Merge remote-tracking branch 'upstream/pydantic_v2_migration_do_not_squash_updates' into pr-osparc-add-dybamic-services-monitor-dashboard2

Author: Andrei Neagu

ci/github/system-testing/e2e-playwright.bash

@@ -24,6 +24,7 @@ test() {
   source .venv/bin/activate
   pushd tests/e2e-playwright
   make test-sleepers
+  make test-platform
   popd
 }
 

packages/common-library/src/common_library/basic_types.py

@@ -0,0 +1,38 @@
+from enum import StrEnum
+
+
+class LogLevel(StrEnum):
+    DEBUG = "DEBUG"
+    INFO = "INFO"
+    WARNING = "WARNING"
+    ERROR = "ERROR"
+
+
+class BootModeEnum(StrEnum):
+    """
+    Values taken by SC_BOOT_MODE environment variable
+    set in Dockerfile and used during docker/boot.sh
+    """
+
+    DEFAULT = "default"
+    LOCAL = "local-development"
+    DEBUG = "debug"
+    PRODUCTION = "production"
+    DEVELOPMENT = "development"
+
+    def is_devel_mode(self) -> bool:
+        """returns True if this boot mode is used for development"""
+        return self in (self.DEBUG, self.DEVELOPMENT, self.LOCAL)
+
+
+class BuildTargetEnum(StrEnum):
+    """
+    Values taken by SC_BUILD_TARGET environment variable
+    set in Dockerfile that defines the stage targeted in the
+    docker image build
+    """
+
+    BUILD = "build"
+    CACHE = "cache"
+    PRODUCTION = "production"
+    DEVELOPMENT = "development"

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

@@ -1,8 +1,9 @@
 from decimal import Decimal
-from enum import StrEnum
+from enum import Enum
 from re import Pattern
 from typing import Annotated, ClassVar, Final, TypeAlias
 
+from common_library.basic_types import BootModeEnum, BuildTargetEnum, LogLevel
 from pydantic import Field, HttpUrl, PositiveInt, StringConstraints
 from pydantic_core import core_schema
 
@@ -13,6 +14,17 @@
     UUID_RE,
 )
 
+assert issubclass(LogLevel, Enum)  # nosec
+assert issubclass(BootModeEnum, Enum)  # nosec
+assert issubclass(BuildTargetEnum, Enum)  # nosec
+
+__all__: tuple[str, ...] = (
+    "LogLevel",
+    "BootModeEnum",
+    "BuildTargetEnum",
+)
+
+
 NonNegativeDecimal: TypeAlias = Annotated[Decimal, Field(ge=0)]
 
 PositiveDecimal: TypeAlias = Annotated[Decimal, Field(gt=0)]
@@ -145,41 +157,4 @@ class HttpUrlWithCustomMinLength(HttpUrl):
     min_length = 0
 
 
-class LogLevel(StrEnum):
-    DEBUG = "DEBUG"
-    INFO = "INFO"
-    WARNING = "WARNING"
-    ERROR = "ERROR"
-
-
-class BootModeEnum(StrEnum):
-    """
-    Values taken by SC_BOOT_MODE environment variable
-    set in Dockerfile and used during docker/boot.sh
-    """
-
-    DEFAULT = "default"
-    LOCAL = "local-development"
-    DEBUG = "debug"
-    PRODUCTION = "production"
-    DEVELOPMENT = "development"
-
-    def is_devel_mode(self) -> bool:
-        """returns True if this boot mode is used for development"""
-        return self in (self.DEBUG, self.DEVELOPMENT, self.LOCAL)
-
-
-class BuildTargetEnum(StrEnum):
-    """
-    Values taken by SC_BUILD_TARGET environment variable
-    set in Dockerfile that defines the stage targeted in the
-    docker image build
-    """
-
-    BUILD = "build"
-    CACHE = "cache"
-    PRODUCTION = "production"
-    DEVELOPMENT = "development"
-
-
 KeyIDStr = Annotated[str, StringConstraints(pattern=PROPERTY_KEY_RE)]

packages/pytest-simcore/src/pytest_simcore/helpers/playwright.py

@@ -2,19 +2,25 @@
 import json
 import logging
 import re
+import typing
 from collections import defaultdict
-from collections.abc import Generator, Iterator
+from collections.abc import Generator
 from dataclasses import dataclass, field
 from datetime import UTC, datetime, timedelta
 from enum import Enum, unique
 from typing import Any, Final
 
 import httpx
+from playwright._impl._sync_base import EventContextManager
 from playwright.sync_api import FrameLocator, Page, Request
 from playwright.sync_api import TimeoutError as PlaywrightTimeoutError
 from playwright.sync_api import WebSocket
 from pydantic import AnyUrl
-from pytest_simcore.helpers.logging_tools import log_context
+
+from .logging_tools import log_context
+
+_logger = logging.getLogger(__name__)
+
 
 SECOND: Final[int] = 1000
 MINUTE: Final[int] = 60 * SECOND
@@ -106,6 +112,94 @@ class SocketIOEvent:
 SOCKETIO_MESSAGE_PREFIX: Final[str] = "42"
 
 
+@dataclass
+class RestartableWebSocket:
+    page: Page
+    ws: WebSocket
+    _registered_events: list[tuple[str, typing.Callable | None]] = field(
+        default_factory=list
+    )
+    _number_of_restarts: int = 0
+
+    def __post_init__(self):
+        self._configure_websocket_events()
+
+    def _configure_websocket_events(self):
+        try:
+            with log_context(
+                logging.DEBUG,
+                msg="handle websocket message (set to --log-cli-level=DEBUG level if you wanna see all of them)",
+            ) as ctx:
+
+                def on_framesent(payload: str | bytes) -> None:
+                    ctx.logger.debug("⬇️ Frame sent: %s", payload)
+
+                def on_framereceived(payload: str | bytes) -> None:
+                    ctx.logger.debug("⬆️ Frame received: %s", payload)
+
+                def on_close(_: WebSocket) -> None:
+                    ctx.logger.warning(
+                        "⚠️ WebSocket closed. Attempting to reconnect..."
+                    )
+                    self._attempt_reconnect(ctx.logger)
+
+                def on_socketerror(error_msg: str) -> None:
+                    ctx.logger.error("❌ WebSocket error: %s", error_msg)
+
+                # Attach core event listeners
+                self.ws.on("framesent", on_framesent)
+                self.ws.on("framereceived", on_framereceived)
+                self.ws.on("close", on_close)
+                self.ws.on("socketerror", on_socketerror)
+
+        finally:
+            # Detach core event listeners
+            self.ws.remove_listener("framesent", on_framesent)
+            self.ws.remove_listener("framereceived", on_framereceived)
+            self.ws.remove_listener("close", on_close)
+            self.ws.remove_listener("socketerror", on_socketerror)
+
+    def _attempt_reconnect(self, logger: logging.Logger) -> None:
+        """
+        Attempt to reconnect the WebSocket and restore event listeners.
+        """
+        try:
+            with self.page.expect_websocket() as ws_info:
+                assert not ws_info.value.is_closed()
+
+            self.ws = ws_info.value
+            self._number_of_restarts += 1
+            logger.info(
+                "🔄 Reconnected to WebSocket successfully. Number of reconnections: %s",
+                self._number_of_restarts,
+            )
+            self._configure_websocket_events()
+            # Re-register all custom event listeners
+            for event, predicate in self._registered_events:
+                self.ws.expect_event(event, predicate)
+
+        except Exception as e:  # pylint: disable=broad-except
+            logger.error("🚨 Failed to reconnect WebSocket: %s", e)
+
+    def expect_event(
+        self,
+        event: str,
+        predicate: typing.Callable | None = None,
+        *,
+        timeout: float | None = None,
+    ) -> EventContextManager:
+        """
+        Register an event listener with support for reconnection.
+        """
+        output = self.ws.expect_event(event, predicate, timeout=timeout)
+        self._registered_events.append((event, predicate))
+        return output
+
+    @classmethod
+    def create(cls, page: Page, ws: WebSocket):
+        return cls(page, ws)
+
+
 def decode_socketio_42_message(message: str) -> SocketIOEvent:
     data = json.loads(message.removeprefix(SOCKETIO_MESSAGE_PREFIX))
     return SocketIOEvent(name=data[0], obj=data[1])
@@ -245,9 +339,14 @@ def __call__(self, message: str) -> bool:
             url = f"https://{self.node_id}.services.{self.get_partial_product_url()}"
             response = httpx.get(url, timeout=10)
             self.logger.info(
-                "Querying the service endpoint from the E2E test. Url: %s Response: %s",
+                "Querying the service endpoint from the E2E test. Url: %s Response: %s TIP: %s",
                 url,
                 response,
+                (
+                    "Response 401 is OK. It means that service is ready."
+                    if response.status_code == 401
+                    else "We are emulating the frontend; a 500 response is acceptable if the service is not yet ready."
+                ),
             )
             if response.status_code <= 401:
                 # NOTE: If the response status is less than 400, it means that the backend is ready (There are some services that respond with a 3XX)
@@ -278,7 +377,7 @@ def get_partial_product_url(self):
 def wait_for_pipeline_state(
     current_state: RunningState,
     *,
-    websocket: WebSocket,
+    websocket: RestartableWebSocket,
     if_in_states: tuple[RunningState, ...],
     expected_states: tuple[RunningState, ...],
     timeout_ms: int,
@@ -301,39 +400,6 @@ def wait_for_pipeline_state(
     return current_state
 
 
-@contextlib.contextmanager
-def web_socket_default_log_handler(web_socket: WebSocket) -> Iterator[None]:
-
-    try:
-        with log_context(
-            logging.DEBUG,
-            msg="handle websocket message (set to --log-cli-level=DEBUG level if you wanna see all of them)",
-        ) as ctx:
-
-            def on_framesent(payload: str | bytes) -> None:
-                ctx.logger.debug("⬇️ Frame sent: %s", payload)
-
-            def on_framereceived(payload: str | bytes) -> None:
-                ctx.logger.debug("⬆️ Frame received: %s", payload)
-
-            def on_close(payload: WebSocket) -> None:
-                ctx.logger.warning("⚠️ Websocket closed: %s", payload)
-
-            def on_socketerror(error_msg: str) -> None:
-                ctx.logger.error("❌ Websocket error: %s", error_msg)
-
-            web_socket.on("framesent", on_framesent)
-            web_socket.on("framereceived", on_framereceived)
-            web_socket.on("close", on_close)
-            web_socket.on("socketerror", on_socketerror)
-            yield
-    finally:
-        web_socket.remove_listener("framesent", on_framesent)
-        web_socket.remove_listener("framereceived", on_framereceived)
-        web_socket.remove_listener("close", on_close)
-        web_socket.remove_listener("socketerror", on_socketerror)
-
-
 def _node_started_predicate(request: Request) -> bool:
     return bool(
         re.search(NODE_START_REQUEST_PATTERN, request.url)
@@ -358,12 +424,14 @@ def expected_service_running(
     *,
     page: Page,
     node_id: str,
-    websocket: WebSocket,
+    websocket: RestartableWebSocket,
     timeout: int,
     press_start_button: bool,
     product_url: AnyUrl,
 ) -> Generator[ServiceRunning, None, None]:
-    with log_context(logging.INFO, msg="Waiting for node to run") as ctx:
+    with log_context(
+        logging.INFO, msg=f"Waiting for node to run. Timeout: {timeout}"
+    ) as ctx:
         waiter = SocketIONodeProgressCompleteWaiter(
             node_id=node_id, logger=ctx.logger, product_url=product_url
         )
@@ -395,15 +463,17 @@ def wait_for_service_running(
     *,
     page: Page,
     node_id: str,
-    websocket: WebSocket,
+    websocket: RestartableWebSocket,
     timeout: int,
     press_start_button: bool,
     product_url: AnyUrl,
 ) -> FrameLocator:
     """NOTE: if the service was already started this will not work as some of the required websocket events will not be emitted again
     In which case this will need further adjutment"""
 
-    with log_context(logging.INFO, msg="Waiting for node to run") as ctx:
+    with log_context(
+        logging.INFO, msg=f"Waiting for node to run. Timeout: {timeout}"
+    ) as ctx:
         waiter = SocketIONodeProgressCompleteWaiter(
             node_id=node_id, logger=ctx.logger, product_url=product_url
         )

packages/pytest-simcore/src/pytest_simcore/helpers/playwright_sim4life.py

@@ -13,6 +13,7 @@
     MINUTE,
     SECOND,
     SOCKETIO_MESSAGE_PREFIX,
+    RestartableWebSocket,
     SocketIOEvent,
     decode_socketio_42_message,
     wait_for_service_running,
@@ -100,7 +101,7 @@ class WaitForS4LDict(TypedDict):
 def wait_for_launched_s4l(
     page: Page,
     node_id,
-    log_in_and_out: WebSocket,
+    log_in_and_out: RestartableWebSocket,
     *,
     autoscaled: bool,
     copy_workspace: bool,

packages/settings-library/src/settings_library/application.py

@@ -1,8 +1,7 @@
-from models_library.basic_types import BootModeEnum
 from pydantic import Field, PositiveInt
 
 from .base import BaseCustomSettings
-from .basic_types import BuildTargetEnum
+from .basic_types import BootModeEnum, BuildTargetEnum
 
 
 class BaseApplicationSettings(BaseCustomSettings):

packages/settings-library/src/settings_library/base.py

@@ -58,7 +58,7 @@ def _is_auto_default_from_env_enabled(field: FieldInfo) -> bool:
     )
 
 
-ENABLED: Final = {}
+_MARKED_AS_UNSET: Final[dict] = {}
 
 
 class EnvSettingsWithAutoDefaultSource(EnvSettingsSource):
@@ -89,7 +89,7 @@ def prepare_field_value(
             _is_auto_default_from_env_enabled(field)
             and field.default_factory
             and field.default is None
-            and prepared_value == ENABLED
+            and prepared_value == _MARKED_AS_UNSET
         ):
             prepared_value = field.default_factory()
         return prepared_value