[Rebase] Align vLLM-Omni with vLLM v0.31.0 (#8459)
Signed-off-by: tzhouam <tzhouam@connect.ust.hk>
This commit is contained in:
1 parent
1f47fa2f29
commit
0b0d2d7009
141 files changed
+2885
-721
No files matched your search
+58
-12
@@ -659,6 +659,14 @@ class OmniRunner:
|
||||
self.seed = seed
|
||||
self._prompt_len_estimate_cache: dict[str, Any] = {}
|
||||
self.omni: Any = None
|
||||
self._runner_process = psutil.Process()
|
||||
self._initial_child_ids: set[tuple[int, float]] = set()
|
||||
self._owned_engine_processes: dict[int, float] = {}
|
||||
for child in self._runner_process.children(recursive=True):
|
||||
try:
|
||||
self._initial_child_ids.add((child.pid, child.create_time()))
|
||||
except psutil.Error:
|
||||
pass
|
||||
try:
|
||||
from vllm_omni.entrypoints.omni import Omni
|
||||
|
||||
@@ -670,12 +678,14 @@ class OmniRunner:
|
||||
deploy_config=deploy_config,
|
||||
**kwargs,
|
||||
)
|
||||
self._remember_engine_processes(startup=True)
|
||||
startup_s = time.perf_counter() - startup_t0
|
||||
if log_stats:
|
||||
print(f"OmniRunner startup took {startup_s:.3f}s (model={model_name})", flush=True)
|
||||
except BaseException:
|
||||
# ``with OmniRunner(...)`` never reaches ``__enter__``/``__exit__``
|
||||
# when construction fails after worker processes have started.
|
||||
self._remember_engine_processes(startup=True)
|
||||
self.__exit__(None, None, None)
|
||||
raise
|
||||
|
||||
@@ -899,18 +909,51 @@ class OmniRunner:
|
||||
def stop_profile(self, stages: list[int] | None = None) -> list[Any]:
|
||||
return self.omni.stop_profile(stages=stages)
|
||||
|
||||
def _remember_engine_processes(self, *, startup: bool = False) -> None:
|
||||
"""Track startup workers, then only descendants of those workers.
|
||||
|
||||
Existing children may belong to another runner or a shared transcriber.
|
||||
Record identities before close() can reparent surviving workers.
|
||||
"""
|
||||
try:
|
||||
if startup:
|
||||
candidates = self._runner_process.children(recursive=True)
|
||||
else:
|
||||
candidates = []
|
||||
for pid, created in list(self._owned_engine_processes.items()):
|
||||
try:
|
||||
proc = psutil.Process(pid)
|
||||
if proc.create_time() == created:
|
||||
candidates.extend(proc.children(recursive=True))
|
||||
except psutil.Error:
|
||||
pass
|
||||
for proc in candidates:
|
||||
try:
|
||||
identity = (proc.pid, proc.create_time())
|
||||
if identity in self._initial_child_ids:
|
||||
continue
|
||||
if startup and any(
|
||||
(parent.pid, parent.create_time()) in self._initial_child_ids for parent in proc.parents()
|
||||
):
|
||||
continue
|
||||
cmdline = " ".join(proc.cmdline()).lower()
|
||||
if "enginecore" in cmdline or "enginecore" in proc.name().lower():
|
||||
self._owned_engine_processes[identity[0]] = identity[1]
|
||||
except psutil.Error:
|
||||
pass
|
||||
except psutil.Error as error:
|
||||
print(f"Warning: could not inspect runner workers: {error}")
|
||||
|
||||
def _cleanup_process(self):
|
||||
try:
|
||||
keywords = ["enginecore"]
|
||||
matched = []
|
||||
for proc in psutil.process_iter(["pid", "name", "cmdline", "username"]):
|
||||
for pid, created in self._owned_engine_processes.items():
|
||||
try:
|
||||
cmdline = " ".join(proc.cmdline()).lower() if proc.cmdline() else ""
|
||||
name = proc.name().lower()
|
||||
if any(k in cmdline for k in keywords) or any(k in name for k in keywords):
|
||||
print(f"Found vllm process: PID={proc.pid}, cmd={cmdline[:100]}")
|
||||
proc = psutil.Process(pid)
|
||||
if proc.create_time() == created:
|
||||
print(f"Found owned vllm process: PID={proc.pid}")
|
||||
matched.append(proc)
|
||||
except (psutil.NoSuchProcess, psutil.AccessDenied):
|
||||
except psutil.Error:
|
||||
pass
|
||||
for proc in matched:
|
||||
try:
|
||||
@@ -938,11 +981,14 @@ class OmniRunner:
|
||||
return self
|
||||
|
||||
def __exit__(self, exc_type, exc_val, exc_tb):
|
||||
omni = getattr(self, "omni", None)
|
||||
if omni is not None and hasattr(omni, "close"):
|
||||
omni.close()
|
||||
self._cleanup_process()
|
||||
cleanup_test_environment()
|
||||
self._remember_engine_processes()
|
||||
try:
|
||||
omni = getattr(self, "omni", None)
|
||||
if omni is not None and hasattr(omni, "close"):
|
||||
omni.close()
|
||||
finally:
|
||||
self._cleanup_process()
|
||||
cleanup_test_environment()
|
||||
|
||||
|
||||
# ---------------------------------------------------------------------------
|
||||
|
||||
Reference in new issue
Block a user