fractal-server 2.16.0__tar.gz → 2.16.2__tar.gz

{fractal_server-2.16.0 → fractal_server-2.16.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: fractal-server
-Version: 2.16.0
+Version: 2.16.2
 Summary: Backend component of the Fractal analytics platform
 License: BSD-3-Clause
 Author: Tommaso Comparin

fractal_server-2.16.2/fractal_server/__init__.py

@@ -0,0 +1 @@
+__VERSION__ = "2.16.2"

{fractal_server-2.16.0 → fractal_server-2.16.2}/fractal_server/app/runner/v2/runner.py

@@ -1,4 +1,3 @@
-import logging
 from collections.abc import Callable
 from copy import copy
 from copy import deepcopy
@@ -38,6 +37,7 @@ from fractal_server.images.status_tools import IMAGE_STATUS_KEY
 from fractal_server.images.tools import filter_image_list
 from fractal_server.images.tools import find_image_by_zarr_url
 from fractal_server.images.tools import merge_type_filters
+from fractal_server.logger import get_logger
 from fractal_server.types import AttributeFilters
 
 
@@ -103,7 +103,7 @@ def execute_tasks_v2(
     job_type_filters: dict[str, bool],
     job_attribute_filters: AttributeFilters,
 ) -> None:
-    logger = logging.getLogger(logger_name)
+    logger = get_logger(logger_name=logger_name)
 
     if not workflow_dir_local.exists():
         logger.warning(
@@ -198,9 +198,6 @@ def execute_tasks_v2(
             # over from a previous task
             job_db = db.get(JobV2, job_id)
             job_db.executor_error_log = None
-            logger.debug(
-                f"Resetting `JobV2[{job_id}].executor_error_log` to None."
-            )
             db.merge(job_db)
             db.commit()
 
@@ -449,10 +446,6 @@ def execute_tasks_v2(
             # Store the SLURM error in the job database
             job_db = db.get(JobV2, job_id)
             job_db.executor_error_log = runner.executor_error_log
-            logger.debug(
-                f"Setting JobV2[{job_id}].executor_error_log "
-                f"to '{job_db.executor_error_log}'."
-            )
             db.merge(job_db)
             db.commit()
 

{fractal_server-2.16.0 → fractal_server-2.16.2}/fractal_server/config.py

@@ -18,16 +18,19 @@ import sys
 from os import environ
 from os import getenv
 from pathlib import Path
+from typing import Annotated
 from typing import Literal
 from typing import TypeVar
 
 from cryptography.fernet import Fernet
 from dotenv import load_dotenv
+from pydantic import AfterValidator
 from pydantic import BaseModel
 from pydantic import EmailStr
 from pydantic import Field
 from pydantic import field_validator
 from pydantic import model_validator
+from pydantic import PositiveInt
 from pydantic import SecretStr
 from pydantic_settings import BaseSettings
 from pydantic_settings import SettingsConfigDict
@@ -36,6 +39,7 @@ from sqlalchemy.engine import URL
 import fractal_server
 from fractal_server.types import AbsolutePathStr
 from fractal_server.types import DictStrStr
+from fractal_server.types import NonEmptyStr
 
 
 class MailSettings(BaseModel):
@@ -64,6 +68,47 @@ class MailSettings(BaseModel):
     use_login: bool
 
 
+def _check_pixi_slurm_memory(mem: str) -> str:
+    if mem[-1] not in ["K", "M", "G", "T"]:
+        raise ValueError(
+            f"Invalid memory requirement {mem=} for `pixi`, "
+            "please set a K/M/G/T units suffix."
+        )
+    return mem
+
+
+class PixiSLURMConfig(BaseModel):
+    """
+    Parameters that are passed directly to a `sbatch` command.
+
+    See https://slurm.schedmd.com/sbatch.html.
+    """
+
+    partition: NonEmptyStr
+    """
+    `-p, --partition=<partition_names>`
+    """
+    cpus: PositiveInt
+    """
+    `-c, --cpus-per-task=<ncpus>
+    """
+    mem: Annotated[NonEmptyStr, AfterValidator(_check_pixi_slurm_memory)]
+    """
+    `--mem=<size>[units]` (examples: `"10M"`, `"10G"`).
+    From `sbatch` docs: Specify the real memory required per node. Default
+    units are megabytes. Different units can be specified using the suffix
+    [K|M|G|T].
+    """
+    time: NonEmptyStr
+    """
+    `-t, --time=<time>`.
+    From `sbatch` docs: "A time limit of zero requests that no time limit be
+    imposed. Acceptable time formats include "minutes", "minutes:seconds",
+    "hours:minutes:seconds", "days-hours", "days-hours:minutes" and
+    "days-hours:minutes:seconds".
+    """
+
+
 class PixiSettings(BaseModel):
     """
     Configuration for Pixi Task collection.
@@ -130,9 +175,15 @@ class PixiSettings(BaseModel):
     """
     DEFAULT_ENVIRONMENT: str = "default"
     """
+    Default pixi environment name.
     """
     DEFAULT_PLATFORM: str = "linux-64"
     """
+    Default platform for pixi.
+    """
+    SLURM_CONFIG: PixiSLURMConfig | None = None
+    """
+    Required when using pixi in a SSH/SLURM deployment.
     """
 
     @model_validator(mode="after")
@@ -554,12 +605,6 @@ class Settings(BaseSettings):
     still running on SLURM.
     """
 
-    FRACTAL_SLURM_SBATCH_SLEEP: float = 0
-    """
-    Interval to wait (in seconds) between two subsequent `sbatch` calls, when
-    running a task that produces multiple SLURM jobs.
-    """
-
     FRACTAL_PIP_CACHE_DIR: AbsolutePathStr | None = None
     """
     Absolute path to the cache directory for `pip`; if unset,
@@ -790,6 +835,10 @@ class Settings(BaseSettings):
                 raise FractalConfigurationError(
                     f"Must set FRACTAL_SLURM_WORKER_PYTHON when {info}"
                 )
+            if self.pixi and self.pixi.SLURM_CONFIG is None:
+                raise FractalConfigurationError(
+                    "Pixi config must include SLURM_CONFIG."
+                )
 
             from fractal_server.app.runner.executors.slurm_common._slurm_config import (  # noqa: E501
                 load_slurm_config_file,

fractal_server-2.16.2/fractal_server/tasks/v2/ssh/_pixi_slurm_ssh.py

@@ -0,0 +1,214 @@
+import os
+import time
+from pathlib import Path
+
+from sqlalchemy.orm import Session
+
+from fractal_server.app.models.v2 import TaskGroupActivityV2
+from fractal_server.config import get_settings
+from fractal_server.config import PixiSLURMConfig
+from fractal_server.logger import get_logger
+from fractal_server.ssh._fabric import FractalSSH
+from fractal_server.syringe import Inject
+from fractal_server.tasks.v2.utils_background import add_commit_refresh
+from fractal_server.tasks.v2.utils_background import get_current_log
+
+FRACTAL_SQUEUE_ERROR_STATE = "__FRACTAL_SQUEUE_ERROR__"
+
+# https://slurm.schedmd.com/squeue.html#lbAG
+STATES_FINISHED = {
+    "BOOT_FAIL",
+    "CANCELLED",
+    "COMPLETED",
+    "DEADLINE",
+    "FAILED",
+    "NODE_FAIL",
+    "OUT_OF_MEMORY",
+    "PREEMPTED",
+    "SPECIAL_EXIT",
+    "TIMEOUT",
+    FRACTAL_SQUEUE_ERROR_STATE,
+}
+
+
+def _log_change_of_job_state(
+    *,
+    old_state: str | None,
+    new_state: str,
+    logger_name: str,
+) -> None:
+    """
+    Emit a log for state changes.
+
+    Args:
+        old_state:
+        new_state:
+        logger_name:
+    """
+    if new_state != old_state:
+        logger = get_logger(logger_name=logger_name)
+        logger.debug(
+            f"SLURM-job state changed from {old_state=} to {new_state=}."
+        )
+
+
+def _run_squeue(
+    *,
+    fractal_ssh: FractalSSH,
+    squeue_cmd: str,
+    logger_name: str,
+) -> str:
+    """
+    Run a `squeue` command and handle exceptions.
+
+    Args:
+        fractal_ssh:
+        logger_name:
+        squeue_cmd:
+
+    Return:
+        state: The SLURM-job state.
+    """
+    try:
+        cmd_stdout = fractal_ssh.run_command(cmd=squeue_cmd)
+        state = cmd_stdout.strip().split()[1]
+        return state
+    except Exception as e:
+        logger = get_logger(logger_name=logger_name)
+        logger.info(f"`squeue` command failed (original error: {e})")
+        return FRACTAL_SQUEUE_ERROR_STATE
+
+
+def _verify_success_file_exists(
+    *,
+    fractal_ssh: FractalSSH,
+    success_file_remote: str,
+    logger_name: str,
+    stderr_remote: str,
+) -> None:
+    """
+    Fail if the success sentinel file does not exist remotely.
+
+    Note: the `FractalSSH` methods in this function may fail, and such failures
+    are not handled in this function. Any such failure, however, will lead to
+    a "failed" task-group lifecycle activity (because it will raise an
+    exception from within `run_script_on_remote_slurm`, which will then be
+    handled at the calling-function level.
+    """
+    if not fractal_ssh.remote_exists(path=success_file_remote):
+        logger = get_logger(logger_name=logger_name)
+        error_msg = f"{success_file_remote=} missing."
+        logger.info(error_msg)
+        if fractal_ssh.remote_exists(stderr_remote):
+            stderr = fractal_ssh.read_remote_text_file(stderr_remote)
+            logger.info(f"SLURM-job stderr:\n{stderr}")
+        raise RuntimeError(error_msg)
+
+
+def run_script_on_remote_slurm(
+    *,
+    script_path: str,
+    slurm_config: PixiSLURMConfig,
+    fractal_ssh: FractalSSH,
+    logger_name: str,
+    log_file_path: Path,
+    prefix: str,
+    db: Session,
+    activity: TaskGroupActivityV2,
+):
+    """
+    Run a `pixi install` script as a SLURM job.
+
+    NOTE: This is called from within a try/except, thus we can use exceptions
+    as a mechanism to propagate failure/errors.
+    """
+
+    logger = get_logger(logger_name=logger_name)
+    settings = Inject(get_settings)
+
+    # (1) Prepare remote submission script
+    workdir_remote = Path(script_path).parent.as_posix()
+    submission_script_remote = os.path.join(
+        workdir_remote, f"{prefix}-submit.sh"
+    )
+    stderr_remote = os.path.join(workdir_remote, f"{prefix}-err.txt")
+    stdout_remote = os.path.join(workdir_remote, f"{prefix}-out.txt")
+    success_file_remote = os.path.join(workdir_remote, f"{prefix}-success.txt")
+    script_lines = [
+        "#!/bin/bash",
+        f"#SBATCH --partition={slurm_config.partition}",
+        f"#SBATCH --cpus-per-task={slurm_config.cpus}",
+        f"#SBATCH --mem={slurm_config.mem}",
+        f"#SBATCH --time={slurm_config.time}",
+        f"#SBATCH --err={stderr_remote}",
+        f"#SBATCH --out={stdout_remote}",
+        f"#SBATCH -D {workdir_remote}",
+        "",
+        f"bash {script_path}",
+        f"touch {success_file_remote}",
+        "",
+    ]
+    script_contents = "\n".join(script_lines)
+    fractal_ssh.write_remote_file(
+        path=submission_script_remote,
+        content=script_contents,
+    )
+    logger.debug(f"Written {submission_script_remote=}.")
+
+    activity.log = get_current_log(log_file_path)
+    activity = add_commit_refresh(obj=activity, db=db)
+
+    # (2) Submit SLURM job
+    logger.debug("Now submit SLURM job.")
+    sbatch_cmd = f"sbatch --parsable {submission_script_remote}"
+    try:
+        stdout = fractal_ssh.run_command(cmd=sbatch_cmd)
+        job_id = int(stdout)
+        logger.debug(f"SLURM-job submission successful ({job_id=}).")
+    except Exception as e:
+        logger.error(
+            (
+                f"Submission of {submission_script_remote} failed. "
+                f"Original error: {str(e)}"
+            )
+        )
+        raise e
+    finally:
+        activity.log = get_current_log(log_file_path)
+        activity = add_commit_refresh(obj=activity, db=db)
+
+    # (3) Monitor job
+    squeue_cmd = (
+        f"squeue --noheader --format='%i %T' --states=all --jobs={job_id}"
+    )
+    logger.debug(f"Start monitoring job with {squeue_cmd=}.")
+    old_state = None
+    while True:
+        new_state = _run_squeue(
+            fractal_ssh=fractal_ssh,
+            squeue_cmd=squeue_cmd,
+            logger_name=logger_name,
+        )
+        _log_change_of_job_state(
+            old_state=old_state,
+            new_state=new_state,
+            logger_name=logger_name,
+        )
+        activity.log = get_current_log(log_file_path)
+        activity = add_commit_refresh(obj=activity, db=db)
+        if new_state in STATES_FINISHED:
+            logger.debug(f"Exit retrieval loop (state={new_state}).")
+            break
+        old_state = new_state
+        time.sleep(settings.FRACTAL_SLURM_POLL_INTERVAL)
+
+    _verify_success_file_exists(
+        fractal_ssh=fractal_ssh,
+        logger_name=logger_name,
+        success_file_remote=success_file_remote,
+        stderr_remote=stderr_remote,
+    )
+
+    logger.info("SLURM-job execution completed successfully, continue.")
+    activity.log = get_current_log(log_file_path)
+    activity = add_commit_refresh(obj=activity, db=db)

{fractal_server-2.16.0 → fractal_server-2.16.2}/fractal_server/tasks/v2/ssh/_utils.py

@@ -17,29 +17,26 @@ from fractal_server.tasks.v2.utils_templates import customize_template
 logger = set_logger(__name__)
 
 
-def _customize_and_run_template(
+def _customize_and_send_template(
     *,
     template_filename: str,
     replacements: set[tuple[str, str]],
     script_dir_local: str,
+    script_dir_remote: str,
     prefix: str,
     fractal_ssh: FractalSSH,
-    script_dir_remote: str,
     logger_name: str,
-    login_shell: bool = False,
 ) -> str:
     """
-    Customize one of the template bash scripts, transfer it to the remote host
-    via SFTP and then run it via SSH.
+    Customize a template bash scripts and transfer it to the remote host.
 
     Args:
         template_filename: Filename of the template file (ends with ".sh").
         replacements: Dictionary of replacements.
         script_dir_local: Local folder where the script will be placed.
+        script_dir_remote: Remote scripts directory
         prefix: Prefix for the script filename.
         fractal_ssh: FractalSSH object
-        script_dir_remote: Remote scripts directory
-        login_shell: If `True`, use `bash --login` for remote script execution.
     """
     logger = get_logger(logger_name=logger_name)
     logger.debug(f"_customize_and_run_template {template_filename} - START")
@@ -67,10 +64,46 @@ def _customize_and_run_template(
         local=script_path_local,
         remote=script_path_remote,
     )
+    return script_path_remote
+
+
+def _customize_and_run_template(
+    *,
+    template_filename: str,
+    replacements: set[tuple[str, str]],
+    script_dir_local: str,
+    script_dir_remote: str,
+    prefix: str,
+    fractal_ssh: FractalSSH,
+    logger_name: str,
+) -> str:
+    """
+    Customize one of the template bash scripts, transfer it to the remote host
+    via SFTP and then run it via SSH.
+
+    Args:
+        template_filename: Filename of the template file (ends with ".sh").
+        replacements: Dictionary of replacements.
+        script_dir_remote: Remote scripts directory
+        script_dir_local: Local folder where the script will be placed.
+        prefix: Prefix for the script filename.
+        fractal_ssh: FractalSSH object
+    """
+    logger = get_logger(logger_name=logger_name)
+    logger.debug(f"_customize_and_run_template {template_filename} - START")
+
+    script_path_remote = _customize_and_send_template(
+        template_filename=template_filename,
+        replacements=replacements,
+        script_dir_local=script_dir_local,
+        script_dir_remote=script_dir_remote,
+        prefix=prefix,
+        fractal_ssh=fractal_ssh,
+        logger_name=logger_name,
+    )
 
     # Execute script remotely
-    bash = "bash --login" if login_shell else "bash"
-    cmd = f"{bash} {script_path_remote}"
+    cmd = f"bash {script_path_remote}"
     logger.debug(f"Now run '{cmd}' over SSH.")
     stdout = fractal_ssh.run_command(cmd=cmd)
 

{fractal_server-2.16.0 → fractal_server-2.16.2}/fractal_server/tasks/v2/ssh/collect_pixi.py

@@ -8,6 +8,7 @@ from ..utils_background import prepare_tasks_metadata
 from ..utils_database import create_db_tasks_and_update_task_group_sync
 from ..utils_pixi import parse_collect_stdout
 from ..utils_pixi import SOURCE_DIR_NAME
+from ._pixi_slurm_ssh import run_script_on_remote_slurm
 from ._utils import check_ssh_or_fail_and_cleanup
 from ._utils import edit_pyproject_toml_in_place_ssh
 from fractal_server.app.db import get_sync_db
@@ -22,6 +23,7 @@ from fractal_server.ssh._fabric import SingleUseFractalSSH
 from fractal_server.ssh._fabric import SSHConfig
 from fractal_server.syringe import Inject
 from fractal_server.tasks.v2.ssh._utils import _customize_and_run_template
+from fractal_server.tasks.v2.ssh._utils import _customize_and_send_template
 from fractal_server.tasks.v2.utils_background import add_commit_refresh
 from fractal_server.tasks.v2.utils_background import get_current_log
 from fractal_server.tasks.v2.utils_templates import SCRIPTS_SUBFOLDER
@@ -201,16 +203,33 @@ def collect_ssh_pixi(
                         pyproject_toml_path=pyproject_toml_path,
                     )
 
-                    stdout = _customize_and_run_template(
+                    # Run script 2 - run pixi-install command
+                    remote_script2_path = _customize_and_send_template(
                         template_filename="pixi_2_install.sh",
                         replacements=replacements,
-                        login_shell=True,
                         **common_args,
                     )
-                    logger.debug(f"STDOUT: {stdout}")
+                    logger.debug(
+                        "Installation script written to "
+                        f"{remote_script2_path=}."
+                    )
+                    activity.log = get_current_log(log_file_path)
+                    activity = add_commit_refresh(obj=activity, db=db)
+
+                    run_script_on_remote_slurm(
+                        script_path=remote_script2_path,
+                        slurm_config=settings.pixi.SLURM_CONFIG,
+                        fractal_ssh=fractal_ssh,
+                        logger_name=LOGGER_NAME,
+                        prefix=common_args["prefix"],
+                        db=db,
+                        activity=activity,
+                        log_file_path=log_file_path,
+                    )
                     activity.log = get_current_log(log_file_path)
                     activity = add_commit_refresh(obj=activity, db=db)
 
+                    # Run script 3 - post-install
                     stdout = _customize_and_run_template(
                         template_filename="pixi_3_post_install.sh",
                         replacements=replacements,

{fractal_server-2.16.0 → fractal_server-2.16.2}/fractal_server/tasks/v2/ssh/reactivate_pixi.py

@@ -5,6 +5,7 @@ from tempfile import TemporaryDirectory
 from ..utils_background import fail_and_cleanup
 from ..utils_background import get_activity_and_task_group
 from ..utils_pixi import SOURCE_DIR_NAME
+from ._pixi_slurm_ssh import run_script_on_remote_slurm
 from ._utils import check_ssh_or_fail_and_cleanup
 from ._utils import edit_pyproject_toml_in_place_ssh
 from fractal_server.app.db import get_sync_db
@@ -18,6 +19,7 @@ from fractal_server.ssh._fabric import SSHConfig
 from fractal_server.syringe import Inject
 from fractal_server.tasks.utils import get_log_path
 from fractal_server.tasks.v2.ssh._utils import _customize_and_run_template
+from fractal_server.tasks.v2.ssh._utils import _customize_and_send_template
 from fractal_server.tasks.v2.utils_background import add_commit_refresh
 from fractal_server.tasks.v2.utils_background import get_current_log
 from fractal_server.tasks.v2.utils_templates import SCRIPTS_SUBFOLDER
@@ -187,13 +189,28 @@ def reactivate_ssh_pixi(
                     )
 
                     # Run script 2 - run pixi-install command
-                    stdout = _customize_and_run_template(
+                    remote_script2_path = _customize_and_send_template(
                         template_filename="pixi_2_install.sh",
                         replacements=replacements,
-                        login_shell=True,
                         **common_args,
                     )
-                    logger.debug(f"STDOUT: {stdout}")
+                    logger.debug(
+                        "Installation script written to "
+                        f"{remote_script2_path=}."
+                    )
+                    activity.log = get_current_log(log_file_path)
+                    activity = add_commit_refresh(obj=activity, db=db)
+
+                    run_script_on_remote_slurm(
+                        script_path=remote_script2_path,
+                        slurm_config=settings.pixi.SLURM_CONFIG,
+                        fractal_ssh=fractal_ssh,
+                        logger_name=LOGGER_NAME,
+                        prefix=common_args["prefix"],
+                        db=db,
+                        activity=activity,
+                        log_file_path=log_file_path,
+                    )
                     activity.log = get_current_log(log_file_path)
                     activity = add_commit_refresh(obj=activity, db=db)
 

{fractal_server-2.16.0 → fractal_server-2.16.2}/fractal_server/tasks/v2/utils_background.py

@@ -143,6 +143,6 @@ def prepare_tasks_metadata(
     return task_list
 
 
-def get_current_log(logger_file_path: str) -> str:
+def get_current_log(logger_file_path: str | Path) -> str:
     with open(logger_file_path) as f:
         return f.read()

{fractal_server-2.16.0 → fractal_server-2.16.2}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "fractal-server"
-version = "2.16.0"
+version = "2.16.2"
 description = "Backend component of the Fractal analytics platform"
 authors = [
     { name="Tommaso Comparin", email="tommaso.comparin@exact-lab.it" },
@@ -95,7 +95,7 @@ filterwarnings = [
 markers = ["container", "ssh", "fails_on_macos"]
 
 [tool.bumpver]
-current_version = "2.16.0"
+current_version = "2.16.2"
 version_pattern = "MAJOR.MINOR.PATCH[PYTAGNUM]"
 commit_message = "bump version {old_version} -> {new_version}"
 commit = true

fractal_server-2.16.0/fractal_server/__init__.py

@@ -1 +0,0 @@
-__VERSION__ = "2.16.0"