Method _schedule_one

gpustack/scheduler/scheduler.py:314–405

Schedule a model instance by picking one candidate. Args: item: Model instance to schedule.

(self, instance: ModelInstance)

                logger.error(f"Failed to get item from schedule queue: {e}")

    async def _schedule_one(self, instance: ModelInstance):  # noqa: C901
        """
        Schedule a model instance by picking one candidate.
        Args:
            item: Model instance to schedule.
        """
        logger.debug(f"Scheduling model instance {instance.name}")

        state_message = ""

        async with async_session() as session:
            workers = await Worker.all(session)
            if not workers:
                state_message = "No available workers"

            model = await Model.one_by_id(session, instance.model_id)
            if model is None:
                state_message = "Model not found"

            model_instance = await ModelInstance.one_by_id(session, instance.id)
            if model_instance is None:
                logger.debug(
                    f"Model instance(ID: {instance.id}) was deleted before scheduling due"
                )
                return

            model_instances = await ModelInstance.all(
                session, options=[selectinload(ModelInstance.model)]
            )

            candidate = None
            messages = []
            if workers and model:
                try:
                    candidate, messages = await find_candidate(
                        self._config, model, workers, model_instances
                    )
                except Exception as e:
                    state_message = f"Failed to find candidate: {e}"

            if candidate is None:
                # update model instance.
                if model_instance.state in (
                    ModelInstanceStateEnum.SCHEDULED,
                    ModelInstanceStateEnum.ANALYZING,
                ):
                    model_instance.state = ModelInstanceStateEnum.PENDING
                    model_instance.state_message = (
                        "No suitable workers.\nDetails:\n" + "".join(messages)
                    )
                if state_message != "":
                    model_instance.state_message = state_message

                await ModelInstanceService(session).update(model_instance)
                logger.debug(
                    f"No suitable workers for model instance {model_instance.name}, state: {model_instance.state}"
                )
            else: