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

Deprecate the parallel getters the context answers, and ratchet them shut (#40342)

Browse Source
This commit is contained in:
Cheng Wan
2026-09-21 12:25:32 -07:00
committed by GitHub
parent 65be3fa71a
commit 73f071db52
44 changed files with 933 additions and 501 deletions
Download Patch File Download Diff File Expand all files Collapse all files
test/registered/unit/disaggregation/test_encode_server.py
+4 -12
View File
@@ -48,6 +48,7 @@ from sglang.srt.mem_cache.multimodal_cache import (
EmbeddingResult,
MultiModalStaticCache,
)
from sglang.srt.runtime_context import get_parallel
from sglang.srt.utils.common import safe_pickle_loads
from sglang.test.ci.ci_register import register_cpu_ci
from sglang.test.test_utils import CustomTestCase
@@ -1011,10 +1012,7 @@ class TestEncoderDelivery(CustomTestCase):
statuses[1].copy_(torch.tensor([400, 1, 0, 0]))
with (
patch(
"sglang.srt.distributed.parallel_state.get_tp_group",
return_value=TPGroup(),
),
get_parallel().override(tp_group=TPGroup()),
patch(
"sglang.srt.disaggregation.encoder.server.torch.distributed.all_gather",
side_effect=all_gather,
@@ -1051,10 +1049,7 @@ class TestEncoderDelivery(CustomTestCase):
statuses[1][2] += 1
with (
patch(
"sglang.srt.distributed.parallel_state.get_tp_group",
return_value=TPGroup(),
),
get_parallel().override(tp_group=TPGroup()),
patch(
"sglang.srt.disaggregation.encoder.server.torch.distributed.all_gather",
side_effect=all_gather,
@@ -1095,10 +1090,7 @@ class TestEncoderDelivery(CustomTestCase):
statuses[1].copy_(local_status)
with (
patch(
"sglang.srt.distributed.parallel_state.get_tp_group",
return_value=TPGroup(),
),
get_parallel().override(tp_group=TPGroup()),
patch(
"sglang.srt.disaggregation.encoder.server.torch.distributed.all_gather",
side_effect=all_gather,
test/registered/unit/disaggregation/test_register_to_bootstrap.py
+11 -8
View File
@@ -9,7 +9,7 @@ import unittest
from unittest.mock import MagicMock, call, patch
from sglang.srt.environ import envs
from sglang.srt.runtime_context import get_context
from sglang.srt.runtime_context import get_context, get_parallel
from sglang.test.test_utils import CustomTestCase
@@ -201,12 +201,12 @@ class TestRegisterToBootstrap(CustomTestCase):
self.assertIn("10.0.0.1", url_used)
@patch("sglang.srt.disaggregation.common.conn.requests.put")
# The consumer reads the group through `get_parallel()`, which reads
# through to the canonical getter, so that is where the stub belongs.
@patch("sglang.srt.distributed.parallel_state.get_world_group")
def test_rust_attention_dp_replicates_complete_topology_across_hosts(
self, mock_world_group, mock_put
self, mock_put
):
# The consumer reads the group through `get_parallel()`, so the
# stub is stated there rather than in the module the build writes.
mock_world_group = MagicMock()
success_resp = MagicMock()
success_resp.status_code = 200
mock_put.return_value = success_resp
@@ -230,9 +230,12 @@ class TestRegisterToBootstrap(CustomTestCase):
for dp_rank, tp_rank, host, rank_port, _ in schedulers
]
mock_world_group.return_value.all_gather_object.side_effect = gather_topology
mock_world_group.all_gather_object.side_effect = gather_topology
with envs.SGLANG_RUST_SERVER.override(True):
with (
get_parallel().override(world_group=mock_world_group),
envs.SGLANG_RUST_SERVER.override(True),
):
for dp_rank, tp_rank, local_ip, _, rust_http_port in schedulers:
manager = self._make_manager()
manager.attn_dp_size = 2
@@ -277,7 +280,7 @@ class TestRegisterToBootstrap(CustomTestCase):
gather_call.args[0]["attn_dp_rank"],
gather_call.args[0]["attn_tp_rank"],
)
for gather_call in mock_world_group.return_value.all_gather_object.call_args_list
for gather_call in mock_world_group.all_gather_object.call_args_list
],
[(dp, tp) for dp, tp, _, _, _ in schedulers],
)