experimaestro 2.0.0a1__tar.gz → 2.0.0a4__tar.gz

{experimaestro-2.0.0a1 → experimaestro-2.0.0a4}/PKG-INFO

@@ -1,8 +1,9 @@
-Metadata-Version: 2.3
+Metadata-Version: 2.4
 Name: experimaestro
-Version: 2.0.0a1
+Version: 2.0.0a4
 Summary: "Experimaestro is a computer science experiment manager"
 License: GPL-3
+License-File: LICENSE
 Keywords: experiment manager
 Author: Benjamin Piwowarski
 Author-email: benjamin@piwowarski.fr

{experimaestro-2.0.0a1 → experimaestro-2.0.0a4}/pyproject.toml

@@ -48,7 +48,7 @@ dependencies = [
     "typing-extensions >=4.2; python_version < \"3.12\"",
     "watchdog >=2"
 ]
-version = "2.0.0-a1"
+version = "2.0.0-a4"
 
 [tool.poetry-dynamic-versioning]
 enable = false
@@ -85,6 +85,9 @@ build-backend = "poetry_dynamic_versioning.backend"
 
 [dependency-groups]
 dev = [
+    "markdown-include>=0.8.1",
+    "mkdocstrings[python]>=0.30.1",
+    "pymdown-extensions>=10.16.1",
     "pytest>=8.4.1",
     "pytest-timeout>=2.4.0",
 ]
@@ -146,7 +149,7 @@ warn_unused_ignores = true
 
 [tool.commitizen]
 name = "cz_conventional_commits"
-version = "2.0.0a1"
+version = "2.0.0a4"
 changelog_start_rev = "v1.0.0"
 tag_format = "v$major.$minor.$patch$prerelease"
 # update_changelog_on_bump = true

{experimaestro-2.0.0a1 → experimaestro-2.0.0a4}/src/experimaestro/connectors/__init__.py

@@ -16,7 +16,7 @@ from experimaestro.utils import logger
 from experimaestro.locking import Lock
 from experimaestro.tokens import Token
 from experimaestro.utils.asyncio import asyncThreadcheck
-import pkg_resources
+from importlib.metadata import entry_points
 
 
 class RedirectType(enum.Enum):
@@ -101,7 +101,7 @@ class Process:
         """Get a handler"""
         if Process.HANDLERS is None:
             Process.HANDLERS = {}
-            for ep in pkg_resources.iter_entry_points(group="experimaestro.process"):
+            for ep in entry_points(group="experimaestro.process"):
                 logging.debug("Adding process handler for type %s", ep.name)
                 handler = ep.load()
                 Process.HANDLERS[ep.name] = handler

{experimaestro-2.0.0a1 → experimaestro-2.0.0a4}/src/experimaestro/core/arguments.py

@@ -80,10 +80,13 @@ class Argument:
 
         self.generator = generator
         self.default = None
+        self.ignore_generated = False
 
         if default is not None:
             assert self.generator is None, "generator and default are exclusive options"
             if isinstance(default, field):
+                self.ignore_generated = default.ignore_generated
+
                 if default.default is not None:
                     self.default = default.default
                 elif default.default_factory is not None:
@@ -184,13 +187,29 @@ DataPath = Annotated[Path, dataHint]
 class field:
     """Extra information for a given experimaestro field (param or meta)"""
 
-    def __init__(self, *, default: Any = None, default_factory: Callable = None):
+    def __init__(
+        self,
+        *,
+        default: Any = None,
+        default_factory: Callable = None,
+        ignore_generated=False,
+    ):
+        """Gives some extra per-field information
+
+        :param default: a default value, defaults to None
+        :param default_factory: a default factory for values, defaults to None
+        :param ignore_generated: True if the value is hidden – it won't be accessible in
+            tasks, defaults to False. The interest of hidden is to add a
+            configuration field that changes the identifier, but will not be
+            used.
+        """
         assert not (
             (default is not None) and (default_factory is not None)
         ), "default and default_factory are mutually exclusive options"
 
         self.default_factory = default_factory
         self.default = default
+        self.ignore_generated = ignore_generated
 
 
 class help(TypeAnnotation):

{experimaestro-2.0.0a1 → experimaestro-2.0.0a4}/src/experimaestro/core/objects/config.py

@@ -169,10 +169,33 @@ class ConfigInformation:
         self._sealed = False
         self._meta = None
 
-        #: This flags is True when a value in this configuration,
-        #: or any sub-configuration, is generated. This prevents problem
-        #: when a configuration with generated values is re-used
-        self._has_generated_value = False
+        # This contains the list of generated values (using context) in this
+        # configuration or any sub-configuration, is generated. This prevents
+        # problem when a configuration with generated values is re-used.
+        self._generated_values = []
+
+    def get_generated_paths(
+        self, path: list[str] = None, paths: list[str] = None
+    ) -> list[str]:
+        """Get the list of generated paths, useful to track down those
+
+        :param path: The current path
+        :param paths: The list of generated paths so far, defaults to None
+        :return: The full list of generated paths
+        """
+        paths = [] if paths is None else paths
+        path = [] if path is None else path
+
+        for key in self._generated_values:
+            value = self.values[key]
+            if isinstance(value, ConfigMixin) and value.__xpm__._generated_values:
+                path.append(key)
+                value.__xpm__.get_generated_paths(path, paths)
+                path.pop()
+            else:
+                paths.append(".".join(path + [key]))
+
+        return paths
 
     def set_meta(self, value: Optional[bool]):
         """Sets the meta flag"""
@@ -191,6 +214,31 @@ class ConfigInformation:
         # Not an argument, bypass
         return object.__getattribute__(self.pyobject, name)
 
+    @staticmethod
+    def is_generated_value(argument, value):
+        if argument.ignore_generated:
+            return False
+
+        if value is None:
+            return False
+
+        if isinstance(value, (int, str, float, bool, Path)):
+            return False
+
+        if isinstance(value, ConfigMixin):
+            return value.__xpm__._generated_values and value.__xpm__.task is None
+
+        if isinstance(value, list):
+            return any(ConfigInformation.is_generated_value(argument, x) for x in value)
+
+        if isinstance(value, dict):
+            return any(
+                ConfigInformation.is_generated_value(argument, x)
+                for x in value.values()
+            )
+
+        assert False, f"Cannot handle values of type {type(value)}"
+
     def set(self, k, v, bypass=False):
         from experimaestro.generators import Generator
 
@@ -207,18 +255,16 @@ class ConfigInformation:
                 "Configuration (and not objects) should be used. Consider using .C(...)"
             )
 
-        if (
-            isinstance(v, ConfigMixin)
-            and v.__xpm__._has_generated_value
-            and v.__xpm__.task is None
-        ):
-            raise AttributeError(
-                f"Cannot set {k} to a configuration with generated values"
-            )
-
         try:
             argument = self.xpmtype.arguments.get(k, None)
             if argument:
+                if ConfigInformation.is_generated_value(argument, v):
+                    raise AttributeError(
+                        f"Cannot set {k} to a configuration with generated values. "
+                        "Here is the list of paths to help you: "
+                        f"""{', '.join(v.__xpm__.get_generated_paths([k]))}"""
+                    )
+
                 if not bypass and (
                     (isinstance(argument.generator, Generator)) or argument.constant
                 ):
@@ -344,14 +390,15 @@ class ConfigInformation:
         Arguments:
             - context: the generation context
         """
-        subconfigs = [
-            v.__xpm__
-            for v in self.values.values()
-            if isinstance(v, Config) and v.__xpm__.task is None
-        ]
-
-        if any(v._has_generated_value for v in subconfigs):
-            raise AttributeError("Cannot seal a configuration with generated values")
+        if generated_keys := [
+            k
+            for k, v in self.values.items()
+            if ConfigInformation.is_generated_value(self.xpmtype.arguments[k], v)
+        ]:
+            raise AttributeError(
+                "Cannot seal a configuration with generated values:"
+                f"""{",".join(generated_keys)} in {context.currentpath}"""
+            )
 
         class Sealer(ConfigWalk):
             def preprocess(self, config: ConfigMixin):
@@ -375,13 +422,18 @@ class ConfigInformation:
                                 if len(sig.parameters) == 0:
                                     value = argument.generator()
                                 elif len(sig.parameters) == 2:
+                                    # Only in that case do we need to flag this configuration
+                                    # as containing generated values
+                                    if not argument.ignore_generated:
+                                        config.__xpm__._generated_values.append(k)
+                                    else:
+                                        logging.warning("Ignoring %s", k)
                                     value = argument.generator(self.context, config)
                                 else:
                                     assert (
                                         False
                                     ), "generator has either two parameters (context and config), or none"
                             config.__xpm__.set(k, value, bypass=True)
-                            config.__xpm__._has_generated_value = True
                         else:
                             value = config.__xpm__.values.get(k)
                     except Exception:
@@ -392,11 +444,14 @@ class ConfigInformation:
 
                     # Propagate the generated value flag
                     if (
-                        (value is not None)
+                        value is not None
                         and isinstance(value, ConfigMixin)
-                        and value.__xpm__._has_generated_value
+                        and value.__xpm__._generated_values
                     ):
-                        self._has_generated_value = True
+                        if not argument.ignore_generated:
+                            config.__xpm__._generated_values.append(k)
+                        else:
+                            logging.warning("Ignoring %s", k)
 
                 config.__xpm__._sealed = True
 
@@ -889,6 +944,7 @@ class ConfigInformation:
                 "workspace": str(context.workspace.path.absolute()),
                 "tags": {key: value for key, value in self.tags().items()},
                 "version": 2,
+                "experimaestro": experimaestro.__version__,
                 "objects": self.__get_objects__([], context),
             },
             out,

{experimaestro-2.0.0a1 → experimaestro-2.0.0a4}/src/experimaestro/core/objects/config_walk.py

@@ -71,6 +71,7 @@ class ConfigWalk:
         return self.context.push(str(i))
 
     def map(self, k: str):
+        """Provides a path context when processing a tree"""
         return self.context.push(k)
 
     def stub(self, config):
@@ -123,7 +124,8 @@ class ConfigWalk:
                 and self.recurse_task
                 and x.__xpm__.task is not x
             ):
-                self(x.__xpm__.task)
+                with self.map("__task__"):
+                    self(x.__xpm__.task)
 
             processed = self.postprocess(stub, x, result)
             self.visited[xid] = processed

experimaestro-2.0.0a4/src/experimaestro/scheduler/__init__.py

@@ -0,0 +1,18 @@
+from .base import Scheduler, Listener
+from .workspace import Workspace, RunMode
+from .experiment import experiment, FailedExperiment
+from .jobs import Job, JobState, JobFailureStatus, JobDependency, JobContext
+
+__all__ = [
+    "Scheduler",
+    "Listener",
+    "Workspace",
+    "RunMode",
+    "experiment",
+    "FailedExperiment",
+    "Job",
+    "JobState",
+    "JobFailureStatus",
+    "JobDependency",
+    "JobContext",
+]

experimaestro-2.0.0a4/src/experimaestro/scheduler/base.py

@@ -0,0 +1,310 @@
+import logging
+import threading
+import time
+from typing import (
+    Optional,
+    Set,
+)
+import asyncio
+from typing import Dict
+
+from experimaestro.scheduler import experiment
+from experimaestro.scheduler.jobs import Job, JobState
+from experimaestro.scheduler.services import Service
+
+
+from experimaestro.utils import logger
+from experimaestro.utils.asyncio import asyncThreadcheck
+import concurrent.futures
+
+
+class Listener:
+    def job_submitted(self, job):
+        pass
+
+    def job_state(self, job):
+        pass
+
+    def service_add(self, service: Service):
+        """Notify when a new service is added"""
+        pass
+
+
+class Scheduler(threading.Thread):
+    """A job scheduler
+
+    The scheduler is based on asyncio for easy concurrency handling
+    """
+
+    def __init__(self, xp: "experiment", name: str):
+        super().__init__(name=f"Scheduler ({name})", daemon=True)
+        self._ready = threading.Event()
+
+        # Name of the experiment
+        self.name = name
+        self.xp = xp
+
+        # Exit mode activated
+        self.exitmode = False
+
+        # List of all jobs
+        self.jobs: Dict[str, "Job"] = {}
+
+        # List of jobs
+        self.waitingjobs: Set[Job] = set()
+
+        # Listeners
+        self.listeners: Set[Listener] = set()
+
+    @staticmethod
+    def create(xp: "experiment", name: str):
+        instance = Scheduler(xp, name)
+        instance.start()
+        instance._ready.wait()
+        return instance
+
+    def run(self):
+        """Run the event loop forever"""
+        logger.debug("Starting event loop thread")
+        # Ported from SchedulerCentral
+        self.loop = asyncio.new_event_loop()
+        asyncio.set_event_loop(self.loop)
+        # Set loop-dependent variables
+        self.exitCondition = asyncio.Condition()
+        self.dependencyLock = asyncio.Lock()
+        self._ready.set()
+        self.loop.run_forever()
+
+    def start_scheduler(self):
+        """Start the scheduler event loop in a thread"""
+        if not self.is_alive():
+            self.start()
+            self._ready.wait()
+        else:
+            logger.warning("Scheduler already started")
+
+    def addlistener(self, listener: Listener):
+        self.listeners.add(listener)
+
+    def removelistener(self, listener: Listener):
+        self.listeners.remove(listener)
+
+    def getJobState(self, job: Job) -> "concurrent.futures.Future[JobState]":
+        # Check if the job belongs to this scheduler
+        if job.identifier not in self.jobs:
+            # If job is not in this scheduler, return its current state directly
+            future = concurrent.futures.Future()
+            future.set_result(job.state)
+            return future
+
+        return asyncio.run_coroutine_threadsafe(self.aio_getjobstate(job), self.loop)
+
+    async def aio_getjobstate(self, job: Job):
+        return job.state
+
+    def submit(self, job: Job) -> Optional[Job]:
+        # Wait for the future containing the submitted job
+        logger.debug("Registering the job %s within the scheduler", job)
+        otherFuture = asyncio.run_coroutine_threadsafe(
+            self.aio_registerJob(job), self.loop
+        )
+        other = otherFuture.result()
+        logger.debug("Job already submitted" if other else "First submission")
+        if other:
+            return other
+
+        job._future = asyncio.run_coroutine_threadsafe(self.aio_submit(job), self.loop)
+        return None
+
+    def prepare(self, job: Job):
+        """Prepares the job for running"""
+        logger.info("Preparing job %s", job.path)
+        job.prepare(overwrite=True)
+
+    async def aio_registerJob(self, job: Job):
+        """Register a job by adding it to the list, and checks
+        whether the job has already been submitted
+        """
+        logger.debug("Registering job %s", job)
+
+        if self.exitmode:
+            logger.warning("Exit mode: not submitting")
+
+        elif job.identifier in self.jobs:
+            other = self.jobs[job.identifier]
+            assert job.type == other.type
+            if other.state == JobState.ERROR:
+                logger.info("Re-submitting job")
+            else:
+                logger.warning("Job %s already submitted", job.identifier)
+                return other
+
+        else:
+            # Register this job
+            self.xp.unfinishedJobs += 1
+            self.jobs[job.identifier] = job
+
+        return None
+
+    def notify_job_submitted(self, job: Job):
+        """Notify the listeners that a job has been submitted"""
+        for listener in self.listeners:
+            try:
+                listener.job_submitted(job)
+            except Exception:
+                logger.exception("Got an error with listener %s", listener)
+
+    def notify_job_state(self, job: Job):
+        """Notify the listeners that a job has changed state"""
+        for listener in self.listeners:
+            try:
+                listener.job_state(job)
+            except Exception:
+                logger.exception("Got an error with listener %s", listener)
+
+    async def aio_submit(self, job: Job) -> JobState:  # noqa: C901
+        """Main scheduler function: submit a job, run it (if needed), and returns
+        the status code
+        """
+        logger.info("Submitting job %s", job)
+        job._readyEvent = asyncio.Event()
+        job.submittime = time.time()
+        job.scheduler = self
+        self.waitingjobs.add(job)
+
+        # Check that we don't have a completed job in
+        # alternate directories
+        for jobspath in experiment.current().alt_jobspaths:
+            # FIXME: check if done
+            pass
+
+        # Creates a link into the experiment folder
+        path = experiment.current().jobspath / job.relpath
+        path.parent.mkdir(parents=True, exist_ok=True)
+        if path.is_symlink():
+            path.unlink()
+        path.symlink_to(job.path)
+
+        job.state = JobState.WAITING
+
+        self.notify_job_submitted(job)
+
+        # Add dependencies, and add to blocking resources
+        if job.dependencies:
+            job.unsatisfied = len(job.dependencies)
+
+            for dependency in job.dependencies:
+                dependency.target = job
+                dependency.loop = self.loop
+                dependency.origin.dependents.add(dependency)
+                dependency.check()
+        else:
+            job._readyEvent.set()
+            job.state = JobState.READY
+
+        if job.donepath.exists():
+            job.state = JobState.DONE
+
+        # Check if we have a running process
+        process = await job.aio_process()
+        if process is not None:
+            # Yep! First we notify the listeners
+            job.state = JobState.RUNNING
+            # Notify the listeners
+            self.notify_job_state(job)
+
+            # Adds to the listeners
+            if self.xp.server is not None:
+                job.add_notification_server(self.xp.server)
+
+            # And now, we wait...
+            logger.info("Got a process for job %s - waiting to complete", job)
+            code = await process.aio_code()
+            logger.info("Job %s completed with code %s", job, code)
+            job.state = JobState.DONE if code == 0 else JobState.ERROR
+
+        # Check if done
+        if job.donepath.exists():
+            job.state = JobState.DONE
+
+        # OK, not done; let's start the job for real
+        while not job.state.finished():
+            # Wait that the job is ready
+            await job._readyEvent.wait()
+            job._readyEvent.clear()
+
+            if job.state == JobState.READY:
+                try:
+                    state = await self.aio_start(job)
+                except Exception:
+                    logger.exception("Got an exception while starting the job")
+                    raise
+
+                if state is None:
+                    # State is None if this is not the main thread
+                    return JobState.ERROR
+
+                job.state = state
+
+        self.notify_job_state(job)
+
+        # Job is finished
+        if job.state != JobState.DONE:
+            self.xp.failedJobs[job.identifier] = job
+
+        # Process all remaining tasks outputs
+        await asyncThreadcheck("End of job processing", job.done_handler)
+
+        # Decrement the number of unfinished jobs and notify
+        self.xp.unfinishedJobs -= 1
+        async with self.exitCondition:
+            logging.debug("Updated number of unfinished jobs")
+            self.exitCondition.notify_all()
+
+        job.endtime = time.time()
+        if job in self.waitingjobs:
+            self.waitingjobs.remove(job)
+
+        with job.dependents as dependents:
+            logger.info("Processing %d dependent jobs", len(dependents))
+            for dependency in dependents:
+                logger.debug("Checking dependency %s", dependency)
+                self.loop.call_soon(dependency.check)
+
+        return job.state
+
+    async def aio_start(self, job: Job) -> Optional[JobState]:
+        """Start a job (scheduler coordination layer)
+
+        This method serves as a coordination layer that delegates the actual
+        job starting logic to the job itself while handling scheduler-specific
+        concerns like state notifications and providing coordination context.
+
+        :param job: The job to start
+        :return: JobState.WAITING if dependencies could not be locked, JobState.DONE
+            if job completed successfully, JobState.ERROR if job failed during execution,
+            or None (should not occur in normal operation)
+        :raises Exception: Various exceptions during scheduler coordination
+        """
+
+        # Assert preconditions
+        assert job.launcher is not None
+
+        try:
+            # Call job's start method with scheduler context
+            state = await job.aio_start(
+                sched_dependency_lock=self.dependencyLock,
+                notification_server=self.xp.server if self.xp else None,
+            )
+
+            if state is None:
+                # Dependencies couldn't be locked, return WAITING state
+                return JobState.WAITING
+
+            # Notify scheduler listeners of job state after successful start
+            self.notify_job_state(job)
+            return state
+
+        except Exception:
+            logger.warning("Error in scheduler job coordination", exc_info=True)
+            return JobState.ERROR