コード例 #1
0
    def submit(self, fun, *args, **kwargs):
        """
        Submit a job to the pool.
        kwargs may contain __cfut_options which currently should look like:
        {
            "output_pickle_path": str
        }
        output_pickle_path defines where the pickled result should be stored.
        That file will not be removed after the job has finished.
        """
        fut = self.create_enriched_future()
        workerid = random_string()

        should_keep_output = False
        if "__cfut_options" in kwargs:
            should_keep_output = True
            output_pickle_path = kwargs["__cfut_options"]["output_pickle_path"]
            del kwargs["__cfut_options"]
        else:
            output_pickle_path = self.format_outfile_name(
                self.cfut_dir, workerid)

        self.ensure_not_shutdown()

        # Start the job.
        serialized_function_info = pickling.dumps(
            (fun, args, kwargs, self.meta_data, output_pickle_path))
        with open(self.format_infile_name(self.cfut_dir, workerid), "wb") as f:
            f.write(serialized_function_info)

        self.store_main_path_to_meta_file(workerid)

        preliminary_output_pickle_path = with_preliminary_postfix(
            output_pickle_path)
        if os.path.exists(preliminary_output_pickle_path):
            logging.warning(
                f"Deleting stale output file at {preliminary_output_pickle_path}..."
            )
            os.unlink(preliminary_output_pickle_path)

        job_name = get_function_name(fun)
        jobids_futures, _ = self._start(workerid, job_name=job_name)
        # Only a single job was submitted
        jobid = jobids_futures[0].result()

        if self.debug:
            print("job submitted: %i" % jobid, file=sys.stderr)

        # Thread will wait for it to finish.
        self.wait_thread.waitFor(preliminary_output_pickle_path, jobid)

        with self.jobs_lock:
            self.jobs[jobid] = (fut, workerid, output_pickle_path,
                                should_keep_output)

        fut.cluster_jobid = jobid
        return fut
コード例 #2
0
ファイル: slurm.py プロジェクト: scalableminds/cluster_tools
    def submit_text(cls, job, cfut_dir):
        """Submits a Slurm job represented as a job file string. Returns
        the job ID.
        """

        filename = cls.get_temp_file_path(
            cfut_dir, "_temp_slurm{}.sh".format(random_string()))
        with open(filename, "w") as f:
            f.write(job)
        job_id, stderr = chcall("sbatch --parsable {}".format(filename))
        os.unlink(filename)

        if len(stderr) > 0:
            logging.warning(f"Submitting batch job emitted warnings: {stderr}")

        return int(job_id)
コード例 #3
0
    def submit_text(self, job):
        """Submits a PBS job represented as a job file string. Returns
        the job ID.
        """

        filename = self.get_temp_file_path(
            self.cfut_dir, "_temp_pbs_{}.sh".format(random_string()))
        with open(filename, "w") as f:
            f.write(job)
        jobid_desc, _ = chcall("qsub -V {}".format(filename))
        match = re.search("^[0-9]+", jobid_desc.decode("utf-8"))
        assert match is not None
        jobid = match.group(0)

        print("jobid", jobid)
        # os.unlink(filename)
        return int(jobid)
コード例 #4
0
    def map_to_futures(self, fun, allArgs, output_pickle_path_getter=None):
        self.ensure_not_shutdown()
        allArgs = list(allArgs)
        if len(allArgs) == 0:
            return []

        should_keep_output = output_pickle_path_getter is not None

        futs_with_output_paths = []
        workerid = random_string()

        pickled_function_path = self.get_function_pickle_path(workerid)
        self.files_to_clean_up.append(pickled_function_path)
        with open(pickled_function_path, "wb") as file:
            pickling.dump(fun, file)
        self.store_main_path_to_meta_file(workerid)

        for index, arg in enumerate(allArgs):
            fut = self.create_enriched_future()
            workerid_with_index = self.get_workerid_with_index(workerid, index)

            if output_pickle_path_getter is None:
                output_pickle_path = self.format_outfile_name(
                    self.cfut_dir, workerid_with_index)
            else:
                output_pickle_path = output_pickle_path_getter(arg)

            preliminary_output_pickle_path = with_preliminary_postfix(
                output_pickle_path)
            if os.path.exists(preliminary_output_pickle_path):
                logging.warning(
                    f"Deleting stale output file at {preliminary_output_pickle_path}..."
                )
                os.unlink(preliminary_output_pickle_path)

            serialized_function_info = pickling.dumps(
                (pickled_function_path, [arg], {}, self.meta_data,
                 output_pickle_path))
            infile_name = self.format_infile_name(self.cfut_dir,
                                                  workerid_with_index)

            with open(infile_name, "wb") as f:
                f.write(serialized_function_info)

            futs_with_output_paths.append((fut, output_pickle_path))

        with self.jobs_lock:
            # Use a separate loop to avoid having to acquire the jobs_lock many times
            # or for the full duration of the above loop
            for index in range(len(futs_with_output_paths)):
                workerid_with_index = self.get_workerid_with_index(
                    workerid, index)
                # Register the job in the jobs array, although the jobid is not known yet.
                # Otherwise it might happen that self.jobs becomes empty, but some of the jobs were
                # not even submitted yet.
                self.jobs[workerid_with_index] = "pending"

        job_count = len(allArgs)
        job_name = get_function_name(fun)
        jobids_futures, job_index_ranges = self._start(workerid, job_count,
                                                       job_name)

        number_of_batches = len(jobids_futures)
        for batch_index, (jobid_future,
                          (job_index_start, job_index_end)) in enumerate(
                              zip(jobids_futures, job_index_ranges)):
            jobid_future.add_done_callback(
                partial(
                    self.register_jobs,
                    futs_with_output_paths[job_index_start:job_index_end],
                    workerid,
                    should_keep_output,
                    job_index_start,
                    f"{batch_index + 1}/{number_of_batches}",
                ))

        return [fut for (fut, _) in futs_with_output_paths]