chenchenghao/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 12 Packages Projects Releases Wiki Activity

[CI] feat: add early exit to wait_for_server when process dies (#18602)

Browse Source
This commit is contained in:
shuwenn
2026-02-13 16:46:09 -08:00
committed by GitHub
parent eccf875d49
commit 3299c4f9c1
33 changed files with 229 additions and 179 deletions
Download Patch File Download Diff File Expand all files Collapse all files
test/registered/distributed/test_disaggregation_aarch64.py
+2 -2
View File
@@ -29,8 +29,8 @@ class TestDisaggregationMooncakeAARCH64Accuracy(PDDisaggregationServerBase):
cls.start_decode()
# Block until both
cls.wait_server_ready(cls.prefill_url + "/health")
cls.wait_server_ready(cls.decode_url + "/health")
cls.wait_server_ready(cls.prefill_url + "/health", process=cls.process_prefill)
cls.wait_server_ready(cls.decode_url + "/health", process=cls.process_decode)
cls.launch_lb()
test/registered/distributed/test_disaggregation_different_tp.py
+8 -8
View File
@@ -32,8 +32,8 @@ class TestDisaggregationMooncakePrefillLargerTP(PDDisaggregationServerBase):
cls.start_decode()
# Block until both
cls.wait_server_ready(cls.prefill_url + "/health")
cls.wait_server_ready(cls.decode_url + "/health")
cls.wait_server_ready(cls.prefill_url + "/health", process=cls.process_prefill)
cls.wait_server_ready(cls.decode_url + "/health", process=cls.process_decode)
cls.launch_lb()
@@ -103,8 +103,8 @@ class TestDisaggregationMooncakeDecodeLargerTP(PDDisaggregationServerBase):
cls.start_decode()
# Block until both
cls.wait_server_ready(cls.prefill_url + "/health")
cls.wait_server_ready(cls.decode_url + "/health")
cls.wait_server_ready(cls.prefill_url + "/health", process=cls.process_prefill)
cls.wait_server_ready(cls.decode_url + "/health", process=cls.process_decode)
cls.launch_lb()
@@ -174,8 +174,8 @@ class TestDisaggregationMooncakeMHAPrefillLargerTP(PDDisaggregationServerBase):
cls.start_decode()
# Block until both
cls.wait_server_ready(cls.prefill_url + "/health")
cls.wait_server_ready(cls.decode_url + "/health")
cls.wait_server_ready(cls.prefill_url + "/health", process=cls.process_prefill)
cls.wait_server_ready(cls.decode_url + "/health", process=cls.process_decode)
cls.launch_lb()
@@ -245,8 +245,8 @@ class TestDisaggregationMooncakeMHADecodeLargerTP(PDDisaggregationServerBase):
cls.start_decode()
# Block until both
cls.wait_server_ready(cls.prefill_url + "/health")
cls.wait_server_ready(cls.decode_url + "/health")
cls.wait_server_ready(cls.prefill_url + "/health", process=cls.process_prefill)
cls.wait_server_ready(cls.decode_url + "/health", process=cls.process_decode)
cls.launch_lb()
test/registered/distributed/test_disaggregation_dp_attention.py
+2 -2
View File
@@ -34,8 +34,8 @@ class TestDisaggregationDPAttention(PDDisaggregationServerBase):
cls.start_decode()
# Block until both
cls.wait_server_ready(cls.prefill_url + "/health")
cls.wait_server_ready(cls.decode_url + "/health")
cls.wait_server_ready(cls.prefill_url + "/health", process=cls.process_prefill)
cls.wait_server_ready(cls.decode_url + "/health", process=cls.process_decode)
cls.launch_lb()
test/registered/distributed/test_disaggregation_hybrid_attention.py
+6 -6
View File
@@ -29,8 +29,8 @@ class TestDisaggregationHybridAttentionMamba(PDDisaggregationServerBase):
cls.start_decode()
# Block until both
cls.wait_server_ready(cls.prefill_url + "/health")
cls.wait_server_ready(cls.decode_url + "/health")
cls.wait_server_ready(cls.prefill_url + "/health", process=cls.process_prefill)
cls.wait_server_ready(cls.decode_url + "/health", process=cls.process_decode)
cls.launch_lb()
@@ -97,8 +97,8 @@ class TestDisaggregationHybridAttentionMambaExtraBuffer(PDDisaggregationServerBa
cls.start_decode()
# Block until both
cls.wait_server_ready(cls.prefill_url + "/health")
cls.wait_server_ready(cls.decode_url + "/health")
cls.wait_server_ready(cls.prefill_url + "/health", process=cls.process_prefill)
cls.wait_server_ready(cls.decode_url + "/health", process=cls.process_decode)
cls.launch_lb()
@@ -175,8 +175,8 @@ class TestDisaggregationHybridAttentionMambaDPDecode(PDDisaggregationServerBase)
cls.start_decode()
# Block until both
cls.wait_server_ready(cls.prefill_url + "/health")
cls.wait_server_ready(cls.decode_url + "/health")
cls.wait_server_ready(cls.prefill_url + "/health", process=cls.process_prefill)
cls.wait_server_ready(cls.decode_url + "/health", process=cls.process_decode)
cls.launch_lb()
test/registered/distributed/test_disaggregation_pp.py
+6 -6
View File
@@ -28,8 +28,8 @@ class TestDisaggregationPrefillPPAccuracy(PDDisaggregationServerBase):
cls.start_decode()
# Block until both
cls.wait_server_ready(cls.prefill_url + "/health")
cls.wait_server_ready(cls.decode_url + "/health")
cls.wait_server_ready(cls.prefill_url + "/health", process=cls.process_prefill)
cls.wait_server_ready(cls.decode_url + "/health", process=cls.process_decode)
cls.launch_lb()
@@ -101,8 +101,8 @@ class TestDisaggregationPrefillPPDynamicChunkAccuracy(PDDisaggregationServerBase
cls.start_decode()
# Block until both
cls.wait_server_ready(cls.prefill_url + "/health")
cls.wait_server_ready(cls.decode_url + "/health")
cls.wait_server_ready(cls.prefill_url + "/health", process=cls.process_prefill)
cls.wait_server_ready(cls.decode_url + "/health", process=cls.process_decode)
cls.launch_lb()
@@ -175,8 +175,8 @@ class TestDisaggregationDecodePPAccuracy(PDDisaggregationServerBase):
cls.start_decode()
# Block until both
cls.wait_server_ready(cls.prefill_url + "/health")
cls.wait_server_ready(cls.decode_url + "/health")
cls.wait_server_ready(cls.prefill_url + "/health", process=cls.process_prefill)
cls.wait_server_ready(cls.decode_url + "/health", process=cls.process_decode)
cls.launch_lb()
test/registered/distributed/test_epd_disaggregation.py
+7 -7
View File
@@ -44,9 +44,9 @@ class TestEPDDisaggregationOneEncoder(PDDisaggregationServerBase):
decode_thread.join()
# Wait for all servers to be ready
cls.wait_server_ready(cls.encode_url + "/health")
cls.wait_server_ready(cls.prefill_url + "/health")
cls.wait_server_ready(cls.decode_url + "/health")
cls.wait_server_ready(cls.encode_url + "/health", process=cls.process_encode)
cls.wait_server_ready(cls.prefill_url + "/health", process=cls.process_prefill)
cls.wait_server_ready(cls.decode_url + "/health", process=cls.process_decode)
cls.launch_lb()
@@ -249,10 +249,10 @@ class TestEPDDisaggregationMultiEncoders(PDDisaggregationServerBase):
prefill_thread.join()
decode_thread.join()
cls.wait_server_ready(cls.encode_url1 + "/health")
cls.wait_server_ready(cls.encode_url2 + "/health")
cls.wait_server_ready(cls.prefill_url + "/health")
cls.wait_server_ready(cls.decode_url + "/health")
cls.wait_server_ready(cls.encode_url1 + "/health", process=cls.process_encode1)
cls.wait_server_ready(cls.encode_url2 + "/health", process=cls.process_encode2)
cls.wait_server_ready(cls.prefill_url + "/health", process=cls.process_prefill)
cls.wait_server_ready(cls.decode_url + "/health", process=cls.process_decode)
cls.launch_lb()