ls-algorithm-plugin-sdk 0.3.0__py3-none-any.whl → 0.3.3__py3-none-any.whl

algorithm_plugin_sdk/cli_impl/configure.py

@@ -1,11 +1,14 @@
 from __future__ import annotations
 
 import argparse
+import ipaddress
 import os
 import re
+import socket
 import subprocess
 import sys
 from pathlib import Path
+from urllib.parse import urlparse
 
 from ..deployment import (
     DEFAULT_CONFIG_NAME,
@@ -13,10 +16,15 @@ from ..deployment import (
     repository_root,
     stable_instance_key,
 )
-from .parsing import environment, gpu_ids
+from .parsing import environment, gpu_ids, json_object
 
 
 SYSTEMD_UNIT_DIR = Path("/etc/systemd/system")
+INPUT_ROOT = Path("/mnt/ldp_uploads/ldp-uploads")
+PRIVATE_IPV4_NETWORKS = tuple(
+    ipaddress.ip_network(network)
+    for network in ("10.0.0.0/8", "172.16.0.0/12", "192.168.0.0/16")
+)
 
 SERVICE_NAME_PATTERN = re.compile(
     r"^[A-Za-z0-9][A-Za-z0-9_.@-]*$"
@@ -48,7 +56,11 @@ def configure_parser(parser: argparse.ArgumentParser) -> None:
         action=argparse.BooleanOptionalAction,
         default=True,
     )
-    parser.add_argument("--max-concurrency", type=int, default=1)
+    parser.add_argument(
+        "--max-concurrency",
+        type=int,
+        help="maximum concurrent executions (default: GPU count, or 1)",
+    )
     parser.add_argument("--scratch-dir")
     parser.add_argument(
         "--gpu-ids",
@@ -56,6 +68,12 @@ def configure_parser(parser: argparse.ArgumentParser) -> None:
         default=[],
         help="GPU IDs injected into every service execution",
     )
+    parser.add_argument(
+        "--default-parameters",
+        type=json_object,
+        default={},
+        help="JSON object used to fill missing request parameters",
+    )
     parser.add_argument("--service-token")
     parser.add_argument(
         "--compute-url",
@@ -65,18 +83,6 @@ def configure_parser(parser: argparse.ArgumentParser) -> None:
         "--api-key",
         default=os.getenv("LDP_INTERNAL_API_KEY"),
     )
-    parser.add_argument(
-        "--public-url",
-        default=os.getenv("LDP_PLUGIN_PUBLIC_URL"),
-    )
-    parser.add_argument(
-        "--input-root",
-        default=os.getenv("LDP_JOB_INPUT_ROOT"),
-    )
-    parser.add_argument(
-        "--workspace-root",
-        default=os.getenv("LDP_JOB_WORKSPACE_ROOT"),
-    )
     parser.add_argument(
         "--instance-key",
         default=os.getenv("LDP_PLUGIN_INSTANCE_KEY"),
@@ -85,14 +91,6 @@ def configure_parser(parser: argparse.ArgumentParser) -> None:
         "--cluster",
         default=os.getenv("LDP_CLUSTER"),
     )
-    parser.add_argument(
-        "--namespace",
-        default=os.getenv("POD_NAMESPACE"),
-    )
-    parser.add_argument(
-        "--node-name",
-        default=os.getenv("NODE_NAME"),
-    )
     parser.add_argument(
         "--local",
         action="store_true",
@@ -105,10 +103,6 @@ def configure_parser(parser: argparse.ArgumentParser) -> None:
         default=[],
         metavar="NAME=VALUE",
     )
-    parser.add_argument(
-        "--service-name",
-        help="systemd unit name without .service; defaults to repository name",
-    )
     parser.add_argument(
         "--force",
         action="store_true",
@@ -119,33 +113,35 @@ def configure_parser(parser: argparse.ArgumentParser) -> None:
 
 def resolve_service_name(
     repository: Path,
-    requested: str | None,
 ) -> str:
-    name = requested or repository.name
+    name = repository.name
     if name.endswith(".service"):
         raise ValueError(
-            "service name must not include the .service suffix"
+            "repository name must not include the .service suffix"
         )
     if not SERVICE_NAME_PATTERN.fullmatch(name):
         raise ValueError(
-            "service name may contain only letters, numbers, "
+            "repository name may contain only letters, numbers, "
             "underscore, dot, @, and hyphen"
         )
     return name
 
 
-def _quote_systemd_path(value: Path) -> str:
+def _escape_systemd_path(value: Path) -> str:
     text = str(value)
     if "\n" in text or "\r" in text or "\0" in text:
         raise ValueError(
             f"systemd path contains control characters: {value}"
         )
-    escaped = (
+    return (
         text.replace("\\", "\\\\")
         .replace('"', '\\"')
         .replace("%", "%%")
     )
-    return f'"{escaped}"'
+
+
+def _quote_systemd_path(value: Path) -> str:
+    return f'"{_escape_systemd_path(value)}"'
 
 
 def render_systemd_unit(
@@ -153,10 +149,18 @@ def render_systemd_unit(
     repository: Path,
     config_path: Path,
     service_name: str,
+    gpu_ids: list[int] | None = None,
 ) -> str:
     root = repository.resolve()
+    working_directory = root / "jobs"
     launcher = root / ".venv" / "bin" / "algorithm-plugin"
     config = config_path.resolve()
+    environment = ["Environment=PYTHONUNBUFFERED=1"]
+    if gpu_ids:
+        environment.append(
+            "Environment=CUDA_VISIBLE_DEVICES="
+            + ",".join(str(gpu_id) for gpu_id in gpu_ids)
+        )
     return "\n".join(
         (
             "[Unit]",
@@ -167,12 +171,13 @@ def render_systemd_unit(
             "[Service]",
             "Type=simple",
             "User=root",
-            f"WorkingDirectory={_quote_systemd_path(root)}",
+            f"WorkingDirectory={_escape_systemd_path(working_directory)}",
             (
                 f"ExecStart={_quote_systemd_path(launcher)} "
                 f"serve --config {_quote_systemd_path(config)}"
             ),
-            "Environment=PYTHONUNBUFFERED=1",
+            *environment,
+            "LimitNOFILE=524288",
             "Restart=on-failure",
             "RestartSec=5s",
             "",
@@ -317,6 +322,35 @@ def resolve_config_path(
     return path.resolve()
 
 
+def resolve_private_ipv4(compute_url: str) -> str:
+    parsed = urlparse(compute_url)
+    if parsed.scheme not in {"http", "https"} or not parsed.hostname:
+        raise ValueError("compute URL must be an http or https URL")
+    try:
+        port = parsed.port or (443 if parsed.scheme == "https" else 80)
+    except ValueError as exc:
+        raise ValueError("compute URL has an invalid port") from exc
+
+    try:
+        with socket.socket(socket.AF_INET, socket.SOCK_DGRAM) as probe:
+            probe.connect((parsed.hostname, port))
+            value = probe.getsockname()[0]
+    except OSError as exc:
+        raise RuntimeError(
+            f"cannot resolve the local IP used to reach {parsed.hostname}: {exc}"
+        ) from exc
+
+    address = ipaddress.ip_address(value)
+    if not isinstance(address, ipaddress.IPv4Address) or not any(
+        address in network for network in PRIVATE_IPV4_NETWORKS
+    ):
+        raise RuntimeError(
+            f"the local IP used to reach {parsed.hostname} is not a private IPv4: "
+            f"{address}"
+        )
+    return str(address)
+
+
 def registration_from_args(
     args: argparse.Namespace,
     repository: Path,
@@ -327,12 +361,9 @@ def registration_from_args(
     supplied = {
         "computeUrl": args.compute_url,
         "apiKey": args.api_key,
-        "publicUrl": args.public_url,
-        "inputRoot": args.input_root,
-        "workspaceRoot": args.workspace_root,
+        "inputRoot": str(INPUT_ROOT),
+        "workspaceRoot": str(repository.resolve()),
         "cluster": args.cluster,
-        "namespace": args.namespace,
-        "nodeName": args.node_name,
     }
     missing = [
         name
@@ -350,6 +381,9 @@ def registration_from_args(
         name: str(value)
         for name, value in supplied.items()
     }
+    private_ipv4 = resolve_private_ipv4(registration["computeUrl"])
+    registration["publicUrl"] = f"http://{private_ipv4}:{args.port}"
+    registration["nodeName"] = private_ipv4
     registration["instanceKey"] = stable_instance_key(
         repository / ".algorithm-plugin" / "instance-key",
         requested=args.instance_key,
@@ -361,6 +395,11 @@ def generate_config(
     args: argparse.Namespace,
     repository: Path,
 ) -> DeploymentConfig:
+    max_concurrency = (
+        args.max_concurrency
+        if args.max_concurrency is not None
+        else max(len(args.gpu_ids), 1)
+    )
     config = DeploymentConfig.generate(
         output=args.config_output,
         root=repository,
@@ -368,12 +407,13 @@ def generate_config(
         host=args.host,
         port=args.port,
         webui=args.webui,
-        max_concurrency=args.max_concurrency,
+        max_concurrency=max_concurrency,
         scratch_dir=args.scratch_dir,
         token=args.service_token,
         registration=registration_from_args(args, repository),
         gpu_ids=args.gpu_ids,
         environment=dict(args.env),
+        default_parameters=args.default_parameters,
     )
     config.save()
     return config
@@ -383,10 +423,7 @@ def execute(args: argparse.Namespace) -> int:
     require_root()
     repository = validate_repository(args.repository)
     validate_configure_runtime(repository)
-    service_name = resolve_service_name(
-        repository,
-        args.service_name,
-    )
+    service_name = resolve_service_name(repository)
     config_path = resolve_config_path(
         repository,
         args.config_output,
@@ -395,6 +432,7 @@ def execute(args: argparse.Namespace) -> int:
         repository=repository,
         config_path=config_path,
         service_name=service_name,
+        gpu_ids=args.gpu_ids,
     )
 
     inspect_systemd_unit(
@@ -402,6 +440,7 @@ def execute(args: argparse.Namespace) -> int:
         content=unit,
         force=args.force,
     )
+    (repository / "jobs").mkdir(exist_ok=True)
     config = generate_config(args, repository)
     unit_path, changed = write_systemd_unit(
         service_name=service_name,
@@ -418,4 +457,5 @@ def execute(args: argparse.Namespace) -> int:
     print("  sudo systemctl daemon-reload")
     print(f"  sudo systemctl enable {service_name}.service")
     print(f"  sudo systemctl restart {service_name}.service")
-    return 0
+    print(f"  sudo systemctl status {service_name}.service --no-pager")
+    return 0

algorithm_plugin_sdk/cli_impl/run.py

@@ -17,6 +17,7 @@ from typing import Any, Iterator
 
 from ..context import ExecutionContext, ProgressSnapshot
 from ..errors import ExecutionCancelled
+from ..gpu_isolation import apply_gpu_isolation
 from ..loader import load_algorithm
 from ..models import AlgorithmRequest
 from ..runner import AlgorithmRunner
@@ -125,6 +126,10 @@ def cancel_on_interrupt(
 
 def execute(args: argparse.Namespace) -> int:
     request = request_from_args(args)
+    request = replace(
+        request,
+        gpu_ids=apply_gpu_isolation(request.gpu_ids),
+    )
     progress = TerminalProgress()
     context = ExecutionContext(
         f"cli-{id(args)}",

algorithm_plugin_sdk/cli_impl/serve.py

@@ -5,6 +5,7 @@ import os
 from pathlib import Path
 
 from ..deployment import DEFAULT_CONFIG_NAME, DeploymentConfig
+from ..gpu_isolation import apply_gpu_isolation
 from ..loader import load_algorithm
 from ..registration import PluginRegistrationAgent
 from ..release import ReleaseManifest
@@ -78,9 +79,12 @@ def configure_parser(parser: argparse.ArgumentParser) -> None:
 
 
 def execute(args: argparse.Namespace) -> int:
+    config: DeploymentConfig | None = None
+    service: dict[str, object] | None = None
     if args.config:
-        config = DeploymentConfig.load(args.config)
-        config.save()
+        # Apply the process mask before refreshing/loading the algorithm.  A
+        # repository module may import CUDA during module import.
+        config = DeploymentConfig.load(args.config, refresh=False)
         service = config.payload["service"]
         args.algorithm = config.algorithm_reference
         args.host = str(service["host"])
@@ -96,6 +100,11 @@ def execute(args: argparse.Namespace) -> int:
     elif not args.algorithm:
         raise ValueError("algorithm is required unless --config is provided")
 
+    args.gpu_ids = apply_gpu_isolation(args.gpu_ids)
+    if config is not None:
+        config.refresh()
+        config.save()
+
     algorithm_reference = str(args.algorithm)
     algorithm = load_algorithm(algorithm_reference)
 
@@ -105,6 +114,10 @@ def execute(args: argparse.Namespace) -> int:
     release = ReleaseManifest.discover(algorithm, start=Path.cwd(), required=False)
     manager = ExecutionManager(
         AlgorithmRunner(algorithm),
+        default_parameters=(
+            service.get("defaultParameters", {})
+            if service is not None else None
+        ),
         max_concurrent_executions=args.max_concurrent_executions,
         scratch_dir=args.scratch_dir,
         gpu_ids=args.gpu_ids,
@@ -113,6 +126,7 @@ def execute(args: argparse.Namespace) -> int:
         retry_seconds=args.retry_seconds,
         retry_jitter_ratio=args.retry_jitter_ratio,
         release_manifest=release,
+        repository=(config.repository if config is not None else Path.cwd()),
     )
     registration = PluginRegistrationAgent.from_env(manager.manifest, manager.heartbeat)
     if registration is not None and release is None:

algorithm_plugin_sdk/deployment.py

@@ -145,6 +145,7 @@ class DeploymentConfig:
         registration: dict[str, str] | None,
         gpu_ids: list[int] | None = None,
         environment: dict[str, str] | None = None,
+        default_parameters: dict[str, Any] | None = None,
     ) -> "DeploymentConfig":
         repository = repository_root(root)
         _make_repository_importable(repository)
@@ -154,6 +155,8 @@ class DeploymentConfig:
             instance,
             repository=repository,
         )
+        if registration is not None:
+            registration = {**registration, "namespace": release.name}
         release_path = release.save(
             repository / "release-manifest.json"
         )
@@ -188,6 +191,7 @@ class DeploymentConfig:
                 "scratchDir": _absolute(scratch_dir, repository),
                 "gpuIds": configured_gpu_ids,
                 "token": token or None,
+                "defaultParameters": dict(default_parameters or {}),
             },
             "registration": registration,
             "environment": dict(sorted((environment or {}).items())),
@@ -271,6 +275,16 @@ class DeploymentConfig:
         ) or len(gpu_ids) != len(set(gpu_ids)):
             raise RuntimeError("service gpuIds must be unique non-negative integers")
 
+        default_parameters = service.get("defaultParameters", {})
+        if not isinstance(default_parameters, dict):
+            raise RuntimeError("defaultParameters must be an object")
+        if any(not isinstance(key, str) for key in default_parameters):
+            raise RuntimeError("defaultParameters keys must be strings")
+        try:
+            json.dumps(default_parameters)
+        except (TypeError, ValueError) as exc:
+            raise RuntimeError("defaultParameters must be JSON serializable") from exc
+
     def _validate_registration(self) -> None:
         registration = self.payload.get("registration")
         if registration is None:

algorithm_plugin_sdk/gpu_isolation.py

@@ -0,0 +1,27 @@
+from __future__ import annotations
+
+import os
+from collections.abc import Sequence
+
+
+def apply_gpu_isolation(gpu_ids: Sequence[int] | None) -> list[int]:
+    """Mask physical GPUs for this process and return local GPU ordinals."""
+    if gpu_ids is None:
+        return []
+
+    physical_gpu_ids = list(gpu_ids)
+    if any(
+        isinstance(gpu_id, bool)
+        or not isinstance(gpu_id, int)
+        or gpu_id < 0
+        for gpu_id in physical_gpu_ids
+    ) or len(physical_gpu_ids) != len(set(physical_gpu_ids)):
+        raise ValueError("gpu_ids must be unique non-negative integers")
+
+    if not physical_gpu_ids:
+        return []
+
+    os.environ["CUDA_VISIBLE_DEVICES"] = ",".join(
+        str(gpu_id) for gpu_id in physical_gpu_ids
+    )
+    return list(range(len(physical_gpu_ids)))

algorithm_plugin_sdk/log_manager.py

@@ -0,0 +1,76 @@
+from __future__ import annotations
+
+import json
+import logging
+import re
+import threading
+from pathlib import Path
+from typing import Any
+
+from .context import utc_now
+
+_SAFE_NAME = re.compile(r"[^A-Za-z0-9._-]+")
+
+
+def _safe_name(value: str) -> str:
+    cleaned = _SAFE_NAME.sub("_", str(value).strip()).strip("._")
+    return cleaned or "job"
+
+
+class ExecutionLog:
+    """Per-execution, 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:
+        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"
+        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._handler.setLevel(logging.DEBUG)
+        self.logger = logger
+        self.logger.setLevel(logging.DEBUG)
+        self.logger.addHandler(self._handler)
+        self._closed = False
+
+    def write_event(self, event: str, **values: Any) -> None:
+        payload = {"event": event, "timestamp": utc_now(), **values}
+        with self._lock:
+            if self._closed:
+                return
+            self._stream.write(json.dumps(payload, ensure_ascii=False, default=str) + "\n")
+            self._stream.flush()
+
+    def close(self) -> None:
+        with self._lock:
+            if self._closed:
+                return
+            self._closed = True
+            self.logger.removeHandler(self._handler)
+            self._handler.close()
+            self._stream.close()
+
+
+class _FlushFileHandler(logging.Handler):
+    def __init__(self, stream: Any) -> None:
+        super().__init__()
+        self.stream = stream
+        self._lock = threading.RLock()
+
+    def emit(self, record: logging.LogRecord) -> None:
+        try:
+            with self._lock:
+                self.stream.write(self.format(record) + "\n")
+                self.stream.flush()
+        except Exception:
+            self.handleError(record)
+
+    def close(self) -> None:
+        try:
+            self.stream.flush()
+        finally:
+            super().close()

algorithm_plugin_sdk/service.py

@@ -10,12 +10,14 @@ import uuid
 from concurrent.futures import Future, ThreadPoolExecutor
 from contextlib import asynccontextmanager
 from dataclasses import dataclass, field, replace
+from pathlib import Path
 from typing import Any, Callable
 
 from .context import ExecutionContext, ProgressSnapshot, utc_now
 from .errors import (ExecutionCancelled, ExecutionNotFinished,
                      ExecutionNotFound, IdempotencyConflict)
 from .models import AlgorithmRequest, AlgorithmResult
+from .log_manager import ExecutionLog
 from .release import ReleaseManifest
 from .runner import AlgorithmRunner
 from .webui_app import announce_webui, mount_webui
@@ -34,6 +36,8 @@ logger = logging.getLogger(__name__)
 class ExecutionRecord:
     execution_id: str
     request: AlgorithmRequest
+    job_id: str
+    task_id: str
     accepted_at: str
     state: str = "accepted"
     started_at: str | None = None
@@ -41,6 +45,7 @@ class ExecutionRecord:
     result: AlgorithmResult | None = None
     error: str | None = None
     context: ExecutionContext | None = None
+    log: ExecutionLog | None = None
     future: Future[None] | None = None
     attempt: int = 0
     max_attempts: int = 1
@@ -113,8 +118,11 @@ class ExecutionManager:
         self,
         runner: AlgorithmRunner,
         *,
+        default_parameters: dict[str, Any] | None = None,
         max_concurrent_executions: int = 1,
         scratch_dir: str | None = None,
+        repository: str | Path | None = None,
+        log_dir: str | Path | None = None,
         gpu_ids: list[int] | None = None,
         runner_factory: RunnerFactory | None = None,
         max_attempts: int = MAX_ATTEMPTS,
@@ -144,7 +152,15 @@ class ExecutionManager:
         self._runner_factory = runner_factory or (
             lambda: AlgorithmRunner(algorithm_class())
         )
+        if default_parameters is not None and not isinstance(default_parameters, dict):
+            raise ValueError("default_parameters must be a JSON object")
+        self.default_parameters = dict(default_parameters or {})
         self.scratch_dir = scratch_dir
+        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)
+        )
         self.gpu_ids = configured_gpu_ids
         self.max_concurrent_executions = max_concurrent_executions
         self.max_attempts = max_attempts
@@ -180,9 +196,16 @@ class ExecutionManager:
         request: AlgorithmRequest,
         *,
         idempotency_key: str | None = None,
+        job_id: str | None = None,
+        task_id: str | None = None,
     ) -> dict[str, Any]:
         if self.gpu_ids is not None:
             request = replace(request, gpu_ids=self.gpu_ids)
+        if self.default_parameters:
+            request = replace(
+                request,
+                parameters={**self.default_parameters, **request.parameters},
+            )
         self.start()
         canonical_request = json.dumps(
             request.to_dict(), ensure_ascii=False, sort_keys=True, separators=(",", ":")
@@ -201,9 +224,13 @@ class ExecutionManager:
                         "acceptedAt": record.accepted_at,
                     }
         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)
         record = ExecutionRecord(
             execution_id=execution_id,
             request=request,
+            job_id=resolved_job_id,
+            task_id=resolved_task_id,
             accepted_at=utc_now(),
             max_attempts=self.max_attempts,
         )
@@ -217,6 +244,30 @@ class ExecutionManager:
             ),
             progress_callback=record.add_progress,
         )
+        if self.log_dir is not None:
+            record.log = ExecutionLog(
+                log_dir=self.log_dir,
+                job_id=record.job_id,
+                execution_id=record.execution_id,
+                logger=record.context.logger,
+            )
+            workspace = record.request.workspace or {}
+            record.log.write_event(
+                "request",
+                jobId=record.job_id,
+                taskId=record.task_id,
+                job_id=record.job_id,
+                task_id=record.task_id,
+                executionId=record.execution_id,
+                execution_id=record.execution_id,
+                inputs=[item.to_dict() for item in record.request.inputs],
+                inputRoot=workspace.get("inputRoot"),
+                outputRoot=workspace.get("outputRoot"),
+                scratchRoot=workspace.get("scratchRoot") or self.scratch_dir,
+                merge=record.request.merge,
+                parameters=record.request.parameters,
+                gpuIds=record.request.gpu_ids,
+            )
         with self._lock:
             if self._closed:
                 raise RuntimeError("execution manager is closed")
@@ -328,6 +379,9 @@ class ExecutionManager:
             cancelled_before_start = bool(record.future and record.future.cancel())
             if cancelled_before_start:
                 record.set_state("cancelled", error=reason)
+                if record.log is not None:
+                    record.log.write_event("result", state="cancelled", error=reason)
+                    record.log.close()
             return {
                 "executionId": execution_id,
                 "state": record.state,
@@ -349,6 +403,9 @@ class ExecutionManager:
             record.context.cancel("service shutdown")
             if record.future and record.future.cancel():
                 record.set_state("cancelled", error="service shutdown")
+                if record.log is not None:
+                    record.log.write_event("result", state="cancelled", error="service shutdown")
+                    record.log.close()
         self._executor.shutdown(wait=True, cancel_futures=True)
         self.runner.close()
 
@@ -362,17 +419,28 @@ class ExecutionManager:
     def _run(self, record: ExecutionRecord) -> None:
         record.set_state("running")
         assert record.context is not None
+        if record.log is not None:
+            record.log.write_event("started", jobId=record.job_id, taskId=record.task_id)
         try:
             result = self._run_with_restarts(record)
         except ExecutionCancelled as exc:
             record.context.mark_unfinished("cancelled", str(exc))
             record.set_state("cancelled", error=str(exc))
+            if record.log is not None:
+                record.log.write_event("result", state="cancelled", error=str(exc))
         except Exception as exc:
             error = f"{type(exc).__name__}: {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)
         else:
             record.set_state(result.status, result=result)
+            if record.log is not None:
+                record.log.write_event("result", state=record.state, result=result.to_dict())
+        finally:
+            if record.log is not None:
+                record.log.close()
 
     def _run_with_restarts(self, record: ExecutionRecord) -> AlgorithmResult:
         assert record.context is not None
@@ -585,7 +653,12 @@ def create_app(
     async def create_execution(request: Request) -> Any:
         try:
             body = await request.json()
-            execution_request = AlgorithmRequest.from_dict(body)
+            if not isinstance(body, dict):
+                raise ValueError("request body must be a mapping")
+            job_id = next((body.get(name) for name in ("jobId", "job_id") if body.get(name)), None)
+            task_id = next((body.get(name) for name in ("taskId", "task_id") if body.get(name)), None)
+            request_body = {key: value for key, value in body.items() if key not in {"jobId", "job_id", "taskId", "task_id"}}
+            execution_request = AlgorithmRequest.from_dict(request_body)
             if execution_request.workspace is None:
                 raise ValueError("workspace is required")
         except Exception as exc:
@@ -593,9 +666,14 @@ def create_app(
                 status_code=422,
                 content={"error": f"{type(exc).__name__}: {exc}"},
             )
+        idempotency_key = request.headers.get("Idempotency-Key")
+        job_id = job_id or next((request.headers.get(name) for name in ("X-LDP-Job-ID", "X-Job-ID", "Job-ID") if request.headers.get(name)), None)
+        task_id = task_id or next((request.headers.get(name) for name in ("X-LDP-Task-ID", "X-Task-ID", "Task-ID") if request.headers.get(name)), None)
         return manager.submit(
             execution_request,
-            idempotency_key=request.headers.get("Idempotency-Key"),
+            idempotency_key=idempotency_key,
+            job_id=job_id or idempotency_key,
+            task_id=task_id,
         )
 
     @application.get("/v1/executions/{execution_id}")

{ls_algorithm_plugin_sdk-0.3.0.dist-info → ls_algorithm_plugin_sdk-0.3.3.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: ls-algorithm-plugin-sdk
-Version: 0.3.0
+Version: 0.3.3
 Summary: Protocol-independent runtime SDK for dataset algorithms
 Author: Ling Robotics
 License: Proprietary
@@ -15,10 +15,6 @@ Requires-Dist: build>=1.2; extra == "dev"
 
 # Algorithm Plugin SDK
 
-算法仓库只实现一个 `Algorithm` 类并声明 entry point。SDK 统一提供单次运行、
-WebUI、HTTP service、release manifest、compute 注册和心跳。算法仓库不需要
-Dockerfile 或 Web 框架代码。
-
 ## 算法侧最小接口
 
 ```python
@@ -30,8 +26,7 @@ class MyAlgorithm(Algorithm):
     def execute(self, request, context): ...
 ```
 
-算法包需要声明 `algorithm_plugin_sdk.algorithms` entry point。同一个 `execute()`
-会被所有运行方式复用。
+算法包需要声明 `algorithm_plugin_sdk.algorithms` entry point。
 
 ## 单次运行
 
@@ -49,10 +44,9 @@ algorithm-plugin run camera-space-mano \
 
 目标算法仓库必须满足：
 
-- 仓库本身是 Git 工作树根目录，并且至少有一个 commit。
+- 仓库本身是 Git 工作树根目录。
 - 仓库包含自己的 `.venv`。
 - `<repository>/.venv/bin/algorithm-plugin` 存在且可执行。
-- 该启动器的 `run --help` 和 `serve --help` 均可正常加载。
 
 建议将生成的 `release-manifest.json` 和 `algorithm-plugin.json` 加入算法仓库的
 `.gitignore`，不要手动编辑或提交它们。
@@ -64,16 +58,11 @@ sudo /srv/camera-space-mano/.venv/bin/algorithm-plugin configure \
   --repository /srv/camera-space-mano \
   --compute-url http://compute:5180 \
   --api-key "$LDP_INTERNAL_API_KEY" \
-  --public-url http://10.0.0.4:9000 \
-  --input-root /data/jobs/input \
-  --workspace-root /data/jobs/workspace \
+  --port 9030 \
   --cluster production \
-  --namespace camera-space-mano \
-  --node-name 10.0.0.4 \
   --gpu-ids 0,1
 ```
 
-service name 默认使用仓库目录名，也可以通过 `--service-name` 指定。
 重复执行 configure 需要显式提供 `--force`。
 
 确认输出后执行：
@@ -82,7 +71,7 @@ service name 默认使用仓库目录名，也可以通过 `--service-name` 指
 sudo systemctl daemon-reload
 sudo systemctl enable camera-space-mano.service
 sudo systemctl restart camera-space-mano.service
-sudo systemctl status camera-space-mano.service
+sudo systemctl status camera-space-mano.service --no-pager
 ```
 
 ## serve
@@ -95,6 +84,6 @@ sudo systemctl status camera-space-mano.service
 ## 安装与测试
 
 ```bash
-python -m pip install '.[service]'
+python -m pip install .
 python -m unittest discover -s tests -v
 ```

{ls_algorithm_plugin_sdk-0.3.0.dist-info → ls_algorithm_plugin_sdk-0.3.3.dist-info}/RECORD

@@ -2,20 +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=hFX6kJXia2gF2mOCDJlgWK-BAUT6tAvr2t70u32mg1A,13823
+algorithm_plugin_sdk/deployment.py,sha256=nx4-26vs5OqO6miBf6GegZIhdvOgFaw9MdjpTHZNDPo,14583
 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/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=Xj99UiCZtqtu7U4IVghAr4Mu3I9nR1AvP6vNzdD4sAk,24002
+algorithm_plugin_sdk/service.py,sha256=HcO8PmtPczogi0an7irajsRCD-11GAH8pnE7rOHOLSM,28004
 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=acCGDJ6Khn3G8hay8ISsYVlKDmhDH1i-UpNZU5bb0fU,11668
+algorithm_plugin_sdk/cli_impl/configure.py,sha256=OlYuNnPZXHbaoWKfBpFivyP8lZk2QQpvMN572hAjriA,13353
 algorithm_plugin_sdk/cli_impl/parsing.py,sha256=TDX45Spu4WMrczuzTMN9ZC2SYyHE11OPBCvN45yPsxs,1399
-algorithm_plugin_sdk/cli_impl/run.py,sha256=EnqqCdGWYK-iB5s1DMlGRXNpnCWA25xoAex0f4l36Ic,5893
-algorithm_plugin_sdk/cli_impl/serve.py,sha256=9_nJcdzerwqczGm7xTIcMYKShpmoltYD6r_nBQjAlWU,4324
+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/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
@@ -23,8 +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-0.3.0.dist-info/METADATA,sha256=YewSjuTdWCstRq4c6h4-fy-7pSbs31JRoui_J5GhV7M,2943
-ls_algorithm_plugin_sdk-0.3.0.dist-info/WHEEL,sha256=YVMoNqKzERt-wjUZwJ33xBGAwnFl-4cqbYkTtWa4itE,91
-ls_algorithm_plugin_sdk-0.3.0.dist-info/entry_points.txt,sha256=TU0R_TxuB5OuOiZ5BHOEmUPtHvN3v8ARUW8d3PztHcE,67
-ls_algorithm_plugin_sdk-0.3.0.dist-info/top_level.txt,sha256=8lsgxZ8HJGLlzJ5Rt8CiVzw9Ccme286i3akbmeMjBos,21
-ls_algorithm_plugin_sdk-0.3.0.dist-info/RECORD,,
+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,,