[CI] Fix SGLANG_JIT_KERNEL_RUN_FULL_TESTS never activating the nightly full jit-kernel sweep (#31042)

Co-authored-by: Claude Fable 5 <noreply@anthropic.com>

.github/workflows/nightly-test-nvidia.yml

@@ -145,7 +145,9 @@ jobs:
         timeout-minutes: 90
         run: |
           cd test
-          python3 run_suite.py --hw cuda --suite nightly-kernel-8-gpu-h200 --nightly --continue-on-error
+          # Full grids run ~7x the in-CI parametrizations per world size; the
+          # default 1200s per-file budget only fits the reduced PR sweep.
+          python3 run_suite.py --hw cuda --suite nightly-kernel-8-gpu-h200 --nightly --continue-on-error --timeout-per-file 3600
 
       - uses: ./.github/actions/upload-cuda-coredumps
         if: failure()

python/sglang/jit_kernel/utils.py

@@ -24,19 +24,19 @@ from typing import (
 
 import torch
 
+from sglang.srt.environ import envs
 from sglang.utils import is_in_ci
 
 if TYPE_CHECKING:
     from tvm_ffi import Module
 
 F = TypeVar("F", bound=Callable[..., Any])
-_FULL_TEST_ENV_VAR = "SGLANG_JIT_KERNEL_RUN_FULL_TESTS"
 
 logger = logging.getLogger(__name__)
 
 
 def should_run_full_tests() -> bool:
-    return os.getenv(_FULL_TEST_ENV_VAR, "false").lower() == "true"
+    return envs.SGLANG_JIT_KERNEL_RUN_FULL_TESTS.get()
 
 
 def get_ci_test_range(full_range: List[Any], ci_range: List[Any]) -> List[Any]:

python/sglang/srt/environ.py

@@ -245,6 +245,8 @@ class Envs:
     # else /tmp); see debug_utils/cuda_coredump.py.
     SGLANG_CUDA_COREDUMP_DIR = EnvStr(None)
     SGLANG_TEST_MAX_RETRY = EnvInt(None)
+    # Expand jit_kernel test grids to their full parameter ranges (nightly).
+    SGLANG_JIT_KERNEL_RUN_FULL_TESTS = EnvBool(False)
 
     # Constrained Decoding (Grammar)
     SGLANG_GRAMMAR_POLL_INTERVAL = EnvFloat(0.005)