fractal-server 2.3.4__tar.gz → 2.3.5__tar.gz

{fractal_server-2.3.4 → fractal_server-2.3.5}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: fractal-server
-Version: 2.3.4
+Version: 2.3.5
 Summary: Server component of the Fractal analytics platform
 Home-page: https://github.com/fractal-analytics-platform/fractal-server
 License: BSD-3-Clause

fractal_server-2.3.5/fractal_server/__init__.py

@@ -0,0 +1 @@
+__VERSION__ = "2.3.5"

{fractal_server-2.3.4 → fractal_server-2.3.5}/fractal_server/app/runner/executors/slurm/_slurm_config.py

@@ -62,6 +62,8 @@ class _SlurmConfigSet(BaseModel, extra=Extra.forbid):
     time: Optional[str]
     account: Optional[str]
     extra_lines: Optional[list[str]]
+    pre_submission_commands: Optional[list[str]]
+    gpus: Optional[str]
 
 
 class _BatchingConfigSet(BaseModel, extra=Extra.forbid):
@@ -219,6 +221,7 @@ class SlurmConfig(BaseModel, extra=Extra.forbid):
         constraint: Corresponds to SLURM option.
         gres: Corresponds to SLURM option.
         account: Corresponds to SLURM option.
+        gpus: Corresponds to SLURM option.
         time: Corresponds to SLURM option (WARNING: not fully supported).
         prefix: Prefix of configuration lines in SLURM submission scripts.
         shebang_line: Shebang line for SLURM submission scripts.
@@ -240,6 +243,8 @@ class SlurmConfig(BaseModel, extra=Extra.forbid):
             Key-value pairs to be included as `export`-ed variables in SLURM
             submission script, after prepending values with the user's cache
             directory.
+        pre_submission_commands: List of commands to be prepended to the sbatch
+            command.
     """
 
     # Required SLURM parameters (note that the integer attributes are those
@@ -254,6 +259,7 @@ class SlurmConfig(BaseModel, extra=Extra.forbid):
     job_name: Optional[str] = None
     constraint: Optional[str] = None
     gres: Optional[str] = None
+    gpus: Optional[str] = None
     time: Optional[str] = None
     account: Optional[str] = None
 
@@ -274,6 +280,8 @@ class SlurmConfig(BaseModel, extra=Extra.forbid):
     target_num_jobs: int
     max_num_jobs: int
 
+    pre_submission_commands: list[str] = Field(default_factory=list)
+
     def _sorted_extra_lines(self) -> list[str]:
         """
         Return a copy of `self.extra_lines`, where lines starting with
@@ -340,7 +348,14 @@ class SlurmConfig(BaseModel, extra=Extra.forbid):
             f"{self.prefix} --cpus-per-task={self.cpus_per_task}",
             f"{self.prefix} --mem={mem_per_job_MB}M",
         ]
-        for key in ["job_name", "constraint", "gres", "time", "account"]:
+        for key in [
+            "job_name",
+            "constraint",
+            "gres",
+            "gpus",
+            "time",
+            "account",
+        ]:
             value = getattr(self, key)
             if value is not None:
                 # Handle the `time` parameter

{fractal_server-2.3.4 → fractal_server-2.3.5}/fractal_server/app/runner/executors/slurm/ssh/executor.py

@@ -869,9 +869,22 @@ class FractalSlurmSSHExecutor(SlurmExecutor):
 
         # Submit job to SLURM, and get jobid
         sbatch_command = f"sbatch --parsable {job.slurm_script_remote}"
-        sbatch_stdout = self.fractal_ssh.run_command(
-            cmd=sbatch_command,
-        )
+        pre_submission_cmds = job.slurm_config.pre_submission_commands
+        if len(pre_submission_cmds) == 0:
+            sbatch_stdout = self.fractal_ssh.run_command(cmd=sbatch_command)
+        else:
+            logger.debug(f"Now using {pre_submission_cmds=}")
+            script_lines = pre_submission_cmds + [sbatch_command]
+            script_content = "\n".join(script_lines)
+            script_content = f"{script_content}\n"
+            script_path_remote = (
+                f"{job.slurm_script_remote.as_posix()}_wrapper.sh"
+            )
+            self.fractal_ssh.write_remote_file(
+                path=script_path_remote, content=script_content
+            )
+            cmd = f"bash {script_path_remote}"
+            sbatch_stdout = self.fractal_ssh.run_command(cmd=cmd)
 
         # Extract SLURM job ID from stdout
         try:
@@ -881,7 +894,9 @@ class FractalSlurmSSHExecutor(SlurmExecutor):
             error_msg = (
                 f"Submit command `{sbatch_command}` returned "
                 f"`{stdout=}` which cannot be cast to an integer "
-                f"SLURM-job ID. Original error:\n{str(e)}"
+                f"SLURM-job ID.\n"
+                f"Note that {pre_submission_cmds=}.\n"
+                f"Original error:\n{str(e)}"
             )
             logger.error(error_msg)
             raise JobExecutionError(info=error_msg)

{fractal_server-2.3.4 → fractal_server-2.3.5}/fractal_server/app/runner/executors/slurm/sudo/executor.py

@@ -1121,6 +1121,12 @@ class FractalSlurmExecutor(SlurmExecutor):
             slurm_err_path=str(job.slurm_stderr),
         )
 
+        # Print warning for ignored parameter
+        if len(job.slurm_config.pre_submission_commands) > 0:
+            logger.warning(
+                f"Ignoring {job.slurm_config.pre_submission_commands=}."
+            )
+
         # Submit job via sbatch, and retrieve jobid
 
         # Write script content to a job.slurm_script

{fractal_server-2.3.4 → fractal_server-2.3.5}/fractal_server/app/runner/v2/__init__.py

@@ -36,8 +36,8 @@ from ._local import process_workflow as local_process_workflow
 from ._local_experimental import (
     process_workflow as local_experimental_process_workflow,
 )
-from ._slurm import process_workflow as slurm_sudo_process_workflow
 from ._slurm_ssh import process_workflow as slurm_ssh_process_workflow
+from ._slurm_sudo import process_workflow as slurm_sudo_process_workflow
 from .handle_failed_job import assemble_filters_failed_job
 from .handle_failed_job import assemble_history_failed_job
 from .handle_failed_job import assemble_images_failed_job

{fractal_server-2.3.4/fractal_server/app/runner/v2/_slurm_ssh → fractal_server-2.3.5/fractal_server/app/runner/v2/_slurm_common}/get_slurm_config.py

@@ -18,8 +18,6 @@ from fractal_server.app.runner.executors.slurm._slurm_config import (
 
 def get_slurm_config(
     wftask: WorkflowTaskV2,
-    workflow_dir_local: Path,
-    workflow_dir_remote: Path,
     which_type: Literal["non_parallel", "parallel"],
     config_path: Optional[Path] = None,
 ) -> SlurmConfig:
@@ -43,13 +41,6 @@ def get_slurm_config(
         wftask:
             WorkflowTask for which the SLURM configuration is is to be
             prepared.
-        workflow_dir_local:
-            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_remote:
-            User-side directory with the same scope as `workflow_dir_local`,
-            and where a user can write.
         config_path:
             Path of a Fractal SLURM configuration file; if `None`, use
             `FRACTAL_SLURM_CONFIG_FILE` variable from settings.
@@ -99,13 +90,13 @@ def get_slurm_config(
     # 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).
+    #    (i.e. from WorkflowTask.meta_parallel).
     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 and slurm_env.gpu_slurm_config is not None:  # FIXME
+    if needs_gpu:
         for key, value in slurm_env.gpu_slurm_config.dict(
             exclude_unset=True, exclude={"mem"}
         ).items():
@@ -143,9 +134,9 @@ def get_slurm_config(
             )
             logger.error(error_msg)
             raise SlurmConfigError(error_msg)
-        for key in ["time", "gres", "constraint"]:
+        for key in ["time", "gres", "gpus", "constraint"]:
             value = wftask_meta.get(key, None)
-            if value:
+            if value is not None:
                 slurm_dict[key] = value
     if wftask_meta is not None:
         extra_lines = wftask_meta.get("extra_lines", [])

{fractal_server-2.3.4 → fractal_server-2.3.5}/fractal_server/app/runner/v2/_slurm_ssh/_submit_setup.py

@@ -17,8 +17,10 @@ from pathlib import Path
 from typing import Literal
 
 from ...task_files import get_task_file_paths
-from .get_slurm_config import get_slurm_config
 from fractal_server.app.models.v2 import WorkflowTaskV2
+from fractal_server.app.runner.v2._slurm_common.get_slurm_config import (
+    get_slurm_config,
+)
 
 
 def _slurm_submit_setup(
@@ -62,8 +64,6 @@ def _slurm_submit_setup(
     # Get SlurmConfig object
     slurm_config = get_slurm_config(
         wftask=wftask,
-        workflow_dir_local=workflow_dir_local,
-        workflow_dir_remote=workflow_dir_remote,
         which_type=which_type,
     )
 

{fractal_server-2.3.4/fractal_server/app/runner/v2/_slurm → fractal_server-2.3.5/fractal_server/app/runner/v2/_slurm_sudo}/_submit_setup.py

@@ -17,8 +17,10 @@ from pathlib import Path
 from typing import Literal
 
 from ...task_files import get_task_file_paths
-from .get_slurm_config import get_slurm_config
 from fractal_server.app.models.v2 import WorkflowTaskV2
+from fractal_server.app.runner.v2._slurm_common.get_slurm_config import (
+    get_slurm_config,
+)
 
 
 def _slurm_submit_setup(
@@ -62,8 +64,6 @@ def _slurm_submit_setup(
     # Get SlurmConfig object
     slurm_config = get_slurm_config(
         wftask=wftask,
-        workflow_dir_local=workflow_dir_local,
-        workflow_dir_remote=workflow_dir_remote,
         which_type=which_type,
     )
 

{fractal_server-2.3.4 → fractal_server-2.3.5}/fractal_server/ssh/_fabric.py

@@ -306,6 +306,28 @@ class FractalSSH(object):
             cmd = f"rm -r {folder}"
             self.run_command(cmd=cmd)
 
+    def write_remote_file(
+        self,
+        *,
+        path: str,
+        content: str,
+        lock_timeout: Optional[float] = None,
+    ) -> None:
+        """
+        Open a remote file via SFTP and write it.
+
+        Args:
+            path: Absolute path
+            contents: File contents
+            lock_timeout:
+        """
+        actual_lock_timeout = self.default_lock_timeout
+        if lock_timeout is not None:
+            actual_lock_timeout = lock_timeout
+        with self.acquire_timeout(timeout=actual_lock_timeout):
+            with self.sftp().open(filename=path, mode="w") as f:
+                f.write(content)
+
 
 def get_ssh_connection(
     *,

{fractal_server-2.3.4 → fractal_server-2.3.5}/pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "fractal-server"
-version = "2.3.4"
+version = "2.3.5"
 description = "Server component of the Fractal analytics platform"
 authors = [
     "Tommaso Comparin <tommaso.comparin@exact-lab.it>",
@@ -91,7 +91,7 @@ filterwarnings = [
 ]
 
 [tool.bumpver]
-current_version = "2.3.4"
+current_version = "2.3.5"
 version_pattern = "MAJOR.MINOR.PATCH[PYTAGNUM]"
 commit_message = "bump version {old_version} -> {new_version}"
 commit = true

fractal_server-2.3.4/fractal_server/__init__.py

@@ -1 +0,0 @@
-__VERSION__ = "2.3.4"

fractal_server-2.3.4/fractal_server/app/runner/v2/_slurm/get_slurm_config.py

@@ -1,182 +0,0 @@
-from pathlib import Path
-from typing import Literal
-from typing import Optional
-
-from fractal_server.app.models.v2 import WorkflowTaskV2
-from fractal_server.app.runner.executors.slurm._slurm_config import (
-    _parse_mem_value,
-)
-from fractal_server.app.runner.executors.slurm._slurm_config import (
-    load_slurm_config_file,
-)
-from fractal_server.app.runner.executors.slurm._slurm_config import logger
-from fractal_server.app.runner.executors.slurm._slurm_config import SlurmConfig
-from fractal_server.app.runner.executors.slurm._slurm_config import (
-    SlurmConfigError,
-)
-
-
-def get_slurm_config(
-    wftask: WorkflowTaskV2,
-    workflow_dir_local: Path,
-    workflow_dir_remote: Path,
-    which_type: Literal["non_parallel", "parallel"],
-    config_path: Optional[Path] = None,
-) -> SlurmConfig:
-    """
-    Prepare a `SlurmConfig` configuration object
-
-    The argument `which_type` determines whether we use `wftask.meta_parallel`
-    or `wftask.meta_non_parallel`. In the following descritpion, let us assume
-    that `which_type="parallel"`.
-
-    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_parallel` (which typically include those in
-       `wftask.task.meta_parallel`). Note that `wftask.meta_parallel` may be
-       `None`.
-
-    Arguments:
-        wftask:
-            WorkflowTask for which the SLURM configuration is is to be
-            prepared.
-        workflow_dir_local:
-            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_remote:
-            User-side directory with the same scope as `workflow_dir_local`,
-            and where a user can write.
-        config_path:
-            Path of a Fractal SLURM configuration file; if `None`, use
-            `FRACTAL_SLURM_CONFIG_FILE` variable from settings.
-        which_type:
-            Determines whether to use `meta_parallel` or `meta_non_parallel`.
-
-    Returns:
-        slurm_config:
-            The SlurmConfig object
-    """
-
-    if which_type == "non_parallel":
-        wftask_meta = wftask.meta_non_parallel
-    elif which_type == "parallel":
-        wftask_meta = wftask.meta_parallel
-    else:
-        raise ValueError(
-            f"get_slurm_config received invalid argument {which_type=}."
-        )
-
-    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
-    if wftask.is_legacy_task:
-        job_name = wftask.task_legacy.name.replace(" ", "_")
-    else:
-        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