
+from models_library.docker import DockerNodeID
+from models_library.projects_nodes_io import NodeID
+from models_library.rabbitmq_basic_types import RPCMethodName, RPCNamespace
+from pydantic import NonNegativeInt, TypeAdapter
+from servicelib.logging_utils import log_decorator
+from servicelib.rabbitmq import RabbitMQRPCClient
+
+_logger = logging.getLogger(__name__)
+
+_REQUEST_TIMEOUT: Final[NonNegativeInt] = int(timedelta(minutes=60).total_seconds())
+
+
+@log_decorator(_logger, level=logging.DEBUG)
+async def force_container_cleanup(
+    rabbitmq_rpc_client: RabbitMQRPCClient,
+    *,
+    docker_node_id: DockerNodeID,
+    swarm_stack_name: str,
+    node_id: NodeID,
+) -> None:
+    result = await rabbitmq_rpc_client.request(
+        RPCNamespace.from_entries(
+            {
+                "service": "agent",
+                "docker_node_id": docker_node_id,
+                "swarm_stack_name": swarm_stack_name,
+            }
+        ),
+        TypeAdapter(RPCMethodName).validate_python("force_container_cleanup"),
+        node_id=node_id,
+        timeout_s=_REQUEST_TIMEOUT,
+    )
+    assert result is None  # nosec

packages/service-library/src/servicelib/rabbitmq/rpc_interfaces/agent/volumes.py

@@ -2,6 +2,7 @@
 from datetime import timedelta
 from typing import Final
 
+from models_library.docker import DockerNodeID
 from models_library.projects_nodes_io import NodeID
 from models_library.rabbitmq_basic_types import RPCMethodName, RPCNamespace
 from pydantic import NonNegativeInt, TypeAdapter
@@ -17,7 +18,7 @@
 async def remove_volumes_without_backup_for_service(
     rabbitmq_rpc_client: RabbitMQRPCClient,
     *,
-    docker_node_id: str,
+    docker_node_id: DockerNodeID,
     swarm_stack_name: str,
     node_id: NodeID,
 ) -> None:
@@ -42,7 +43,7 @@ async def remove_volumes_without_backup_for_service(
 async def backup_and_remove_volumes_for_all_services(
     rabbitmq_rpc_client: RabbitMQRPCClient,
     *,
-    docker_node_id: str,
+    docker_node_id: DockerNodeID,
     swarm_stack_name: str,
 ) -> None:
     result = await rabbitmq_rpc_client.request(

packages/service-library/src/servicelib/rabbitmq/rpc_interfaces/dynamic_scheduler/services.py

@@ -1,7 +1,10 @@
 import logging
 from typing import Final
 
-from models_library.api_schemas_directorv2.dynamic_services import DynamicServiceGet
+from models_library.api_schemas_directorv2.dynamic_services import (
+    DynamicServiceGet,
+    RetrieveDataOutEnveloped,
+)
 from models_library.api_schemas_dynamic_scheduler import DYNAMIC_SCHEDULER_RPC_NAMESPACE
 from models_library.api_schemas_dynamic_scheduler.dynamic_services import (
     DynamicServiceStart,
@@ -11,6 +14,7 @@
 from models_library.projects import ProjectID
 from models_library.projects_nodes_io import NodeID
 from models_library.rabbitmq_basic_types import RPCMethodName
+from models_library.services_types import ServicePortKey
 from models_library.users import UserID
 from pydantic import NonNegativeInt, TypeAdapter
 from servicelib.logging_utils import log_decorator
@@ -95,6 +99,41 @@ async def stop_dynamic_service(
     assert result is None  # nosec
 
 
+@log_decorator(_logger, level=logging.DEBUG)
+async def restart_user_services(
+    rabbitmq_rpc_client: RabbitMQRPCClient,
+    *,
+    node_id: NodeID,
+    timeout_s: NonNegativeInt,
+) -> None:
+    result = await rabbitmq_rpc_client.request(
+        DYNAMIC_SCHEDULER_RPC_NAMESPACE,
+        _RPC_METHOD_NAME_ADAPTER.validate_python("restart_user_services"),
+        node_id=node_id,
+        timeout_s=timeout_s,
+    )
+    assert result is None  # nosec
+
+
+@log_decorator(_logger, level=logging.DEBUG)
+async def retrieve_inputs(
+    rabbitmq_rpc_client: RabbitMQRPCClient,
+    *,
+    node_id: NodeID,
+    port_keys: list[ServicePortKey],
+    timeout_s: NonNegativeInt,
+) -> RetrieveDataOutEnveloped:
+    result = await rabbitmq_rpc_client.request(
+        DYNAMIC_SCHEDULER_RPC_NAMESPACE,
+        _RPC_METHOD_NAME_ADAPTER.validate_python("retrieve_inputs"),
+        node_id=node_id,
+        port_keys=port_keys,
+        timeout_s=timeout_s,
+    )
+    assert isinstance(result, RetrieveDataOutEnveloped)  # nosec
+    return result
+
+
 @log_decorator(_logger, level=logging.DEBUG)
 async def update_projects_networks(
     rabbitmq_rpc_client: RabbitMQRPCClient, *, project_id: ProjectID

services/agent/src/simcore_service_agent/api/rpc/_containers.py

@@ -0,0 +1,20 @@
+import logging
+
+from fastapi import FastAPI
+from models_library.projects_nodes_io import NodeID
+from servicelib.logging_utils import log_context
+from servicelib.rabbitmq import RPCRouter
+
+from ...services.containers_manager import ContainersManager
+
+_logger = logging.getLogger(__name__)
+
+router = RPCRouter()
+
+
+@router.expose()
+async def force_container_cleanup(app: FastAPI, *, node_id: NodeID) -> None:
+    with log_context(
+        _logger, logging.INFO, f"removing all orphan container for {node_id=}"
+    ):
+        await ContainersManager.get_from_app_state(app).force_container_cleanup(node_id)

services/agent/src/simcore_service_agent/api/rpc/_volumes.py

@@ -7,7 +7,8 @@
 from servicelib.rabbitmq.rpc_interfaces.agent.errors import (
     NoServiceVolumesFoundRPCError,
 )
-from simcore_service_agent.services.volumes_manager import VolumesManager
+
+from ...services.volumes_manager import VolumesManager
 
 _logger = logging.getLogger(__name__)
 

services/agent/src/simcore_service_agent/api/rpc/routes.py

@@ -4,9 +4,10 @@
 from simcore_service_agent.core.settings import ApplicationSettings
 
 from ...services.rabbitmq import get_rabbitmq_rpc_server
-from . import _volumes
+from . import _containers, _volumes
 
 ROUTERS: list[RPCRouter] = [
+    _containers.router,
     _volumes.router,
 ]
 

services/agent/src/simcore_service_agent/core/application.py

@@ -18,6 +18,7 @@
 )
 from ..api.rest.routes import setup_rest_api
 from ..api.rpc.routes import setup_rpc_api_routes
+from ..services.containers_manager import setup_containers_manager
 from ..services.instrumentation import setup_instrumentation
 from ..services.rabbitmq import setup_rabbitmq
 from ..services.volumes_manager import setup_volume_manager
@@ -28,8 +29,8 @@
 
 def _setup_logger(settings: ApplicationSettings):
     # SEE https://github.com/ITISFoundation/osparc-simcore/issues/3148
-    logging.basicConfig(level=settings.LOGLEVEL.value)  # NOSONAR
-    logging.root.setLevel(settings.LOGLEVEL.value)
+    logging.basicConfig(level=settings.LOG_LEVEL.value)  # NOSONAR
+    logging.root.setLevel(settings.LOG_LEVEL.value)
     config_all_loggers(
         log_format_local_dev_enabled=settings.AGENT_VOLUMES_LOG_FORMAT_LOCAL_DEV_ENABLED,
         logger_filter_mapping=settings.AGENT_VOLUMES_LOG_FILTER_MAPPING,
@@ -58,6 +59,7 @@ def create_app() -> FastAPI:
 
     setup_rabbitmq(app)
     setup_volume_manager(app)
+    setup_containers_manager(app)
     setup_rest_api(app)
     setup_rpc_api_routes(app)
 

services/agent/src/simcore_service_agent/core/settings.py

@@ -1,6 +1,7 @@
 from datetime import timedelta
 
 from models_library.basic_types import BootModeEnum, LogLevel
+from models_library.docker import DockerNodeID
 from pydantic import AliasChoices, AnyHttpUrl, Field, field_validator
 from servicelib.logging_utils_filtering import LoggerName, MessageSubstring
 from settings_library.base import BaseCustomSettings
@@ -11,7 +12,7 @@
 
 
 class ApplicationSettings(BaseCustomSettings, MixinLoggingSettings):
-    LOGLEVEL: LogLevel = Field(
+    LOG_LEVEL: LogLevel = Field(
         LogLevel.WARNING,
         validation_alias=AliasChoices(
             "AGENT_LOGLEVEL",
@@ -79,7 +80,9 @@ class ApplicationSettings(BaseCustomSettings, MixinLoggingSettings):
 
     AGENT_PROMETHEUS_INSTRUMENTATION_ENABLED: bool = True
 
-    AGENT_DOCKER_NODE_ID: str = Field(..., description="used by the rabbitmq module")
+    AGENT_DOCKER_NODE_ID: DockerNodeID = Field(
+        ..., description="used by the rabbitmq module"
+    )
 
     AGENT_RABBITMQ: RabbitSettings = Field(
         description="settings for service/rabbitmq",
@@ -91,7 +94,7 @@ class ApplicationSettings(BaseCustomSettings, MixinLoggingSettings):
         json_schema_extra={"auto_default_from_env": True},
     )
 
-    @field_validator("LOGLEVEL")
+    @field_validator("LOG_LEVEL")
     @classmethod
     def valid_log_level(cls, value) -> LogLevel:
         return LogLevel(cls.validate_log_level(value))