diff --git a/python/sglang/multimodal_gen/test/server/test_server_common.py b/python/sglang/multimodal_gen/test/server/test_server_common.py index ed8dc9d36e..d9423b99fc 100644 --- a/python/sglang/multimodal_gen/test/server/test_server_common.py +++ b/python/sglang/multimodal_gen/test/server/test_server_common.py @@ -390,12 +390,17 @@ class DiffusionServerBase: def _validate_and_record( self, + ctx: ServerContext, case: DiffusionTestCase, perf_record: RequestPerfRecord, request_index: int = 1, - load_time_ms: float | None = None, ) -> None: - """Validate metrics and record results.""" + """Validate metrics and record results. + + The load duration is read off ``ctx`` rather than passed separately: + recording a result requires it, and a caller that has the server always + has the measurement. + """ if perf_record is None: raise PerformanceValidationError( f"[performance] {case.id}: request performance record is missing" @@ -426,7 +431,7 @@ class DiffusionServerBase: ) summary = validator.collect_metrics(perf_record) - summary.load_time_ms = load_time_ms + summary.load_time_ms = ctx.load_time_ms self._record_performance_result(case, summary, request_index) self._print_performance_log(case, summary, scenario) @@ -1727,7 +1732,7 @@ Pinned revision used by this check: {SGL_TEST_FILES_CI_DATA_REVISION} run_case_check( "performance", lambda: self._validate_and_record( - case, perf_record, request_index, diffusion_server.load_time_ms + diffusion_server, case, perf_record, request_index ), ) diff --git a/python/sglang/multimodal_gen/test/unit/test_performance_failure_policy.py b/python/sglang/multimodal_gen/test/unit/test_performance_failure_policy.py index 10211180e5..b5b7d438aa 100644 --- a/python/sglang/multimodal_gen/test/unit/test_performance_failure_policy.py +++ b/python/sglang/multimodal_gen/test/unit/test_performance_failure_policy.py @@ -3,6 +3,7 @@ import subprocess import sys import textwrap import time +from types import SimpleNamespace import pytest @@ -80,7 +81,7 @@ def test_e2e_only_does_not_require_stage_metrics(monkeypatch, generate_baseline) steps=[], total_duration_ms=2000 if generate_baseline else 1000, ) - server._validate_and_record(case, record, load_time_ms=100) + server._validate_and_record(SimpleNamespace(load_time_ms=100), case, record) assert len(server._perf_results) == 1 assert bool(common._PENDING_BASELINE_DUMPS) == generate_baseline diff --git a/test/registered/amd/test_zimage_turbo.py b/test/registered/amd/test_zimage_turbo.py index 7ccf32f875..332fb7e566 100644 --- a/test/registered/amd/test_zimage_turbo.py +++ b/test/registered/amd/test_zimage_turbo.py @@ -125,7 +125,7 @@ class TestZImageTurboAMD(DiffusionServerBase): diffusion_server, case.id, generate_fn ) - self._validate_and_record(case, perf_record) + self._validate_and_record(diffusion_server, case, perf_record) self._test_v1_models_endpoint(diffusion_server, case) prompt = case.sampling_params.prompt or "" diff --git a/test/registered/xpu/test_xpu_flux2_dev.py b/test/registered/xpu/test_xpu_flux2_dev.py index f8d2f0ff7e..93d95a8c06 100644 --- a/test/registered/xpu/test_xpu_flux2_dev.py +++ b/test/registered/xpu/test_xpu_flux2_dev.py @@ -164,7 +164,7 @@ class TestFlux2DevXPU(DiffusionServerBase): diffusion_server, case.id, generate_fn ) - self._validate_and_record(case, perf_record) + self._validate_and_record(diffusion_server, case, perf_record) self._test_v1_models_endpoint(diffusion_server, case) prompt = case.sampling_params.prompt or "" diff --git a/test/registered/xpu/test_xpu_zimage_turbo.py b/test/registered/xpu/test_xpu_zimage_turbo.py index 128d98028f..4e4f737dac 100644 --- a/test/registered/xpu/test_xpu_zimage_turbo.py +++ b/test/registered/xpu/test_xpu_zimage_turbo.py @@ -156,7 +156,7 @@ class TestZImageTurboXPU(DiffusionServerBase): diffusion_server, case.id, generate_fn ) - self._validate_and_record(case, perf_record) + self._validate_and_record(diffusion_server, case, perf_record) self._test_v1_models_endpoint(diffusion_server, case) prompt = case.sampling_params.prompt or ""