Fix the chunked-prefix-cache gate writing config the backends never read (#33168)

The load-time gate (maybe_disable_chunked_prefix_cache) wrote its
ServerArgs instance while every reader has moved to the published
config: the attention backends assert / branch on
get_schedule().disable_chunked_prefix_cache when they initialize, so the
flip never reached them and a backend outside
CHUNKED_PREFIX_CACHE_SUPPORTED_ATTENTION_BACKENDS kept chunked prefix
enabled.

Reroute the writer through get_context().override (which writes the
published bags) and flip the two remaining instance reads — the gate's
own log check and the prefill cuda-graph runner's capture flag — to the
bag. A regression test pins the three contracts: the gate lands on the
bag, the pristine ServerArgs instance stays untouched, and the
draft-worker guard never writes.

The ServerArgs.override call-site ratchet drops 39 -> 38.

test/registered/unit/model_executor/test_chunked_prefix_cache_gate.py

@@ -0,0 +1,72 @@
+"""The load-time chunked-prefix gate must land on the published config bag.
+
+Regression: the gate wrote the ServerArgs instance while every attention
+backend reads ``get_schedule().disable_chunked_prefix_cache`` — the bag never
+saw the flip, so an unsupported backend kept chunked prefix enabled.
+"""
+
+from sglang.test.ci.ci_register import register_cpu_ci
+
+register_cpu_ci(est_time=4, suite="base-a-test-cpu")
+
+import unittest
+
+from sglang.srt.model_executor.model_runner_components.misc_utils import (
+    maybe_disable_chunked_prefix_cache,
+)
+from sglang.srt.runtime_context import get_context, get_schedule, get_server_args
+from sglang.test.test_utils import CustomTestCase
+
+
+class TestChunkedPrefixCacheGate(CustomTestCase):
+    def _seed(self, **fields):
+        override = get_context().override_server_args(**fields)
+        override.install()
+        self.addCleanup(override.restore)
+
+    def test_unsupported_backend_disables_on_the_bag(self):
+        self._seed(attention_backend="triton")
+        maybe_disable_chunked_prefix_cache(use_mla_backend=True, is_draft_worker=False)
+        self.assertTrue(get_schedule().disable_chunked_prefix_cache)
+        self.assertFalse(get_server_args().disable_chunked_prefix_cache)
+
+    def test_supported_backend_keeps_chunked_prefix(self):
+        self._seed(attention_backend="fa3")
+        maybe_disable_chunked_prefix_cache(use_mla_backend=True, is_draft_worker=False)
+        self.assertFalse(get_schedule().disable_chunked_prefix_cache)
+
+    def test_draft_worker_never_writes(self):
+        self._seed(attention_backend="triton")
+        maybe_disable_chunked_prefix_cache(use_mla_backend=False, is_draft_worker=True)
+        self.assertFalse(get_schedule().disable_chunked_prefix_cache)
+
+    def test_republish_discards_the_gate_so_it_must_run_after_publish(self):
+        # Pins the ordering contract in ModelRunner.__init__: publishing
+        # rebuilds the bags from the pristine instance, so the gate runs after
+        # the target-worker publish.
+        self._seed(attention_backend="triton")
+        sa = get_server_args()
+        maybe_disable_chunked_prefix_cache(use_mla_backend=True, is_draft_worker=False)
+        self.assertTrue(get_schedule().disable_chunked_prefix_cache)
+        get_context().set_server_args(sa)  # what a later republish would do
+        self.assertFalse(get_schedule().disable_chunked_prefix_cache)
+
+    def test_draft_copy_overrides_carry_the_gate(self):
+        # The draft copy comes from the pristine instance, which never sees
+        # the bag-only gate; the copy's pre-publish overrides carry it.
+        from types import SimpleNamespace
+
+        from sglang.srt.speculative.draft_worker_common import (
+            draft_server_args_overrides,
+        )
+
+        self._seed(attention_backend="triton")
+        maybe_disable_chunked_prefix_cache(use_mla_backend=True, is_draft_worker=False)
+        fields = draft_server_args_overrides(
+            SimpleNamespace(context_len=64), draft_backend="fa3"
+        )
+        self.assertTrue(fields["disable_chunked_prefix_cache"])
+
+
+if __name__ == "__main__":
+    unittest.main()

test/registered/unit/test_server_args_writer_ratchet.py

@@ -49,7 +49,7 @@ _EXCLUDED = (
     "multimodal_gen",
 )
 
-_BASELINE = 39
+_BASELINE = 38
 
 
 class TestServerArgsWriterRatchet(CustomTestCase):