import unittest
from types import SimpleNamespace

from sglang.test.ci.ci_register import register_cuda_ci
from sglang.test.few_shot_gsm8k import run_eval as run_eval_few_shot_gsm8k
from sglang.test.server_fixtures.disaggregation_fixture import (
    PDDisaggregationServerBase,
)
from sglang.test.test_utils import (
    DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
    is_in_ci,
    popen_launch_pd_server,
)

register_cuda_ci(
    est_time=400, suite="stage-c-test-8-gpu-h200", disabled="TCP fallback flaky"
)


@unittest.skipIf(is_in_ci(), "Temporarily disable the flaky test.")
class TestDisaggregationHybridAttentionMamba(PDDisaggregationServerBase):
    @classmethod
    def setUpClass(cls):
        super().setUpClass()
        cls.model = "Qwen/Qwen3-Next-80B-A3B-Instruct"

        # Non blocking start servers
        cls.start_prefill()
        cls.start_decode()

        # Block until both
        cls.wait_server_ready(cls.prefill_url + "/health", process=cls.process_prefill)
        cls.wait_server_ready(cls.decode_url + "/health", process=cls.process_decode)

        cls.launch_lb()

    @classmethod
    def start_prefill(cls):
        prefill_args = [
            "--trust-remote-code",
            "--disaggregation-mode",
            "prefill",
            "--tp",
            "4",
        ]
        prefill_args += cls.transfer_backend + cls.rdma_devices
        cls.process_prefill = popen_launch_pd_server(
            cls.model,
            cls.prefill_url,
            timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
            other_args=prefill_args,
        )

    @classmethod
    def start_decode(cls):
        decode_args = [
            "--trust-remote-code",
            "--disaggregation-mode",
            "decode",
            "--tp",
            "4",
            "--base-gpu-id",
            "4",
        ]
        decode_args += cls.transfer_backend + cls.rdma_devices
        cls.process_decode = popen_launch_pd_server(
            cls.model,
            cls.decode_url,
            timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
            other_args=decode_args,
        )

    def test_gsm8k(self):
        args = SimpleNamespace(
            num_shots=5,
            data_path=None,
            num_questions=200,
            max_new_tokens=512,
            parallel=128,
            host=f"http://{self.base_host}",
            port=int(self.lb_port),
        )
        metrics = run_eval_few_shot_gsm8k(args)
        print(f"Evaluation metrics: {metrics}")

        self.assertGreater(metrics["accuracy"], 0.93)


class TestDisaggregationHybridAttentionMambaExtraBuffer(PDDisaggregationServerBase):
    @classmethod
    def setUpClass(cls):
        super().setUpClass()
        cls.model = "Qwen/Qwen3-Next-80B-A3B-Instruct"

        # Non blocking start servers
        cls.start_prefill()
        cls.start_decode()

        # Block until both
        cls.wait_server_ready(cls.prefill_url + "/health", process=cls.process_prefill)
        cls.wait_server_ready(cls.decode_url + "/health", process=cls.process_decode)

        cls.launch_lb()

    @classmethod
    def start_prefill(cls):
        prefill_args = [
            "--trust-remote-code",
            "--disaggregation-mode",
            "prefill",
            "--tp",
            "4",
            "--mamba-scheduler-strategy",
            "extra_buffer",
        ]
        prefill_args += cls.transfer_backend + cls.rdma_devices
        cls.process_prefill = popen_launch_pd_server(
            cls.model,
            cls.prefill_url,
            timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
            other_args=prefill_args,
        )

    @classmethod
    def start_decode(cls):
        decode_args = [
            "--trust-remote-code",
            "--disaggregation-mode",
            "decode",
            "--tp",
            "4",
            "--base-gpu-id",
            "4",
            "--mamba-scheduler-strategy",
            "extra_buffer",
        ]
        decode_args += cls.transfer_backend + cls.rdma_devices
        cls.process_decode = popen_launch_pd_server(
            cls.model,
            cls.decode_url,
            timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
            other_args=decode_args,
        )

    def test_gsm8k(self):
        args = SimpleNamespace(
            num_shots=5,
            data_path=None,
            num_questions=200,
            max_new_tokens=512,
            parallel=128,
            host=f"http://{self.base_host}",
            port=int(self.lb_port),
        )
        metrics = run_eval_few_shot_gsm8k(args)
        print(f"Evaluation metrics: {metrics}")

        self.assertGreater(metrics["accuracy"], 0.93)


@unittest.skipIf(
    is_in_ci(),
    "Temporarily disable the flaky test: tcp fallback is not stable currently.",
)
class TestDisaggregationHybridAttentionMambaDPDecode(PDDisaggregationServerBase):
    """Test with prefill tp=2 and decode tp=2/dp=2 with dp-attention enabled."""

    @classmethod
    def setUpClass(cls):
        super().setUpClass()
        cls.model = "Qwen/Qwen3-Next-80B-A3B-Instruct"

        # Non blocking start servers
        cls.start_prefill()
        cls.start_decode()

        # Block until both
        cls.wait_server_ready(cls.prefill_url + "/health", process=cls.process_prefill)
        cls.wait_server_ready(cls.decode_url + "/health", process=cls.process_decode)

        cls.launch_lb()

    @classmethod
    def start_prefill(cls):
        prefill_args = [
            "--trust-remote-code",
            "--disaggregation-mode",
            "prefill",
            "--tp",
            "2",
        ]
        prefill_args += cls.transfer_backend + cls.rdma_devices
        cls.process_prefill = popen_launch_pd_server(
            cls.model,
            cls.prefill_url,
            timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
            other_args=prefill_args,
        )

    @classmethod
    def start_decode(cls):
        decode_args = [
            "--trust-remote-code",
            "--disaggregation-mode",
            "decode",
            "--tp",
            "2",
            "--dp",
            "2",
            "--enable-dp-attention",
            "--enable-dp-lm-head",
            "--base-gpu-id",
            "2",
        ]
        decode_args += cls.transfer_backend + cls.rdma_devices
        cls.process_decode = popen_launch_pd_server(
            cls.model,
            cls.decode_url,
            timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
            other_args=decode_args,
        )

    def test_gsm8k(self):
        args = SimpleNamespace(
            num_shots=5,
            data_path=None,
            num_questions=200,
            max_new_tokens=512,
            parallel=128,
            host=f"http://{self.base_host}",
            port=int(self.lb_port),
        )
        metrics = run_eval_few_shot_gsm8k(args)
        print(f"Evaluation metrics: {metrics}")

        self.assertGreater(metrics["accuracy"], 0.93)


if __name__ == "__main__":
    unittest.main()