fractal-server 2.16.2a0__py3-none-any.whl → 2.16.3__py3-none-any.whl

fractal_server/__init__.py

@@ -1 +1 @@
-__VERSION__ = "2.16.2a0"
+__VERSION__ = "2.16.3"

fractal_server/tasks/v2/ssh/_pixi_slurm_ssh.py

@@ -2,12 +2,18 @@ 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 = {
@@ -21,20 +27,130 @@ STATES_FINISHED = {
     "PREEMPTED",
     "SPECIAL_EXIT",
     "TIMEOUT",
+    FRACTAL_SQUEUE_ERROR_STATE,
 }
 
 
+def _get_workdir_remote(script_paths: list[str]) -> str:
+    """
+    Check that there is one and only one `workdir`, and return it.
+
+    Note: The `is_absolute` check is to filter out a `chmod` command.
+    """
+    workdirs = [
+        Path(script_path).parent.as_posix()
+        for script_path in script_paths
+        if Path(script_path).is_absolute()
+    ]
+    if not len(set(workdirs)) == 1:
+        raise ValueError(f"Invalid {script_paths=}.")
+    return workdirs[0]
+
+
+def _read_file_if_exists(
+    *,
+    fractal_ssh: FractalSSH,
+    path: str,
+) -> str:
+    """
+    Read a remote file if it exists, or return an empty string.
+    """
+    if fractal_ssh.remote_exists(path=path):
+        return fractal_ssh.read_remote_text_file(path)
+    else:
+        return ""
+
+
+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)
+
+        stderr = _read_file_if_exists(
+            fractal_ssh=fractal_ssh, path=stderr_remote
+        )
+        if stderr:
+            logger.info(f"SLURM-job stderr:\n{stderr}")
+        raise RuntimeError(error_msg)
+
+
 def run_script_on_remote_slurm(
     *,
-    script_path: str,
+    script_paths: list[str],
     slurm_config: PixiSLURMConfig,
     fractal_ssh: FractalSSH,
     logger_name: str,
+    log_file_path: Path,
     prefix: str,
+    db: Session,
+    activity: TaskGroupActivityV2,
 ):
     """
-    FIXME
-
+    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.
@@ -44,7 +160,7 @@ def run_script_on_remote_slurm(
     settings = Inject(get_settings)
 
     # (1) Prepare remote submission script
-    workdir_remote = Path(script_path).parent.as_posix()
+    workdir_remote = _get_workdir_remote(script_paths)
     submission_script_remote = os.path.join(
         workdir_remote, f"{prefix}-submit.sh"
     )
@@ -61,55 +177,79 @@ def run_script_on_remote_slurm(
         f"#SBATCH --out={stdout_remote}",
         f"#SBATCH -D {workdir_remote}",
         "",
-        f"bash {script_path}",
-        f"touch {success_file_remote}",
-        "",
     ]
+    for script_path in script_paths:
+        script_lines.append(f"bash {script_path}")
+    script_lines.append(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
-    sbatch_cmd = f"sbatch --parsable {submission_script_remote} "
+    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)}"
+            (
+                f"Submission of {submission_script_remote} failed. "
+                f"Original error: {str(e)}"
+            )
         )
         raise e
-    logger.debug(f"Now submit job {submission_script_remote} to SLURM.")
-    job_id = int(stdout)
-    logger.debug(f"SLURM-job submission successful ({job_id=}).")
+    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:
-        try:
-            stdout = fractal_ssh.run_command(cmd=squeue_cmd)
-        except Exception as e:
-            # FIXME: review this logic
-            logger.info(
-                f"`squeue` command failed (original error: {e}), "
-                "consider the job as complete."
-            )
-            break
-        state = stdout.strip().split()[1]
-        logger.debug(f"Status of SLURM job {job_id}: {state}")
-        if state in STATES_FINISHED:
-            logger.debug(f"Exit retrieval loop ({state=}).")
+        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)
 
-    if fractal_ssh.remote_exists(path=success_file_remote):
-        logger.info(f"{success_file_remote=} exists.")
-    else:
-        raise RuntimeError(
-            "SLURM job did not complete correctly "
-            f"({success_file_remote=} missing)."
-        )
+    _verify_success_file_exists(
+        fractal_ssh=fractal_ssh,
+        logger_name=logger_name,
+        success_file_remote=success_file_remote,
+        stderr_remote=stderr_remote,
+    )
+
+    stdout = _read_file_if_exists(
+        fractal_ssh=fractal_ssh,
+        path=stdout_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)
+
+    return stdout

fractal_server/tasks/v2/ssh/collect_pixi.py

@@ -203,12 +203,21 @@ def collect_ssh_pixi(
                         pyproject_toml_path=pyproject_toml_path,
                     )
 
-                    # Run script 2 - run pixi-install command
+                    # Prepare scripts 2 and 3
                     remote_script2_path = _customize_and_send_template(
                         template_filename="pixi_2_install.sh",
                         replacements=replacements,
                         **common_args,
                     )
+                    remote_script3_path = _customize_and_send_template(
+                        template_filename="pixi_3_post_install.sh",
+                        replacements=replacements,
+                        **common_args,
+                    )
+                    logger.debug(
+                        "Post-installation script written to "
+                        f"{remote_script3_path=}."
+                    )
                     logger.debug(
                         "Installation script written to "
                         f"{remote_script2_path=}."
@@ -216,26 +225,24 @@ def collect_ssh_pixi(
                     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,
+                    # Run scripts 2 and 3
+                    stdout = run_script_on_remote_slurm(
+                        script_paths=[
+                            remote_script2_path,
+                            remote_script3_path,
+                            f"chmod -R 755 {source_dir}",
+                        ],
                         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,
-                        **common_args,
-                    )
-                    logger.debug(f"STDOUT: {stdout}")
-                    activity.log = get_current_log(log_file_path)
-                    activity = add_commit_refresh(obj=activity, db=db)
-
                     # Parse stdout
                     parsed_output = parse_collect_stdout(stdout)
                     package_root_remote = parsed_output["package_root"]
@@ -245,8 +252,6 @@ def collect_ssh_pixi(
                         "project_python_wrapper"
                     ]
 
-                    fractal_ssh.run_command(cmd=f"chmod -R 755 {source_dir}")
-
                     # Read and validate remote manifest file
                     manifest_path_remote = (
                         f"{package_root_remote}/__FRACTAL_MANIFEST__.json"

fractal_server/tasks/v2/ssh/reactivate_pixi.py

@@ -188,12 +188,21 @@ def reactivate_ssh_pixi(
                         remote=pixi_lock_remote,
                     )
 
-                    # Run script 2 - run pixi-install command
+                    # Prepare scripts 2 and 3
                     remote_script2_path = _customize_and_send_template(
                         template_filename="pixi_2_install.sh",
                         replacements=replacements,
                         **common_args,
                     )
+                    remote_script3_path = _customize_and_send_template(
+                        template_filename="pixi_3_post_install.sh",
+                        replacements=replacements,
+                        **common_args,
+                    )
+                    logger.debug(
+                        "Post-installation script written to "
+                        f"{remote_script3_path=}."
+                    )
                     logger.debug(
                         "Installation script written to "
                         f"{remote_script2_path=}."
@@ -201,28 +210,24 @@ def reactivate_ssh_pixi(
                     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,
+                    # Run scripts 2 and 3
+                    stdout = run_script_on_remote_slurm(
+                        script_paths=[
+                            remote_script2_path,
+                            remote_script3_path,
+                            f"chmod -R 755 {source_dir}",
+                        ],
                         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,
-                        **common_args,
-                    )
-                    logger.debug(f"STDOUT: {stdout}")
-                    activity.log = get_current_log(log_file_path)
-                    activity = add_commit_refresh(obj=activity, db=db)
-
-                    fractal_ssh.run_command(cmd=f"chmod -R 755 {source_dir}")
-
                     # Finalize (write metadata to DB)
                     activity.status = TaskGroupActivityStatusV2.OK
                     activity.timestamp_ended = get_timestamp()

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-2.16.2a0.dist-info → fractal_server-2.16.3.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: fractal-server
-Version: 2.16.2a0
+Version: 2.16.3
 Summary: Backend component of the Fractal analytics platform
 License: BSD-3-Clause
 Author: Tommaso Comparin
@@ -12,7 +12,7 @@ Classifier: Programming Language :: Python :: 3.11
 Classifier: Programming Language :: Python :: 3.12
 Classifier: Programming Language :: Python :: 3.13
 Requires-Dist: alembic (>=1.13.1,<2.0.0)
-Requires-Dist: cryptography (>=45.0.3,<45.1.0)
+Requires-Dist: cryptography (>=46.0.0,<47.0.0)
 Requires-Dist: fabric (>=3.2.2,<3.3.0)
 Requires-Dist: fastapi (>=0.116.0,<0.117.0)
 Requires-Dist: fastapi-users[oauth] (>=14,<15)

{fractal_server-2.16.2a0.dist-info → fractal_server-2.16.3.dist-info}/RECORD

@@ -1,4 +1,4 @@
-fractal_server/__init__.py,sha256=ywvZY_b51fGszgBtc9wZwOVbz6O5w5tzfaYERx4Ui_s,25
+fractal_server/__init__.py,sha256=XhckXI3NQgXzQFmbqjMIwxfxShIlVm8JhcoaOhWWvyY,23
 fractal_server/__main__.py,sha256=rkM8xjY1KeS3l63irB8yCrlVobR-73uDapC4wvrIlxI,6957
 fractal_server/alembic.ini,sha256=MWwi7GzjzawI9cCAK1LW7NxIBQDUqD12-ptJoq5JpP0,3153
 fractal_server/app/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
@@ -204,15 +204,15 @@ fractal_server/tasks/v2/local/delete.py,sha256=TbNcg0IhbJNXzXsO2Vgv6Dq4jWmFUVtSu
 fractal_server/tasks/v2/local/reactivate.py,sha256=Q43DOadNeFyyfgNP67lUqaXmZsS6onv67XwxH_-5ANA,5756
 fractal_server/tasks/v2/local/reactivate_pixi.py,sha256=IuxDRaj8i6Rc582TIbc9HVKQ9pOvR4IepyXSaJx2PfQ,7565
 fractal_server/tasks/v2/ssh/__init__.py,sha256=dPK6BtEZVh1GiFP05j1RKTEnZvjJez8o2KkMC2hWXaw,339
-fractal_server/tasks/v2/ssh/_pixi_slurm_ssh.py,sha256=P-QyoKlQ9rbq3DPuG0DjgnbJ9KOCBXqbFcJZ2qB4VzA,3484
+fractal_server/tasks/v2/ssh/_pixi_slurm_ssh.py,sha256=24BX64KbnBr4Wd3k9KmSNz4WHzvLf9AF7Hh1-FDlC9w,7472
 fractal_server/tasks/v2/ssh/_utils.py,sha256=AjbH4o6-ENe-ZiHvE_pGe_yw8e2lapGMYtLX7VuBkA8,6111
 fractal_server/tasks/v2/ssh/collect.py,sha256=WesUBtNaax9ST7CtgqAD2qd-ZII4t1JFPzRzltEOuM8,14333
-fractal_server/tasks/v2/ssh/collect_pixi.py,sha256=5CRr_XiuRaFjydZqxgfoMMj3p1TfT7aqNOxHcVxrOQI,15020
+fractal_server/tasks/v2/ssh/collect_pixi.py,sha256=i7_xqAy9kkF19AHHGSAmvs-QUEQIkPJvVxXUiau1WhU,15210
 fractal_server/tasks/v2/ssh/deactivate.py,sha256=7gJ2mOah0wKwUnK0S9QpaXg08_WE95P0rC-oExAGhLE,12438
 fractal_server/tasks/v2/ssh/deactivate_pixi.py,sha256=K0yK_NPUqhFMj6cp6G_0Kfn0Yo7oQux4kT5dFPulnos,4748
 fractal_server/tasks/v2/ssh/delete.py,sha256=9DCe5QXZ14pY8NRb_FCqpIr31r5Gboz6bnFrS8Oa044,4277
 fractal_server/tasks/v2/ssh/reactivate.py,sha256=NJIgMNFKaXMhbvK0iZOsMwMtsms6Boj9f8N4L01X9Bo,8271
-fractal_server/tasks/v2/ssh/reactivate_pixi.py,sha256=LZk7WaC2uCo3cHTwrpRcc4F_I7HUDA-8rE-RJ2agLZA,11020
+fractal_server/tasks/v2/ssh/reactivate_pixi.py,sha256=yRoq7xTCpKgx8Fxc5RTpEwQI-L0-9Tjq7vnX_v3v7JE,11210
 fractal_server/tasks/v2/templates/1_create_venv.sh,sha256=PK0jdHKtQpda1zULebBaVPORt4t6V17wa4N1ohcj5ac,548
 fractal_server/tasks/v2/templates/2_pip_install.sh,sha256=vN9DX-eucJjB-XCuQuZmkuATGBzL4FlDQJTQdVewJG8,2155
 fractal_server/tasks/v2/templates/3_pip_freeze.sh,sha256=JldREScEBI4cD_qjfX4UK7V4aI-FnX9ZvVNxgpSOBFc,168
@@ -222,7 +222,7 @@ fractal_server/tasks/v2/templates/6_pip_install_from_freeze.sh,sha256=jHoALPagSC
 fractal_server/tasks/v2/templates/pixi_1_extract.sh,sha256=Jdy5OyKo2jxe_qIDB9Zi4a0FL0cMBysxvBPHlUrARQM,1099
 fractal_server/tasks/v2/templates/pixi_2_install.sh,sha256=h6-M101Q1AdAfZNZyPfSUc8AlZ-uS84Hae4vJdDSglY,1601
 fractal_server/tasks/v2/templates/pixi_3_post_install.sh,sha256=99J8KXkNeQk9utuEtUxfAZS6VCThC32X7I7HAp2gdTU,2501
-fractal_server/tasks/v2/utils_background.py,sha256=LhiJtlWSsXo_FIednOclhWpBjmIYwzs83hecWAfoGNo,4833
+fractal_server/tasks/v2/utils_background.py,sha256=jjWxNbHPuvAkXNIQ9Bqs67X72xHSRqmY8_BCoj0HM3E,4840
 fractal_server/tasks/v2/utils_database.py,sha256=yi7793Uue32O59OBVUgomO42oUrVKdSKXoShBUNDdK0,1807
 fractal_server/tasks/v2/utils_package_names.py,sha256=RDg__xrvQs4ieeVzmVdMcEh95vGQYrv9Hfal-5EDBM8,2393
 fractal_server/tasks/v2/utils_pixi.py,sha256=tqCnxMdxs7KGWncWvk0alrdvDbX-w77P3fAot68Bqh4,3257
@@ -236,8 +236,8 @@ fractal_server/types/validators/_workflow_task_arguments_validators.py,sha256=HL
 fractal_server/urls.py,sha256=QjIKAC1a46bCdiPMu3AlpgFbcv6a4l3ABcd5xz190Og,471
 fractal_server/utils.py,sha256=Vn35lApt1T1J8nc09sAVqd10Cy0sa3dLipcljI-hkuk,2185
 fractal_server/zip_tools.py,sha256=H0w7wS5yE4ebj7hw1_77YQ959dl2c-L0WX6J_ro1TY4,4884
-fractal_server-2.16.2a0.dist-info/LICENSE,sha256=QKAharUuhxL58kSoLizKJeZE3mTCBnX6ucmz8W0lxlk,1576
-fractal_server-2.16.2a0.dist-info/METADATA,sha256=6x9kEFUU9jQkukimP25U3iu8UKP1JY2bCA67qTSxx2I,4336
-fractal_server-2.16.2a0.dist-info/WHEEL,sha256=b4K_helf-jlQoXBBETfwnf4B04YC67LOev0jo4fX5m8,88
-fractal_server-2.16.2a0.dist-info/entry_points.txt,sha256=8tV2kynvFkjnhbtDnxAqImL6HMVKsopgGfew0DOp5UY,58
-fractal_server-2.16.2a0.dist-info/RECORD,,
+fractal_server-2.16.3.dist-info/LICENSE,sha256=QKAharUuhxL58kSoLizKJeZE3mTCBnX6ucmz8W0lxlk,1576
+fractal_server-2.16.3.dist-info/METADATA,sha256=pw_mNLXMPEkIRXLZGpIG5a4T_S4ehtX1HAuqTkKZrXQ,4334
+fractal_server-2.16.3.dist-info/WHEEL,sha256=b4K_helf-jlQoXBBETfwnf4B04YC67LOev0jo4fX5m8,88
+fractal_server-2.16.3.dist-info/entry_points.txt,sha256=8tV2kynvFkjnhbtDnxAqImL6HMVKsopgGfew0DOp5UY,58
+fractal_server-2.16.3.dist-info/RECORD,,