feat(api): make tile retries configurable
This commit is contained in:
parent
cd06f9291b
commit
6fb0366f52
|
@ -37,6 +37,7 @@ class ServerContext:
|
||||||
memory_limit: Optional[int] = None,
|
memory_limit: Optional[int] = None,
|
||||||
admin_token: Optional[str] = None,
|
admin_token: Optional[str] = None,
|
||||||
server_version: Optional[str] = DEFAULT_SERVER_VERSION,
|
server_version: Optional[str] = DEFAULT_SERVER_VERSION,
|
||||||
|
worker_retries: Optional[int] = 3,
|
||||||
) -> None:
|
) -> None:
|
||||||
self.bundle_path = bundle_path
|
self.bundle_path = bundle_path
|
||||||
self.model_path = model_path
|
self.model_path = model_path
|
||||||
|
@ -56,6 +57,7 @@ class ServerContext:
|
||||||
self.memory_limit = memory_limit
|
self.memory_limit = memory_limit
|
||||||
self.admin_token = admin_token or token_urlsafe()
|
self.admin_token = admin_token or token_urlsafe()
|
||||||
self.server_version = server_version
|
self.server_version = server_version
|
||||||
|
self.worker_retries = worker_retries
|
||||||
|
|
||||||
self.cache = ModelCache(self.cache_limit)
|
self.cache = ModelCache(self.cache_limit)
|
||||||
|
|
||||||
|
@ -88,6 +90,7 @@ class ServerContext:
|
||||||
server_version=environ.get(
|
server_version=environ.get(
|
||||||
"ONNX_WEB_SERVER_VERSION", DEFAULT_SERVER_VERSION
|
"ONNX_WEB_SERVER_VERSION", DEFAULT_SERVER_VERSION
|
||||||
),
|
),
|
||||||
|
worker_retries=int(environ.get("ONNX_WEB_WORKER_RETRIES", 3)),
|
||||||
)
|
)
|
||||||
|
|
||||||
def torch_dtype(self):
|
def torch_dtype(self):
|
||||||
|
|
|
@ -36,6 +36,7 @@ class WorkerContext:
|
||||||
progress: "Queue[ProgressCommand]",
|
progress: "Queue[ProgressCommand]",
|
||||||
active_pid: "Value[int]",
|
active_pid: "Value[int]",
|
||||||
idle: "Value[bool]",
|
idle: "Value[bool]",
|
||||||
|
retries: int,
|
||||||
):
|
):
|
||||||
self.job = None
|
self.job = None
|
||||||
self.name = name
|
self.name = name
|
||||||
|
@ -47,8 +48,8 @@ class WorkerContext:
|
||||||
self.active_pid = active_pid
|
self.active_pid = active_pid
|
||||||
self.last_progress = None
|
self.last_progress = None
|
||||||
self.idle = idle
|
self.idle = idle
|
||||||
|
self.retries = retries
|
||||||
self.timeout = 1.0
|
self.timeout = 1.0
|
||||||
self.retries = 3 # TODO: get from env
|
|
||||||
|
|
||||||
def start(self, job: str) -> None:
|
def start(self, job: str) -> None:
|
||||||
self.job = job
|
self.job = job
|
||||||
|
|
|
@ -124,6 +124,7 @@ class DevicePoolExecutor:
|
||||||
pending=self.pending[name],
|
pending=self.pending[name],
|
||||||
active_pid=current,
|
active_pid=current,
|
||||||
idle=self.worker_idle[name],
|
idle=self.worker_idle[name],
|
||||||
|
retries=self.server.worker_retries,
|
||||||
)
|
)
|
||||||
self.context[name] = context
|
self.context[name] = context
|
||||||
|
|
||||||
|
|
Loading…
Reference in New Issue