[PD-Disagg] Support query dp rank from bootstrap server. (#19168)

Signed-off-by: Chang Huaixin (OpenAnolis) <changhuaixin@linux.alibaba.com>
Co-authored-by: Chang Huaixin (OpenAnolis) <changhuaixin@linux.alibaba.com>
This commit is contained in:
Liangsheng Yin
2026-02-23 10:59:30 -08:00
committed by GitHub
co-authored by Chang Huaixin
parent 2cdde5d4ab
commit 2274bfebb1
9 changed files with 321 additions and 139 deletions
@@ -20,6 +20,7 @@ register_cuda_ci(est_time=155, suite="stage-c-test-8-gpu-h20")
class TestDisaggregationDPAttention(PDDisaggregationServerBase):
PREFILL_DP_SIZE = 4
DECODE_DP_SIZE = 4
LOAD_BALANCE_METHOD = "auto"
@classmethod
def setUpClass(cls):
@@ -50,6 +51,8 @@ class TestDisaggregationDPAttention(PDDisaggregationServerBase):
"--dp",
str(cls.PREFILL_DP_SIZE),
"--enable-dp-attention",
"--load-balance-method",
cls.LOAD_BALANCE_METHOD,
]
prefill_args += cls.transfer_backend + cls.rdma_devices
cls.process_prefill = popen_launch_pd_server(
@@ -72,6 +75,8 @@ class TestDisaggregationDPAttention(PDDisaggregationServerBase):
"--enable-dp-attention",
"--base-gpu-id",
str(cls.PREFILL_DP_SIZE),
"--load-balance-method",
cls.LOAD_BALANCE_METHOD,
]
decode_args += cls.transfer_backend + cls.rdma_devices
cls.process_decode = popen_launch_pd_server(
@@ -97,5 +102,9 @@ class TestDisaggregationDPAttention(PDDisaggregationServerBase):
self.assertGreater(metrics["accuracy"], 0.60)
class TestDisaggregationDPAttentionRoundRobin(TestDisaggregationDPAttention):
LOAD_BALANCE_METHOD = "round_robin"
if __name__ == "__main__":
unittest.main()