ls-algorithm-plugin-sdk 0.3.1__py3-none-any.whl → 0.3.4__py3-none-any.whl

algorithm_plugin_sdk/cli_impl/configure.py

@@ -16,7 +16,7 @@ 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")
@@ -56,14 +56,34 @@ 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(
+        "--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,
         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",
@@ -139,10 +159,18 @@ 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]",
@@ -153,12 +181,13 @@ def render_systemd_unit(
             "[Service]",
             "Type=simple",
             "User=root",
-            f"WorkingDirectory={_escape_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",
             "",
@@ -376,6 +405,11 @@ 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,
@@ -383,12 +417,15 @@ 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,
+        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,
         environment=dict(args.env),
+        default_parameters=args.default_parameters,
     )
     config.save()
     return config
@@ -407,6 +444,7 @@ def execute(args: argparse.Namespace) -> int:
         repository=repository,
         config_path=config_path,
         service_name=service_name,
+        gpu_ids=args.gpu_ids,
     )
 
     inspect_systemd_unit(
@@ -414,6 +452,7 @@ 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,

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
@@ -78,15 +79,22 @@ 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:
-        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"])
         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()
@@ -96,6 +104,11 @@ 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)
 
@@ -105,14 +118,21 @@ 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,
+        local_scratch=local_scratch,
+        local_output=local_output,
         gpu_ids=args.gpu_ids,
         runner_factory=create_runner,
         max_attempts=args.max_attempts,
         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 +145,9 @@ 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,
+        local_scratch: bool = False,
+        local_output: bool = False,
     ) -> "DeploymentConfig":
         repository = repository_root(root)
         _make_repository_importable(repository)
@@ -188,8 +191,11 @@ 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 {}),
             },
             "registration": registration,
             "environment": dict(sorted((environment or {}).items())),
@@ -266,6 +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
@@ -273,6 +282,16 @@ 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 +1,125 @@
+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._-]+")
+_TIMESTAMP_GLOB = "[0-9]" * 14
+_FILE_LOCKS_GUARD = threading.Lock()
+_FILE_LOCKS: dict[Path, threading.RLock] = {}
+
+
+def _safe_name(value: str) -> str:
+    cleaned = _SAFE_NAME.sub("_", str(value).strip()).strip("._")
+    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-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,
+        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 = _job_log_path(directory, job_id)
+        self.path = path
+        self._stream = path.open("a", encoding="utf-8", buffering=1)
+        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)
+        self.logger.addHandler(self._handler)
+        self._closed = False
+
+    def write_event(self, event: str, **values: Any) -> None:
+        payload = {
+            "event": event,
+            "timestamp": utc_now(),
+            **self._event_context,
+            **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, lock: threading.RLock) -> None:
+        super().__init__()
+        self.stream = stream
+        self._lock = lock
+
+    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

@@ -6,16 +6,19 @@ import logging
 import random
 import threading
 import time
+import traceback
 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 +37,8 @@ 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 +46,7 @@ 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 +119,13 @@ 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,
+        local_scratch: bool = False,
+        local_output: bool = False,
         gpu_ids: list[int] | None = None,
         runner_factory: RunnerFactory | None = None,
         max_attempts: int = MAX_ATTEMPTS,
@@ -130,6 +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(
@@ -144,7 +166,24 @@ 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._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 (repository_path / "logs" if repository_path is not None else None)
+        )
         self.gpu_ids = configured_gpu_ids
         self.max_concurrent_executions = max_concurrent_executions
         self.max_attempts = max_attempts
@@ -175,14 +214,61 @@ 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,
         *,
         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=(",", ":")
@@ -201,22 +287,51 @@ 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)
+        request = self._apply_local_paths(request, resolved_job_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,
         )
         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,
         )
+        if self.log_dir is not None:
+            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,
+            )
+            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 +443,9 @@ 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 +467,9 @@ 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 +483,34 @@ 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}"
+            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,
+                    traceback=stack_trace,
+                )
         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
@@ -398,7 +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,
@@ -407,16 +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()
@@ -585,7 +725,12 @@ 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 +738,14 @@ 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-0.3.1.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: 0.3.1
+Version: 0.3.4
 Summary: Protocol-independent runtime SDK for dataset algorithms
 Author: Ling Robotics
 License: Proprietary

{ls_algorithm_plugin_sdk-0.3.1.dist-info → ls_algorithm_plugin_sdk-0.3.4.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=yv6UCMtbX1pglfq7H7N2xI4_-bM-YLSPXFST5m_pW-g,13931
+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=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=Xj99UiCZtqtu7U4IVghAr4Mu3I9nR1AvP6vNzdD4sAk,24002
+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=fr_xDLh82YKV8Csuxi5M2z4dPpXWv9j7SZUHa1n14dA,12496
+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=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=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
@@ -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-0.3.1.dist-info/METADATA,sha256=m3-0HP49-i6sFKAgZjcsWLMGdUdVL-jLUag4siJ9i3Q,2317
-ls_algorithm_plugin_sdk-0.3.1.dist-info/WHEEL,sha256=YVMoNqKzERt-wjUZwJ33xBGAwnFl-4cqbYkTtWa4itE,91
-ls_algorithm_plugin_sdk-0.3.1.dist-info/entry_points.txt,sha256=TU0R_TxuB5OuOiZ5BHOEmUPtHvN3v8ARUW8d3PztHcE,67
-ls_algorithm_plugin_sdk-0.3.1.dist-info/top_level.txt,sha256=8lsgxZ8HJGLlzJ5Rt8CiVzw9Ccme286i3akbmeMjBos,21
-ls_algorithm_plugin_sdk-0.3.1.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,,