config: route DCP topology reads through get_parallel() (#33925)

Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>

test/registered/dcp/test_dcp_layout_unit.py

@@ -162,10 +162,12 @@ class TestGetDcpLens(CustomTestCase):
                 device="cpu",
                 is_draft_worker=False,
             )
+            # The allocator widens from get_parallel(), not from the injected
+            # server_args stand-in -- drive the cause, not the effect.
             with patch(
                 "sglang.srt.mem_cache.kv_cache_configurator.current_platform.is_out_of_tree",
                 return_value=False,
-            ):
+            ), rc.get_parallel().override(attn_dcp_size=dcp_size):
                 allocators[dcp_size] = (
                     KVCacheConfigurator._build_token_to_kv_pool_allocator(
                         configurator,

test/registered/unit/disaggregation/test_disaggregation_wire.py

@@ -200,7 +200,6 @@ class TestEagleDsaSeedTransfer(unittest.TestCase):
             enable_multi_layer_eagle=False,
             disaggregation_mode="decode",
             enable_hisparse=False,
-            dcp_size=1,
         )
 
         with envs.SGLANG_DSA_FUSE_TOPK.override(True), patch(

test/registered/unit/managers/test_scheduler_init_req_max_new_tokens.py

@@ -46,7 +46,6 @@ class TestSchedulerInitReqMaxNewTokens(unittest.TestCase):
         scheduler.max_req_len = max_req_len
         scheduler.max_total_num_tokens = max_total_num_tokens
         scheduler.page_size = page_size
-        scheduler.server_args = SimpleNamespace(dcp_size=1)
         scheduler.max_new_tokens_limit = envs.SGLANG_MAX_NEW_TOKENS_LIMIT.get()
         return scheduler
 

test/registered/unit/test_global_config_read_ratchet.py

@@ -27,8 +27,6 @@ What legitimately remains:
     the short circuit is the point: with PP off the group is never touched, which
     is what lets the ``Indexer`` be constructed before distributed init. The live
     property would demand the group either way.
-  - ``allocation.dcp_size`` asks whether DCP was *configured*; the live property
-    reads ``get_dcp_group()``, and that group is only installed when DCP is on.
   - ``cuda_ipc_transport_utils.tp_size`` runs in the tokenizer process, which has
     no groups at all (the call site already guards for "not published yet").
   - ``dp_attention.attn_cp_size`` / ``moe_dp_size``: the configuration the
@@ -75,7 +73,6 @@ _CONFIG_INTENT_SIZES = frozenset(
         ("srt/layers/attention/dsa/dsa_indexer.py", "pp_size"),
         ("srt/layers/dp_attention.py", "attn_cp_size"),
         ("srt/layers/dp_attention.py", "moe_dp_size"),
-        ("srt/mem_cache/allocation.py", "dcp_size"),
         ("srt/model_loader/loader.py", "moe_dp_size"),
         ("srt/utils/cuda_ipc_transport_utils.py", "tp_size"),
     }