ls-algorithm-plugin-sdk 0.3.3__py3-none-any.whl → 0.3.4__py3-none-any.whl

algorithm_plugin_sdk/cli_impl/configure.py

@@ -62,6 +62,16 @@ def configure_parser(parser: argparse.ArgumentParser) -> None:
         help="maximum concurrent executions (default: GPU count, or 1)",
     )
     parser.add_argument("--scratch-dir")
+    parser.add_argument(
+        "--local-scratch",
+        action="store_true",
+        help="override request scratchRoot with <repository>/.scratch",
+    )
+    parser.add_argument(
+        "--local-output",
+        action="store_true",
+        help="override request output paths with <repository>/.output/<job-id>",
+    )
     parser.add_argument(
         "--gpu-ids",
         type=gpu_ids,
@@ -409,6 +419,8 @@ def generate_config(
         webui=args.webui,
         max_concurrency=max_concurrency,
         scratch_dir=args.scratch_dir,
+        local_scratch=args.local_scratch,
+        local_output=args.local_output,
         token=args.service_token,
         registration=registration_from_args(args, repository),
         gpu_ids=args.gpu_ids,

algorithm_plugin_sdk/cli_impl/serve.py

@@ -81,6 +81,8 @@ def configure_parser(parser: argparse.ArgumentParser) -> None:
 def execute(args: argparse.Namespace) -> int:
     config: DeploymentConfig | None = None
     service: dict[str, object] | None = None
+    local_scratch = False
+    local_output = False
     if args.config:
         # Apply the process mask before refreshing/loading the algorithm.  A
         # repository module may import CUDA during module import.
@@ -91,6 +93,8 @@ def execute(args: argparse.Namespace) -> int:
         args.port = int(service["port"])
         args.max_concurrent_executions = int(service["maxConcurrency"])
         args.scratch_dir = service.get("scratchDir")
+        local_scratch = bool(service.get("localScratch", False))
+        local_output = bool(service.get("localOutput", False))
         args.gpu_ids = list(service.get("gpuIds", []))
         args.webui = bool(service.get("webui"))
         values = config.process_environment()
@@ -120,6 +124,8 @@ def execute(args: argparse.Namespace) -> int:
         ),
         max_concurrent_executions=args.max_concurrent_executions,
         scratch_dir=args.scratch_dir,
+        local_scratch=local_scratch,
+        local_output=local_output,
         gpu_ids=args.gpu_ids,
         runner_factory=create_runner,
         max_attempts=args.max_attempts,

algorithm_plugin_sdk/deployment.py

@@ -146,6 +146,8 @@ class DeploymentConfig:
         gpu_ids: list[int] | None = None,
         environment: dict[str, str] | None = None,
         default_parameters: dict[str, Any] | None = None,
+        local_scratch: bool = False,
+        local_output: bool = False,
     ) -> "DeploymentConfig":
         repository = repository_root(root)
         _make_repository_importable(repository)
@@ -189,6 +191,8 @@ class DeploymentConfig:
                 "webui": webui,
                 "maxConcurrency": max_concurrency,
                 "scratchDir": _absolute(scratch_dir, repository),
+                "localScratch": local_scratch,
+                "localOutput": local_output,
                 "gpuIds": configured_gpu_ids,
                 "token": token or None,
                 "defaultParameters": dict(default_parameters or {}),
@@ -268,6 +272,9 @@ class DeploymentConfig:
             raise RuntimeError("service port must be in [1, 65535]")
         if concurrency < 1:
             raise RuntimeError("service maxConcurrency must be positive")
+        for name in ("localScratch", "localOutput"):
+            if not isinstance(service.get(name, False), bool):
+                raise RuntimeError(f"service {name} must be a boolean")
         gpu_ids = service.get("gpuIds")
         if not isinstance(gpu_ids, list) or any(
             isinstance(gpu_id, bool) or not isinstance(gpu_id, int) or gpu_id < 0

algorithm_plugin_sdk/log_manager.py

@@ -10,6 +10,9 @@ from typing import Any
 from .context import utc_now
 
 _SAFE_NAME = re.compile(r"[^A-Za-z0-9._-]+")
+_TIMESTAMP_GLOB = "[0-9]" * 14
+_FILE_LOCKS_GUARD = threading.Lock()
+_FILE_LOCKS: dict[Path, threading.RLock] = {}
 
 
 def _safe_name(value: str) -> str:
@@ -17,20 +20,61 @@ def _safe_name(value: str) -> str:
     return cleaned or "job"
 
 
+def _file_lock(path: Path) -> threading.RLock:
+    with _FILE_LOCKS_GUARD:
+        return _FILE_LOCKS.setdefault(path, threading.RLock())
+
+
+def _timestamp_prefix() -> str:
+    return utc_now().replace("-", "").replace(":", "").replace("T", "")[:14]
+
+
+def _job_log_path(directory: Path, job_id: str) -> Path:
+    safe_job_id = _safe_name(job_id)
+    pattern = f"{_TIMESTAMP_GLOB}-{safe_job_id}.log"
+    with _FILE_LOCKS_GUARD:
+        existing = sorted(directory.glob(pattern))
+        if existing:
+            return existing[0]
+        timestamp = _timestamp_prefix()
+        path = directory / f"{timestamp}-{safe_job_id}.log"
+        path.touch(exist_ok=True)
+        return path
+
+
 class ExecutionLog:
-    """Per-execution, line-buffered log sink attached to context.logger."""
+    """Per-job, line-buffered log sink attached to context.logger."""
 
-    def __init__(self, *, log_dir: str | Path, job_id: str, execution_id: str, logger: logging.Logger) -> None:
+    def __init__(
+        self,
+        *,
+        log_dir: str | Path,
+        job_id: str,
+        execution_id: str,
+        logger: logging.Logger,
+        task_id: str | None = None,
+    ) -> None:
+        resolved_task_id = task_id or execution_id
+        self._event_context = {
+            "jobId": job_id,
+            "taskId": resolved_task_id,
+            "executionId": execution_id,
+            "job_id": job_id,
+            "task_id": resolved_task_id,
+            "execution_id": execution_id,
+        }
         directory = Path(log_dir).expanduser()
         directory.mkdir(parents=True, exist_ok=True)
-        path = directory / f"{_safe_name(job_id)}.log"
-        if path.exists():
-            path = directory / f"{_safe_name(job_id)}-{_safe_name(execution_id)}.log"
+        path = _job_log_path(directory, job_id)
         self.path = path
         self._stream = path.open("a", encoding="utf-8", buffering=1)
-        self._lock = threading.RLock()
-        self._handler = _FlushFileHandler(self._stream)
-        self._handler.setFormatter(logging.Formatter("%(asctime)s %(levelname)s %(message)s"))
+        self._lock = _file_lock(path)
+        self._handler = _FlushFileHandler(self._stream, self._lock)
+        self._handler.setFormatter(
+            logging.Formatter(
+                f"%(asctime)s %(levelname)s [executionId={execution_id}] %(message)s"
+            )
+        )
         self._handler.setLevel(logging.DEBUG)
         self.logger = logger
         self.logger.setLevel(logging.DEBUG)
@@ -38,7 +82,12 @@ class ExecutionLog:
         self._closed = False
 
     def write_event(self, event: str, **values: Any) -> None:
-        payload = {"event": event, "timestamp": utc_now(), **values}
+        payload = {
+            "event": event,
+            "timestamp": utc_now(),
+            **self._event_context,
+            **values,
+        }
         with self._lock:
             if self._closed:
                 return
@@ -56,10 +105,10 @@ class ExecutionLog:
 
 
 class _FlushFileHandler(logging.Handler):
-    def __init__(self, stream: Any) -> None:
+    def __init__(self, stream: Any, lock: threading.RLock) -> None:
         super().__init__()
         self.stream = stream
-        self._lock = threading.RLock()
+        self._lock = lock
 
     def emit(self, record: logging.LogRecord) -> None:
         try:

algorithm_plugin_sdk/service.py

@@ -6,6 +6,7 @@ import logging
 import random
 import threading
 import time
+import traceback
 import uuid
 from concurrent.futures import Future, ThreadPoolExecutor
 from contextlib import asynccontextmanager
@@ -123,6 +124,8 @@ class ExecutionManager:
         scratch_dir: str | None = None,
         repository: str | Path | None = None,
         log_dir: str | Path | None = None,
+        local_scratch: bool = False,
+        local_output: bool = False,
         gpu_ids: list[int] | None = None,
         runner_factory: RunnerFactory | None = None,
         max_attempts: int = MAX_ATTEMPTS,
@@ -138,6 +141,17 @@ class ExecutionManager:
             raise ValueError("retry_seconds must be positive")
         if not 0 <= retry_jitter_ratio <= 1:
             raise ValueError("retry_jitter_ratio must be in [0, 1]")
+        if not isinstance(local_scratch, bool) or not isinstance(
+            local_output, bool
+        ):
+            raise ValueError("local_scratch and local_output must be booleans")
+        repository_path = (
+            Path(repository).expanduser().resolve()
+            if repository is not None
+            else None
+        )
+        if (local_scratch or local_output) and repository_path is None:
+            raise ValueError("repository is required for local scratch or output")
         configured_gpu_ids = None if gpu_ids is None else list(gpu_ids)
         if configured_gpu_ids is not None and (
             any(
@@ -156,10 +170,19 @@ class ExecutionManager:
             raise ValueError("default_parameters must be a JSON object")
         self.default_parameters = dict(default_parameters or {})
         self.scratch_dir = scratch_dir
+        self._local_scratch_root = (
+            repository_path / ".scratch" if local_scratch else None
+        )
+        self._local_output_root = (
+            repository_path / ".output" if local_output else None
+        )
+        for local_root in (self._local_scratch_root, self._local_output_root):
+            if local_root is not None:
+                local_root.mkdir(parents=True, exist_ok=True)
         self.log_dir = (
             Path(log_dir).expanduser()
             if log_dir is not None
-            else (Path(repository).expanduser() / "logs" if repository is not None else None)
+            else (repository_path / "logs" if repository_path is not None else None)
         )
         self.gpu_ids = configured_gpu_ids
         self.max_concurrent_executions = max_concurrent_executions
@@ -191,6 +214,46 @@ class ExecutionManager:
             self.runner.start()
             self._started = True
 
+    def _apply_local_paths(
+        self,
+        request: AlgorithmRequest,
+        job_id: str,
+    ) -> AlgorithmRequest:
+        workspace = dict(request.workspace or {})
+        inputs = request.inputs
+        if self._local_scratch_root is not None:
+            workspace["scratchRoot"] = str(self._local_scratch_root)
+        if self._local_output_root is not None:
+            workspace["outputRoot"] = str(self._local_output_root)
+            job_output = self._job_output_dir(job_id)
+            if len(inputs) == 1:
+                output_paths = [job_output]
+            else:
+                output_paths = [
+                    job_output / f"{index}-{Path(item.output).name or 'output'}"
+                    for index, item in enumerate(inputs)
+                ]
+                for output_path in output_paths:
+                    output_path.mkdir(parents=True, exist_ok=True)
+            inputs = [
+                replace(item, output=str(output_path))
+                for item, output_path in zip(inputs, output_paths)
+            ]
+        return replace(request, inputs=inputs, workspace=workspace or None)
+
+    def _job_output_dir(self, job_id: str) -> Path:
+        assert self._local_output_root is not None
+        if Path(job_id).name != job_id or job_id in {"", ".", ".."}:
+            raise ValueError(f"job ID is unsafe for local output: {job_id!r}")
+        output = (self._local_output_root / job_id).resolve()
+        if (
+            output == self._local_output_root
+            or self._local_output_root not in output.parents
+        ):
+            raise ValueError(f"job output escapes local root: {job_id!r}")
+        output.mkdir(parents=True, exist_ok=True)
+        return output
+
     def submit(
         self,
         request: AlgorithmRequest,
@@ -226,6 +289,7 @@ class ExecutionManager:
         execution_id = f"exec-{uuid.uuid4().hex}"
         resolved_job_id = str(job_id or execution_id)
         resolved_task_id = str(task_id or execution_id)
+        request = self._apply_local_paths(request, resolved_job_id)
         record = ExecutionRecord(
             execution_id=execution_id,
             request=request,
@@ -237,10 +301,9 @@ class ExecutionManager:
         record.context = ExecutionContext(
             execution_id,
             [item.input_dataset for item in request.inputs],
-            scratch_dir=(
-                record.request.workspace["scratchRoot"]
-                if record.request.workspace is not None
-                else self.scratch_dir
+            scratch_dir=(record.request.workspace or {}).get(
+                "scratchRoot",
+                self.scratch_dir,
             ),
             progress_callback=record.add_progress,
         )
@@ -248,6 +311,7 @@ class ExecutionManager:
             record.log = ExecutionLog(
                 log_dir=self.log_dir,
                 job_id=record.job_id,
+                task_id=record.task_id,
                 execution_id=record.execution_id,
                 logger=record.context.logger,
             )
@@ -430,10 +494,16 @@ class ExecutionManager:
                 record.log.write_event("result", state="cancelled", error=str(exc))
         except Exception as exc:
             error = f"{type(exc).__name__}: {exc}"
+            stack_trace = traceback.format_exc()
             record.context.mark_unfinished("failed", error)
             record.set_state("failed", error=error)
             if record.log is not None:
-                record.log.write_event("result", state="failed", error=error)
+                record.log.write_event(
+                    "result",
+                    state="failed",
+                    error=error,
+                    traceback=stack_trace,
+                )
         else:
             record.set_state(result.status, result=result)
             if record.log is not None:
@@ -466,7 +536,7 @@ class ExecutionManager:
                         self.retry_seconds * (1 - self.retry_jitter_ratio),
                         self.retry_seconds * (1 + self.retry_jitter_ratio),
                     )
-                    logger.warning(
+                    record.context.logger.warning(
                         "algorithm execution %s failed on attempt %d/%d; "
                         "releasing and restarting algorithm in %.3f seconds: %s: %s",
                         record.execution_id,
@@ -475,16 +545,18 @@ class ExecutionManager:
                         delay,
                         type(exc).__name__,
                         exc,
+                        exc_info=True,
                     )
                 else:
                     delay = 0.0
-                    logger.error(
+                    record.context.logger.error(
                         "algorithm execution %s exhausted %d attempts; "
                         "releasing and restarting algorithm for future requests: %s: %s",
                         record.execution_id,
                         self.max_attempts,
                         type(exc).__name__,
                         exc,
+                        exc_info=True,
                     )
             finally:
                 self._release_runner()

{ls_algorithm_plugin_sdk-0.3.3.dist-info → ls_algorithm_plugin_sdk-0.3.4.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: ls-algorithm-plugin-sdk
-Version: 0.3.3
+Version: 0.3.4
 Summary: Protocol-independent runtime SDK for dataset algorithms
 Author: Ling Robotics
 License: Proprietary

{ls_algorithm_plugin_sdk-0.3.3.dist-info → ls_algorithm_plugin_sdk-0.3.4.dist-info}/RECORD

@@ -2,22 +2,22 @@ algorithm_plugin_sdk/__init__.py,sha256=OKj2VcNXTwZmRgPlECI6QUBlxW1YLtflr4RAgs6w
 algorithm_plugin_sdk/algorithm.py,sha256=INjX-ygMQfHk8D1AZbEEdCw1P4VNQLGF989lP1J1H1k,1025
 algorithm_plugin_sdk/cli.py,sha256=V9HFJ98iyhM1q6PZqd-366g1AoKuieC_Fn4u9RJcMNI,1252
 algorithm_plugin_sdk/context.py,sha256=Kj-eDZg7PgX6xNzrWrQs37UfepTP8fXv4_8hCgsAIDA,11016
-algorithm_plugin_sdk/deployment.py,sha256=nx4-26vs5OqO6miBf6GegZIhdvOgFaw9MdjpTHZNDPo,14583
+algorithm_plugin_sdk/deployment.py,sha256=5WlZfML9ymWrAFsmdX-bBHTP2kuVY4WO5wKHs35Dkd4,14936
 algorithm_plugin_sdk/errors.py,sha256=DNmI2c36Pd7K6tTaonqfEgpBzwX0o6pXzYrc8r2Z--k,730
 algorithm_plugin_sdk/gpu_isolation.py,sha256=toXUQ1vfzfLPiideJ8dWkfa9NW2Lc8V0sbNx-75YTjg,794
 algorithm_plugin_sdk/loader.py,sha256=XKmyqbJMoUMdKoo2eRQFGdbsmlz6NBgtUeIdqzumJFs,2561
-algorithm_plugin_sdk/log_manager.py,sha256=ktVVvf4khqXiYENr1QRpSf7yTG4HBzxW2AOHIZElDuE,2457
+algorithm_plugin_sdk/log_manager.py,sha256=fmqSivxi-BCOWs4Qomms9nHZC7MRl5NUVuRiTL0J12Q,3723
 algorithm_plugin_sdk/models.py,sha256=FIwUm8QDK8L37pV7AcCMnmYlDP-2l93zp8yc0KSu4yo,12196
 algorithm_plugin_sdk/registration.py,sha256=s5oQkf12_uRjdA4CYMYM8SDQjhsk1rzDDF-3w36U_lA,10474
 algorithm_plugin_sdk/release.py,sha256=KM-kRM1oFYMgjDFnXdu6BLR1b4dsvWWDCEgTNq8cb3w,9439
 algorithm_plugin_sdk/runner.py,sha256=583Y9_7efvW5rrmM2nt316NEpCRO6kd8dKYjFRjgtCk,3183
-algorithm_plugin_sdk/service.py,sha256=HcO8PmtPczogi0an7irajsRCD-11GAH8pnE7rOHOLSM,28004
+algorithm_plugin_sdk/service.py,sha256=HQLDW5nlGGXJNZ0KMFU4I1rPzWHMf_LlE-tsY42PHnc,31045
 algorithm_plugin_sdk/webui_app.py,sha256=Gvt2FPt0xZBjYi39sqTQqcosllifYeJam49_mGfkEVw,1530
 algorithm_plugin_sdk/cli_impl/__init__.py,sha256=uLknQVGipNd7eSJegDHZS0NVowMa0qAaOf63wFI3zHM,78
-algorithm_plugin_sdk/cli_impl/configure.py,sha256=OlYuNnPZXHbaoWKfBpFivyP8lZk2QQpvMN572hAjriA,13353
+algorithm_plugin_sdk/cli_impl/configure.py,sha256=hO9vwFfZ4yqeQUyYc3heE1MDLK1lppN69OoF2zeVVaM,13761
 algorithm_plugin_sdk/cli_impl/parsing.py,sha256=TDX45Spu4WMrczuzTMN9ZC2SYyHE11OPBCvN45yPsxs,1399
 algorithm_plugin_sdk/cli_impl/run.py,sha256=C4raNwhj4KZJc7_aTJeoiwNMQtgeXyMWyLTZEVUL0mM,6041
-algorithm_plugin_sdk/cli_impl/serve.py,sha256=BBxjFjOI8bORzNij9DbKccAdt4zNzX_vqZyyKtliGSg,4936
+algorithm_plugin_sdk/cli_impl/serve.py,sha256=1Ma08MHMjKk9wKO8tGdymVvedcGC-jaKdaUCVBTPMro,5187
 algorithm_plugin_sdk/examples/__init__.py,sha256=WxZZzUzxKhqZg3KibVY-7CB960DNjt6IfjciZ8JXbq0,57
 algorithm_plugin_sdk/examples/example_algorithm.py,sha256=j2RsVaRPxpJn3kK6_H_JqAceKTpJE8txNm-oTrH2v6E,3787
 algorithm_plugin_sdk/examples/simulated_algorithm.py,sha256=bIaneSMM09CJfV5-oGuZ5QwBDWV1u-7Fn0WDN1TuwGw,2944
@@ -25,8 +25,8 @@ algorithm_plugin_sdk/webui/__init__.py,sha256=cvtaktJXz_DYG4QbV5ppoqY2bFaMLuiUcT
 algorithm_plugin_sdk/webui/app.css,sha256=5HGDKlbUFLZMgCknlRuwCjiNcCr8poIF5YGHchyaamE,2965
 algorithm_plugin_sdk/webui/app.js,sha256=0YsXsocEY3yDseG3jh0RhzI8xo_bYmsbSxdN3FdCswQ,7657
 algorithm_plugin_sdk/webui/index.html,sha256=Z222OuaacWdbr0s-sjJq6wmg45u96tIneM5-LP55Z8I,2261
-ls_algorithm_plugin_sdk-0.3.3.dist-info/METADATA,sha256=iLerkePM3Ff_GG0oA4TAdTsidnxYJxP1pmsSaIARMJw,2317
-ls_algorithm_plugin_sdk-0.3.3.dist-info/WHEEL,sha256=YVMoNqKzERt-wjUZwJ33xBGAwnFl-4cqbYkTtWa4itE,91
-ls_algorithm_plugin_sdk-0.3.3.dist-info/entry_points.txt,sha256=TU0R_TxuB5OuOiZ5BHOEmUPtHvN3v8ARUW8d3PztHcE,67
-ls_algorithm_plugin_sdk-0.3.3.dist-info/top_level.txt,sha256=8lsgxZ8HJGLlzJ5Rt8CiVzw9Ccme286i3akbmeMjBos,21
-ls_algorithm_plugin_sdk-0.3.3.dist-info/RECORD,,
+ls_algorithm_plugin_sdk-0.3.4.dist-info/METADATA,sha256=S21Rk6HWDt6q5H94F_pj8DpUtY-bKCicc7Vlmq1zWxA,2317
+ls_algorithm_plugin_sdk-0.3.4.dist-info/WHEEL,sha256=YVMoNqKzERt-wjUZwJ33xBGAwnFl-4cqbYkTtWa4itE,91
+ls_algorithm_plugin_sdk-0.3.4.dist-info/entry_points.txt,sha256=TU0R_TxuB5OuOiZ5BHOEmUPtHvN3v8ARUW8d3PztHcE,67
+ls_algorithm_plugin_sdk-0.3.4.dist-info/top_level.txt,sha256=8lsgxZ8HJGLlzJ5Rt8CiVzw9Ccme286i3akbmeMjBos,21
+ls_algorithm_plugin_sdk-0.3.4.dist-info/RECORD,,