[CI] Ignore ruff D209, rejoin the docstrings it split, and silence incompatible-rule warnings (#57212)

Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
This commit is contained in:
Harry Mellor
2026-09-16 10:57:24 -07:00
committed by GitHub
co-authored by Claude Opus 5
parent 1085b64425
commit 651a88c09a
693 changed files with 2088 additions and 4171 deletions
+5 -10
View File
@@ -59,8 +59,7 @@ def test_with_env_set(monkeypatch, mode):
@create_new_process_for_each_test()
def test_other_threads_are_not_policed(monkeypatch):
"""A background thread that syncs deliberately must not be broken by the
check being armed on the thread running the decorated function.
"""
check being armed on the thread running the decorated function."""
monkeypatch.setattr(gsd, "_SYNC_CHECK_MODE", "error")
monkeypatch.setattr(gsd, "_sync_check_enabled", True)
@@ -85,8 +84,7 @@ def test_other_threads_are_not_policed(monkeypatch):
def test_allow_on_other_thread_does_not_disarm(monkeypatch):
"""`gpu_sync_allowed()` on one thread must not suppress the check on
another. It is scoped by ContextVar rather than torch's process-global
sync debug mode, which a previous implementation mutated.
"""
sync debug mode, which a previous implementation mutated."""
monkeypatch.setattr(gsd, "_SYNC_CHECK_MODE", "error")
monkeypatch.setattr(gsd, "_sync_check_enabled", True)
@@ -115,8 +113,7 @@ def test_allow_on_other_thread_does_not_disarm(monkeypatch):
def test_suppressing_works_while_compiling(monkeypatch):
"""`_suppressing` wraps torch compile entry points, which run with
`torch.compiler.is_compiling()` true. `gpu_sync_allowed()` deliberately
no-ops in that state, so `_suppressing` must not route through it.
"""
no-ops in that state, so `_suppressing` must not route through it."""
monkeypatch.setattr(gsd, "_SYNC_CHECK_MODE", "error")
monkeypatch.setattr(gsd, "_sync_check_enabled", True)
# Emulate being inside a torch compile, as inductor passes are.
@@ -148,8 +145,7 @@ def test_copy_checker_is_traceable_while_compiling(monkeypatch):
def test_sync_debug_mode_restored_after_checked_call(monkeypatch):
"""The mode is armed only for the duration of a checked call. Leaving it
on process-wide made every sync outside a checked region emit a
`UserWarning` whenever our handler was not the installed one.
"""
`UserWarning` whenever our handler was not the installed one."""
monkeypatch.setattr(gsd, "_SYNC_CHECK_MODE", "error")
monkeypatch.setattr(gsd, "_sync_check_enabled", True)
@@ -294,8 +290,7 @@ def test_genuinely_async_transfers_pass(monkeypatch):
asynchronous in both directions -- even with a dtype conversion, which
runs GPU-side, and even permuted (e.g. transposed), which uses pitched
cudaMemcpy2D/3DAsync; D2H `Tensor.to` allocates a pinned destination.
None of these may be flagged.
"""
None of these may be flagged."""
monkeypatch.setattr(gsd, "_SYNC_CHECK_MODE", "error")
monkeypatch.setattr(gsd, "_sync_check_enabled", True)
gsd._install_copy_checkers()
+5 -10
View File
@@ -29,8 +29,7 @@ def test_get_file_store_init_method_is_unique():
def _call_with_timeout(func, timeout: float = 10.0):
"""Run func in a daemon thread so a livelock regression fails the test
quickly instead of hanging the suite.
"""
quickly instead of hanging the suite."""
result: dict = {}
def target():
@@ -97,8 +96,7 @@ def test_get_open_port_skips_reserved_dp_master_ports(
monkeypatch: pytest.MonkeyPatch,
):
"""VLLM_PORT == VLLM_DP_MASTER_PORT must not livelock by returning the
same free-but-reserved port forever (upstream issue #50024).
"""
same free-but-reserved port forever (upstream issue #50024)."""
base = 42123
with monkeypatch.context() as m:
m.setenv("VLLM_PORT", str(base))
@@ -111,8 +109,7 @@ def test_get_open_ports_list_skips_reserved_dp_master_ports(
monkeypatch: pytest.MonkeyPatch,
):
"""The VLLM_PORT band scan must not hand out ports inside the window
reserved for the data parallel master process.
"""
reserved for the data parallel master process."""
base = 42223
reserved = range(base, base + 10)
with monkeypatch.context() as m:
@@ -128,8 +125,7 @@ def test_get_open_port_ephemeral_skips_reserved_range(
monkeypatch: pytest.MonkeyPatch,
):
"""An ephemeral port that happens to fall in the reserved range is
replaced by a rescan starting past the range.
"""
replaced by a rescan starting past the range."""
base = 42323
with monkeypatch.context() as m:
m.delenv("VLLM_PORT", raising=False)
@@ -148,8 +144,7 @@ def test_get_open_port_ephemeral_without_dp_master_port(
monkeypatch: pytest.MonkeyPatch,
):
"""Without VLLM_PORT and VLLM_DP_MASTER_PORT, an ephemeral port is
returned as before.
"""
returned as before."""
with monkeypatch.context() as m:
m.delenv("VLLM_PORT", raising=False)
m.delenv("VLLM_DP_MASTER_PORT", raising=False)
+6 -12
View File
@@ -157,8 +157,7 @@ def test_pct_binding_filters_cpus(monkeypatch):
)
def test_pct_binding_fires_on_every_capable_sku(monkeypatch, sku, expected_cpus):
"""Each SKU in ``_PCT_CAPABLE_SKUS`` engages the gate at its own
expected ``highest_perf`` and uses its own priority-core stride.
"""
expected ``highest_perf`` and uses its own priority-core stride."""
sku_config = numa_utils._PCT_CAPABLE_SKUS[sku]
_patch_pct_gates(
monkeypatch,
@@ -243,8 +242,7 @@ def test_get_numactl_args_engine_core_baseline_spans_shard_numa_nodes():
def test_get_numactl_args_engine_core_pct_spans_shard_numa_nodes(monkeypatch):
"""PCT: EngineCore for a multi-NUMA shard binds to the union of priority
cores across all shard nodes, so worker `--physcpubind` is always a
subset of EngineCore's `cpus_allowed`.
"""
subset of EngineCore's `cpus_allowed`."""
_patch_pct_gates(
monkeypatch,
model_match=True,
@@ -294,8 +292,7 @@ def test_get_numactl_args_engine_core_pct_external_launcher_spans_local_nodes(
"""external_launcher (or multi-node-within-DP, or Ray) hits the
fallback branch. EngineCore must still span every local NUMA node
so it can mp-spawn its local workers without ``--physcpubind``
strict-validation failures.
"""
strict-validation failures."""
_patch_pct_gates(
monkeypatch,
model_match=True,
@@ -319,8 +316,7 @@ def test_get_numactl_args_engine_core_pct_external_launcher_spans_local_nodes(
def test_get_numactl_args_engine_core_baseline_multi_node_within_dp_spans_locals():
"""Multi-node-within-DP fallback: bind EngineCore to all local NUMA
nodes that the visible ``numa_bind_nodes`` reference.
"""
nodes that the visible ``numa_bind_nodes`` reference."""
vllm_config = _make_config(
numa_bind=True,
numa_bind_nodes=[0, 0, 0, 0, 1, 1, 1, 1],
@@ -343,8 +339,7 @@ def test_get_numactl_args_engine_core_skips_user_cpu_list(monkeypatch):
``--physcpubind`` spawns need. We fall back to ``--cpunodebind`` over
the shard's NUMA nodes instead. PCT auto-detect is also bypassed when
the user is explicit (its priority-core union may not be a superset
of the user's per-worker cores).
"""
of the user's per-worker cores)."""
_patch_pct_gates(monkeypatch, model_match=True, highest_perf=46)
vllm_config = _make_config(
numa_bind=True,
@@ -397,8 +392,7 @@ def test_get_numactl_args_requires_detectable_nodes(monkeypatch):
def test_configure_subprocess_rejects_unknown_process_kind():
"""configure_subprocess only knows 'worker' and 'EngineCore'; anything
else must raise ValueError instead of silently routing to the worker
path.
"""
path."""
vllm_config = _make_config(numa_bind=True, numa_bind_nodes=[0])
with (
pytest.raises(ValueError, match="process_kind"),