fractal-server 1.4.6__py3-none-any.whl → 2.0.0__py3-none-any.whl

fractal_server/app/runner/{_slurm → executors/slurm}/_slurm_config.py

@@ -22,10 +22,9 @@ from pydantic import Extra
 from pydantic import Field
 from pydantic.error_wrappers import ValidationError
 
-from ....config import get_settings
-from ....logger import set_logger
-from ....syringe import Inject
-from ...models import WorkflowTask
+from .....config import get_settings
+from .....logger import set_logger
+from .....syringe import Inject
 
 logger = set_logger(__name__)
 
@@ -459,151 +458,3 @@ def get_default_slurm_config():
         target_num_jobs=2,
         max_num_jobs=4,
     )
-
-
-def get_slurm_config(
-    wftask: WorkflowTask,
-    workflow_dir: Path,
-    workflow_dir_user: Path,
-    config_path: Optional[Path] = None,
-) -> SlurmConfig:
-    """
-    Prepare a `SlurmConfig` configuration object
-
-    The sources for `SlurmConfig` attributes, in increasing priority order, are
-
-    1. The general content of the Fractal SLURM configuration file.
-    2. The GPU-specific content of the Fractal SLURM configuration file, if
-        appropriate.
-    3. Properties in `wftask.meta` (which, for `WorkflowTask`s added through
-       `Workflow.insert_task`, also includes `wftask.task.meta`);
-
-    Note: `wftask.meta` may be `None`.
-
-    Arguments:
-        wftask:
-            WorkflowTask for which the SLURM configuration is is to be
-            prepared.
-        workflow_dir:
-            Server-owned directory to store all task-execution-related relevant
-            files (inputs, outputs, errors, and all meta files related to the
-            job execution). Note: users cannot write directly to this folder.
-        workflow_dir_user:
-            User-side directory with the same scope as `workflow_dir`, and
-            where a user can write.
-        config_path:
-            Path of aFractal  SLURM configuration file; if `None`, use
-            `FRACTAL_SLURM_CONFIG_FILE` variable from settings.
-
-    Returns:
-        slurm_config:
-            The SlurmConfig object
-    """
-
-    logger.debug(
-        "[get_slurm_config] WorkflowTask meta attribute: {wftask.meta=}"
-    )
-
-    # Incorporate slurm_env.default_slurm_config
-    slurm_env = load_slurm_config_file(config_path=config_path)
-    slurm_dict = slurm_env.default_slurm_config.dict(
-        exclude_unset=True, exclude={"mem"}
-    )
-    if slurm_env.default_slurm_config.mem:
-        slurm_dict["mem_per_task_MB"] = slurm_env.default_slurm_config.mem
-
-    # Incorporate slurm_env.batching_config
-    for key, value in slurm_env.batching_config.dict().items():
-        slurm_dict[key] = value
-
-    # Incorporate slurm_env.user_local_exports
-    slurm_dict["user_local_exports"] = slurm_env.user_local_exports
-
-    logger.debug(
-        "[get_slurm_config] Fractal SLURM configuration file: "
-        f"{slurm_env.dict()=}"
-    )
-
-    # GPU-related options
-    # Notes about priority:
-    # 1. This block of definitions takes priority over other definitions from
-    #    slurm_env which are not under the `needs_gpu` subgroup
-    # 2. This block of definitions has lower priority than whatever comes next
-    #    (i.e. from WorkflowTask.meta).
-    if wftask.meta is not None:
-        needs_gpu = wftask.meta.get("needs_gpu", False)
-    else:
-        needs_gpu = False
-    logger.debug(f"[get_slurm_config] {needs_gpu=}")
-    if needs_gpu:
-        for key, value in slurm_env.gpu_slurm_config.dict(
-            exclude_unset=True, exclude={"mem"}
-        ).items():
-            slurm_dict[key] = value
-        if slurm_env.gpu_slurm_config.mem:
-            slurm_dict["mem_per_task_MB"] = slurm_env.gpu_slurm_config.mem
-
-    # Number of CPUs per task, for multithreading
-    if wftask.meta is not None and "cpus_per_task" in wftask.meta:
-        cpus_per_task = int(wftask.meta["cpus_per_task"])
-        slurm_dict["cpus_per_task"] = cpus_per_task
-
-    # Required memory per task, in MB
-    if wftask.meta is not None and "mem" in wftask.meta:
-        raw_mem = wftask.meta["mem"]
-        mem_per_task_MB = _parse_mem_value(raw_mem)
-        slurm_dict["mem_per_task_MB"] = mem_per_task_MB
-
-    # Job name
-    job_name = wftask.task.name.replace(" ", "_")
-    slurm_dict["job_name"] = job_name
-
-    # Optional SLURM arguments and extra lines
-    if wftask.meta is not None:
-        account = wftask.meta.get("account", None)
-        if account is not None:
-            error_msg = (
-                f"Invalid {account=} property in WorkflowTask `meta` "
-                "attribute.\n"
-                "SLURM account must be set in the request body of the "
-                "apply-workflow endpoint, or by modifying the user properties."
-            )
-            logger.error(error_msg)
-            raise SlurmConfigError(error_msg)
-        for key in ["time", "gres", "constraint"]:
-            value = wftask.meta.get(key, None)
-            if value:
-                slurm_dict[key] = value
-    if wftask.meta is not None:
-        extra_lines = wftask.meta.get("extra_lines", [])
-    else:
-        extra_lines = []
-    extra_lines = slurm_dict.get("extra_lines", []) + extra_lines
-    if len(set(extra_lines)) != len(extra_lines):
-        logger.debug(
-            "[get_slurm_config] Removing repeated elements "
-            f"from {extra_lines=}."
-        )
-        extra_lines = list(set(extra_lines))
-    slurm_dict["extra_lines"] = extra_lines
-
-    # Job-batching parameters (if None, they will be determined heuristically)
-    if wftask.meta is not None:
-        tasks_per_job = wftask.meta.get("tasks_per_job", None)
-        parallel_tasks_per_job = wftask.meta.get(
-            "parallel_tasks_per_job", None
-        )
-    else:
-        tasks_per_job = None
-        parallel_tasks_per_job = None
-    slurm_dict["tasks_per_job"] = tasks_per_job
-    slurm_dict["parallel_tasks_per_job"] = parallel_tasks_per_job
-
-    # Put everything together
-    logger.debug(
-        "[get_slurm_config] Now create a SlurmConfig object based "
-        f"on {slurm_dict=}"
-    )
-    slurm_config = SlurmConfig(**slurm_dict)
-
-    return slurm_config

fractal_server/app/runner/{_slurm → executors/slurm}/_subprocess_run_as_user.py

@@ -19,7 +19,7 @@ import shlex
 import subprocess  # nosec
 from typing import Optional
 
-from ....logger import set_logger
+from .....logger import set_logger
 
 logger = set_logger(__name__)
 
@@ -113,6 +113,47 @@ def _glob_as_user(
     return output
 
 
+def _glob_as_user_strict(
+    *,
+    folder: str,
+    user: str,
+    startswith: str,
+) -> list[str]:
+    """
+    Run `ls` in a folder (as a user) and filter results
+
+    Execute `ls` on a folder (impersonating a user, if `user` is not `None`)
+    and select results that comply with a set of rules. They all start with
+    `startswith` (if not `None`), and they match one of the known filename
+    patterns. See details in
+    https://github.com/fractal-analytics-platform/fractal-server/issues/1240
+
+
+    Arguments:
+        folder: Absolute path to the folder
+        user: If not `None`, the user to be impersonated via `sudo -u`
+        startswith: If not `None`, this is used to filter output of `ls`.
+    """
+
+    res = _run_command_as_user(cmd=f"ls {folder}", user=user, check=True)
+    output = res.stdout.split()
+
+    new_output = []
+    known_filenames = [
+        f"{startswith}{suffix}"
+        for suffix in [".args.json", ".metadiff.json", ".err", ".out"]
+    ]
+    for filename in output:
+        if filename in known_filenames:
+            new_output.append(filename)
+        elif filename.startswith(f"{startswith}_out_") and filename.endswith(
+            ".pickle"
+        ):
+            new_output.append(filename)
+
+    return new_output
+
+
 def _path_exists_as_user(*, path: str, user: Optional[str] = None) -> bool:
     """
     Impersonate a user and check if `path` exists via `ls`

fractal_server/app/runner/{_slurm → executors/slurm}/executor.py

@@ -29,22 +29,24 @@ import cloudpickle
 from cfut import SlurmExecutor
 from cfut.util import random_string
 
-from ....config import get_settings
-from ....logger import set_logger
-from ....syringe import Inject
-from .._common import get_task_file_paths
-from .._common import SHUTDOWN_FILENAME
-from .._common import TaskFiles
-from ..common import JobExecutionError
-from ..common import TaskExecutionError
+from .....config import get_settings
+from .....logger import set_logger
+from .....syringe import Inject
+from ...exceptions import JobExecutionError
+from ...exceptions import TaskExecutionError
+from ...filenames import SHUTDOWN_FILENAME
+from ...task_files import get_task_file_paths
+from ...task_files import TaskFiles
 from ._batching import heuristics
 from ._executor_wait_thread import FractalSlurmWaitThread
 from ._slurm_config import get_default_slurm_config
 from ._slurm_config import SlurmConfig
 from ._subprocess_run_as_user import _glob_as_user
+from ._subprocess_run_as_user import _glob_as_user_strict
 from ._subprocess_run_as_user import _path_exists_as_user
 from ._subprocess_run_as_user import _run_command_as_user
 from fractal_server import __VERSION__
+from fractal_server.app.runner.components import _COMPONENT_KEY_
 
 
 logger = set_logger(__name__)
@@ -415,8 +417,6 @@ class FractalSlurmExecutor(SlurmExecutor):
                 A `TaskFiles` object; if `None`, use
                 `self.get_default_task_files()`.
 
-        Returns:
-            An iterator of results.
         """
 
         def _result_or_cancel(fut):
@@ -543,7 +543,7 @@ class FractalSlurmExecutor(SlurmExecutor):
         single_task_submission: bool = False,
         args: Optional[Sequence[Any]] = None,
         kwargs: Optional[dict] = None,
-        components: list[Any] = None,
+        components: Optional[list[Any]] = None,
     ) -> Future:
         """
         Submit a multi-task job to the pool, where each task is handled via the
@@ -579,6 +579,10 @@ class FractalSlurmExecutor(SlurmExecutor):
 
         # Define slurm-job-related files
         if single_task_submission:
+            if components is not None:
+                raise ValueError(
+                    f"{single_task_submission=} but components is not None"
+                )
             job = SlurmJob(
                 slurm_file_prefix=slurm_file_prefix,
                 num_tasks_tot=1,
@@ -602,15 +606,23 @@ class FractalSlurmExecutor(SlurmExecutor):
                 num_tasks_tot=num_tasks_tot,
                 slurm_config=slurm_config,
             )
-            job.wftask_file_prefixes = tuple(
-                get_task_file_paths(
-                    workflow_dir=task_files.workflow_dir,
-                    workflow_dir_user=task_files.workflow_dir_user,
-                    task_order=task_files.task_order,
-                    component=component,
-                ).file_prefix
-                for component in components
-            )
+
+            _prefixes = []
+            for component in components:
+                if isinstance(component, dict):
+                    # This is needed for V2
+                    actual_component = component.get(_COMPONENT_KEY_, None)
+                else:
+                    actual_component = component
+                _prefixes.append(
+                    get_task_file_paths(
+                        workflow_dir=task_files.workflow_dir,
+                        workflow_dir_user=task_files.workflow_dir_user,
+                        task_order=task_files.task_order,
+                        component=actual_component,
+                    ).file_prefix
+                )
+            job.wftask_file_prefixes = tuple(_prefixes)
 
         # Define I/O pickle file names/paths
         job.input_pickle_files = tuple(
@@ -930,11 +942,19 @@ class FractalSlurmExecutor(SlurmExecutor):
 
         for prefix in prefixes:
 
-            files_to_copy = _glob_as_user(
-                folder=str(self.working_dir_user),
-                user=self.slurm_user,
-                startswith=prefix,
-            )
+            if prefix == job.slurm_file_prefix:
+                files_to_copy = _glob_as_user(
+                    folder=str(self.working_dir_user),
+                    user=self.slurm_user,
+                    startswith=prefix,
+                )
+            else:
+                files_to_copy = _glob_as_user_strict(
+                    folder=str(self.working_dir_user),
+                    user=self.slurm_user,
+                    startswith=prefix,
+                )
+
             logger.debug(
                 "[_copy_files_from_user_to_server] "
                 f"{prefix=}, {len(files_to_copy)=}"
@@ -949,7 +969,6 @@ class FractalSlurmExecutor(SlurmExecutor):
                 source_file_path = str(
                     self.working_dir_user / source_file_name
                 )
-                dest_file_path = str(self.working_dir / source_file_name)
 
                 # Read source_file_path (requires sudo)
                 # NOTE: By setting encoding=None, we read/write bytes instead
@@ -993,7 +1012,7 @@ class FractalSlurmExecutor(SlurmExecutor):
             cmdlines.append(
                 (
                     f"{python_worker_interpreter}"
-                    " -m fractal_server.app.runner._slurm.remote "
+                    " -m fractal_server.app.runner.executors.slurm.remote "
                     f"--input-file {input_pickle_file} "
                     f"--output-file {output_pickle_file}"
                 )

fractal_server/app/runner/filenames.py

@@ -0,0 +1,6 @@
+HISTORY_FILENAME = "history.json"
+FILTERS_FILENAME = "filters.json"
+IMAGES_FILENAME = "images.json"
+METADATA_FILENAME = "metadata.json"
+SHUTDOWN_FILENAME = "shutdown"
+WORKFLOW_LOG_FILENAME = "workflow.log"

fractal_server/app/runner/set_start_and_last_task_index.py

@@ -0,0 +1,39 @@
+from typing import Optional
+
+
+def set_start_and_last_task_index(
+    num_tasks: int,
+    first_task_index: Optional[int] = None,
+    last_task_index: Optional[int] = None,
+) -> tuple[int, int]:
+    """
+    Handle `first_task_index` and `last_task_index`, by setting defaults and
+    validating values.
+
+    num_tasks:
+        Total number of tasks in a workflow task list
+    first_task_index:
+        Positional index of the first task to execute
+    last_task_index:
+        Positional index of the last task to execute
+    """
+    # Set default values
+    if first_task_index is None:
+        first_task_index = 0
+    if last_task_index is None:
+        last_task_index = num_tasks - 1
+
+    # Perform checks
+    if first_task_index < 0:
+        raise ValueError(f"{first_task_index=} cannot be negative")
+    if last_task_index < 0:
+        raise ValueError(f"{last_task_index=} cannot be negative")
+    if last_task_index > num_tasks - 1:
+        raise ValueError(
+            f"{last_task_index=} cannot be larger than {(num_tasks-1)=}"
+        )
+    if first_task_index > last_task_index:
+        raise ValueError(
+            f"{first_task_index=} cannot be larger than {last_task_index=}"
+        )
+    return (first_task_index, last_task_index)

fractal_server/app/runner/task_files.py

@@ -0,0 +1,103 @@
+from pathlib import Path
+from typing import Optional
+
+
+def sanitize_component(value: str) -> str:
+    """
+    Remove {" ", "/", "."} form a string, e.g. going from
+    'plate.zarr/B/03/0' to 'plate_zarr_B_03_0'.
+    """
+    return value.replace(" ", "_").replace("/", "_").replace(".", "_")
+
+
+class TaskFiles:
+    """
+    Group all file paths pertaining to a task
+
+    Attributes:
+        workflow_dir:
+            Server-owned directory to store all task-execution-related relevant
+            files (inputs, outputs, errors, and all meta files related to the
+            job execution). Note: users cannot write directly to this folder.
+        workflow_dir_user:
+            User-side directory with the same scope as `workflow_dir`, and
+            where a user can write.
+        task_order:
+            Positional order of the task within a workflow.
+        component:
+            Specific component to run the task for (relevant for tasks that
+            will be executed in parallel over many components).
+        file_prefix:
+            Prefix for all task-related files.
+        args:
+            Path for input json file.
+        metadiff:
+            Path for output json file with metadata update.
+        out:
+            Path for task-execution stdout.
+        err:
+            Path for task-execution stderr.
+    """
+
+    workflow_dir: Path
+    workflow_dir_user: Path
+    task_order: Optional[int] = None
+    component: Optional[str] = None
+
+    file_prefix: str
+    args: Path
+    out: Path
+    err: Path
+    log: Path
+    metadiff: Path
+
+    def __init__(
+        self,
+        workflow_dir: Path,
+        workflow_dir_user: Path,
+        task_order: Optional[int] = None,
+        component: Optional[str] = None,
+    ):
+        self.workflow_dir = workflow_dir
+        self.workflow_dir_user = workflow_dir_user
+        self.task_order = task_order
+        self.component = component
+
+        if self.component is not None:
+            component_safe = sanitize_component(str(self.component))
+            component_safe = f"_par_{component_safe}"
+        else:
+            component_safe = ""
+
+        if self.task_order is not None:
+            order = str(self.task_order)
+        else:
+            order = "task"
+        self.file_prefix = f"{order}{component_safe}"
+        self.args = self.workflow_dir_user / f"{self.file_prefix}.args.json"
+        self.out = self.workflow_dir_user / f"{self.file_prefix}.out"
+        self.err = self.workflow_dir_user / f"{self.file_prefix}.err"
+        self.log = self.workflow_dir_user / f"{self.file_prefix}.log"
+        self.metadiff = (
+            self.workflow_dir_user / f"{self.file_prefix}.metadiff.json"
+        )
+
+
+def get_task_file_paths(
+    workflow_dir: Path,
+    workflow_dir_user: Path,
+    task_order: Optional[int] = None,
+    component: Optional[str] = None,
+) -> TaskFiles:
+    """
+    Return the corrisponding TaskFiles object
+
+    This function is mainly used as a cache to avoid instantiating needless
+    objects.
+    """
+    return TaskFiles(
+        workflow_dir=workflow_dir,
+        workflow_dir_user=workflow_dir_user,
+        task_order=task_order,
+        component=component,
+    )