jobs.py

__author__ = "Johannes Köster"
__copyright__ = "Copyright 2022, Johannes Köster"
__email__ = "johannes.koester@uni-due.de"
__license__ = "MIT"

import os
import sys
import base64
import tempfile
import json
import shutil

from collections import defaultdict
from itertools import chain, filterfalse
from operator import attrgetter

from snakemake.io import (
    IOFile,
    Wildcards,
    Resources,
    _IOFile,
    is_flagged,
    get_flag_value,
)
from snakemake.utils import format, listfiles
from snakemake.exceptions import RuleException, ProtectedOutputException, WorkflowError
from snakemake.logging import logger
from snakemake.common import (
    DYNAMIC_FILL,
    is_local_file,
    parse_uri,
    lazy_property,
    get_uuid,
    TBDString,
)


def format_files(job, io, dynamicio):
    for f in io:
        if f in dynamicio:
            yield f"{f.format_dynamic()} (dynamic)"
        elif is_flagged(f, "pipe"):
            yield f"{f} (pipe)"
        elif is_flagged(f, "service"):
            yield f"{f} (service)"
        elif is_flagged(f, "checkpoint_target"):
            yield TBDString()
        else:
            yield f


def jobfiles(jobs, type):
    return chain(*map(attrgetter(type), jobs))


class AbstractJob:
    def is_group(self):
        raise NotImplementedError()

    def log_info(self, skip_dynamic=False):
        raise NotImplementedError()

    def log_error(self, msg=None, **kwargs):
        raise NotImplementedError()

    def remove_existing_output(self):
        raise NotImplementedError()

    def download_remote_input(self):
        raise NotImplementedError()

    def properties(self, omit_resources=["_cores", "_nodes"], **aux_properties):
        raise NotImplementedError()

    def reset_params_and_resources(self):
        raise NotImplementedError()


class JobFactory:
    def __init__(self):
        self.cache = dict()

    def new(
        self,
        rule,
        dag,
        wildcards_dict=None,
        format_wildcards=None,
        targetfile=None,
        update=False,
        groupid=None,
    ):
        if rule.is_branched:
            # for distinguishing branched rules, we need input and output in addition
            key = (
                rule.name,
                *rule.output,
                *rule.input,
                *sorted(wildcards_dict.items()),
            )
        else:
            key = (rule.name, *sorted(wildcards_dict.items()))
        if update:
            # cache entry has to be replaced because job shall be constructed from scratch
            obj = Job(rule, dag, wildcards_dict, format_wildcards, targetfile, groupid)
            self.cache[key] = obj
        else:
            try:
                # try to get job from cache
                obj = self.cache[key]
            except KeyError:
                obj = Job(rule, dag, wildcards_dict, format_wildcards, targetfile)
                self.cache[key] = obj
        return obj


class Job(AbstractJob):
    HIGHEST_PRIORITY = sys.maxsize

    obj_cache = dict()

    __slots__ = [
        "rule",
        "dag",
        "wildcards_dict",
        "wildcards",
        "_format_wildcards",
        "input",
        "dependencies",
        "output",
        "_params",
        "_log",
        "_benchmark",
        "_resources",
        "_conda_env_file",
        "_conda_env",
        "shadow_dir",
        "_inputsize",
        "dynamic_output",
        "dynamic_input",
        "temp_output",
        "protected_output",
        "touch_output",
        "subworkflow_input",
        "_hash",
        "_attempt",
        "_group",
        "targetfile",
        "incomplete_input_expand",
    ]

    def __init__(
        self,
        rule,
        dag,
        wildcards_dict=None,
        format_wildcards=None,
        targetfile=None,
        groupid=None,
    ):
        self.rule = rule
        self.dag = dag

        # the targetfile that led to the job
        # it is important to record this, since we need it to submit the
        # job on a cluster. In contrast, an arbitrary targetfile could
        # lead to a different composition of wildcard values (in case of
        # ambiguity in matching).
        self.targetfile = targetfile
        self.wildcards_dict = wildcards_dict
        self.wildcards = Wildcards(fromdict=self.wildcards_dict)
        self._format_wildcards = (
            self.wildcards
            if format_wildcards is None
            else Wildcards(fromdict=format_wildcards)
        )

        (
            self.input,
            input_mapping,
            self.dependencies,
            self.incomplete_input_expand,
        ) = self.rule.expand_input(
            self.wildcards_dict,
            groupid=groupid,
        )

        self.output, output_mapping = self.rule.expand_output(self.wildcards_dict)
        # other properties are lazy to be able to use additional parameters and check already existing files
        self._params = None
        self._log = None
        self._benchmark = None
        self._resources = None
        self._conda_env_spec = None
        self._conda_env = None
        self._group = None

        self.shadow_dir = None
        self._inputsize = None
        self.is_updated = False

        self._attempt = self.dag.workflow.attempt

        # TODO get rid of these
        self.dynamic_output, self.dynamic_input = set(), set()
        self.temp_output, self.protected_output = set(), set()
        self.touch_output = set()
        self.subworkflow_input = dict()
        for f in self.output:
            f_ = output_mapping[f]
            if f_ in self.rule.dynamic_output:
                self.dynamic_output.add(f)
            if f_ in self.rule.temp_output:
                self.temp_output.add(f)
            if f_ in self.rule.protected_output:
                self.protected_output.add(f)
            if f_ in self.rule.touch_output:
                self.touch_output.add(f)
        for f in self.input:
            f_ = input_mapping[f]
            if f_ in self.rule.dynamic_input:
                self.dynamic_input.add(f)
            if f_ in self.rule.subworkflow_input:
                self.subworkflow_input[f] = self.rule.subworkflow_input[f_]
            elif "subworkflow" in f.flags:
                sub = f.flags["subworkflow"]
                if f in self.subworkflow_input:
                    other = self.subworkflow_input[f]
                    if sub != other:
                        raise WorkflowError(
                            "The input file {} is ambiguously "
                            "associated with two subworkflows {} "
                            "and {}.".format(f, sub, other),
                            rule=self.rule,
                        )
                self.subworkflow_input[f] = sub

    def updated(self):
        group = self.dag.get_job_group(self)
        groupid = None
        if group is None:
            if self.dag.workflow.run_local or self.is_local:
                groupid = self.dag.workflow.local_groupid
        else:
            groupid = group.jobid

        job = self.dag.job_factory.new(
            self.rule,
            self.dag,
            wildcards_dict=self.wildcards_dict,
            targetfile=self.targetfile,
            update=True,
            groupid=groupid,
        )
        job.is_updated = True
        return job

    def is_valid(self):
        """Check if job is valid"""
        # these properties have to work in dry-run as well. Hence we check them here:
        self.rule.expand_benchmark(self.wildcards_dict)
        self.rule.expand_log(self.wildcards_dict)

    def outputs_older_than_script_or_notebook(self):
        """return output that's older than script, i.e. script has changed"""
        path = self.rule.script or self.rule.notebook
        if not path:
            return
        if self.rule.basedir:
            # needed if rule is included from another subdirectory
            path = self.rule.basedir.join(path).get_path_or_uri()
        if is_local_file(path) and os.path.exists(path):
            script_mtime = os.lstat(path).st_mtime
            for f in self.expanded_output:
                if f.exists:
                    if not f.is_newer(script_mtime):
                        yield f
        # TODO also handle remote file case here.

    @property
    def threads(self):
        return self.resources._cores

    @property
    def params(self):
        if self._params is None:
            self._params = self.rule.expand_params(
                self.wildcards_dict, self.input, self.output, self.resources
            )
        return self._params

    @property
    def log(self):
        if self._log is None:
            self._log = self.rule.expand_log(self.wildcards_dict)
        return self._log

    @property
    def benchmark(self):
        if self._benchmark is None:
            self._benchmark = self.rule.expand_benchmark(self.wildcards_dict)
        return self._benchmark

    @property
    def benchmark_repeats(self):
        if self.benchmark is not None:
            return get_flag_value(self.benchmark, "repeat") or 1

    @property
    def group(self):
        if self._group is None:
            self._group = self.rule.expand_group(self.wildcards_dict)
        return self._group

    @group.setter
    def group(self, group):
        self._group = group

    @property
    def attempt(self):
        return self._attempt

    @attempt.setter
    def attempt(self, attempt):
        # reset resources
        self._resources = None
        self._attempt = attempt

    @property
    def resources(self):
        if self._resources is None:
            self._resources = self.rule.expand_resources(
                self.wildcards_dict, self.input, self.attempt
            )
        return self._resources

    def reset_params_and_resources(self):
        self._resources = None
        self._params = None

    @property
    def conda_env_spec(self):
        if self._conda_env_spec is None:
            self._conda_env_spec = self.rule.expand_conda_env(self.wildcards_dict)
        return self._conda_env_spec

    @property
    def conda_env(self):
        if self.conda_env_spec:
            if self._conda_env is None:
                self._conda_env = self.dag.conda_envs.get(
                    (self.conda_env_spec, self.container_img_url)
                )
            return self._conda_env
        return None

    def archive_conda_env(self):
        """Archive a conda environment into a custom local channel."""
        if self.conda_env_spec:
            if self.conda_env.is_named:
                raise WorkflowError(
                    "Workflow archives cannot be created for workflows using named conda environments."
                    "Please use paths to YAML files for all your conda directives.",
                    rule=self.rule,
                )
            return self.conda_env.create_archive()
        return None

    @property
    def needs_singularity(self):
        return self.container_img is not None

    @property
    def container_img_url(self):
        return self.rule.container_img

    @property
    def is_containerized(self):
        return self.rule.is_containerized

    @property
    def container_img(self):
        if self.dag.workflow.use_singularity and self.container_img_url:
            return self.dag.container_imgs[self.container_img_url]
        return None

    @property
    def env_modules(self):
        return self.rule.env_modules

    @property
    def container_img_path(self):
        return self.container_img.path if self.container_img else None

    @property
    def is_shadow(self):
        return self.rule.shadow_depth is not None

    @property
    def priority(self):
        return self.dag.priority(self)

    @property
    def b64id(self):
        return base64.b64encode(
            (self.rule.name + "".join(self.output)).encode("utf-8")
        ).decode("utf-8")

    @property
    def inputsize(self):
        """
        Return the size of the input files.
        Input files need to be present.
        """
        if self._inputsize is None:
            self._inputsize = sum(f.size for f in self.input)
        return self._inputsize

    @property
    def message(self):
        """Return the message for this job."""
        try:
            return (
                self.format_wildcards(self.rule.message) if self.rule.message else None
            )
        except AttributeError as ex:
            raise RuleException(str(ex), rule=self.rule)
        except KeyError as ex:
            raise RuleException(
                "Unknown variable in message " "of shell command: {}".format(str(ex)),
                rule=self.rule,
            )

    @property
    def shellcmd(self):
        """Return the shell command."""
        try:
            return (
                self.format_wildcards(self.rule.shellcmd)
                if self.rule.shellcmd
                else None
            )
        except AttributeError as ex:
            raise RuleException(str(ex), rule=self.rule)
        except KeyError as ex:
            raise RuleException(
                "Unknown variable when printing " "shell command: {}".format(str(ex)),
                rule=self.rule,
            )

    @property
    def is_shell(self):
        return self.rule.is_shell

    @property
    def is_norun(self):
        return self.rule.norun

    @property
    def is_script(self):
        return self.rule.is_script

    @property
    def is_notebook(self):
        return self.rule.is_notebook

    @property
    def is_wrapper(self):
        return self.rule.is_wrapper

    @property
    def is_cwl(self):
        return self.rule.is_cwl

    @property
    def is_template_engine(self):
        return self.rule.is_template_engine

    @property
    def is_run(self):
        return not (
            self.is_shell
            or self.is_norun
            or self.is_script
            or self.is_notebook
            or self.is_wrapper
            or self.is_cwl
        )

    @property
    def is_pipe(self):
        return any(is_flagged(o, "pipe") for o in self.output)

    @property
    def is_service(self):
        return any(is_flagged(o, "service") for o in self.output)

    @property
    def expanded_output(self):
        """Iterate over output files while dynamic output is expanded."""
        for f, f_ in zip(self.output, self.rule.output):
            if f in self.dynamic_output:
                expansion = self.expand_dynamic(f_)
                if not expansion:
                    yield f_
                for f, _ in expansion:
                    file_to_yield = IOFile(f, self.rule)
                    file_to_yield.clone_flags(f_)
                    yield file_to_yield
            else:
                yield f

    def shadowed_path(self, f):
        """Get the shadowed path of IOFile f."""
        if not self.shadow_dir:
            return f
        f_ = IOFile(os.path.join(self.shadow_dir, f), self.rule)
        f_.clone_flags(f)
        return f_

    @property
    def dynamic_wildcards(self):
        """Return all wildcard values determined from dynamic output."""
        combinations = set()
        for f, f_ in zip(self.output, self.rule.output):
            if f in self.dynamic_output:
                for f, w in self.expand_dynamic(f_):
                    combinations.add(tuple(w.items()))
        wildcards = defaultdict(list)
        for combination in combinations:
            for name, value in combination:
                wildcards[name].append(value)
        return wildcards

    @property
    def missing_input(self):
        """Return missing input files."""
        # omit file if it comes from a subworkflow
        return set(
            f for f in self.input if not f.exists and not f in self.subworkflow_input
        )

    @property
    def existing_remote_input(self):
        files = set()

        for f in self.input:
            if f.is_remote:
                if f.exists_remote:
                    files.add(f)
        return files

    @property
    def existing_remote_output(self):
        files = set()

        for f in self.remote_output:
            if f.exists_remote:
                files.add(f)
        return files

    @property
    def missing_remote_input(self):
        return self.remote_input - self.existing_remote_input

    @property
    def missing_remote_output(self):
        return self.remote_output - self.existing_remote_output

    @property
    def output_mintime(self):
        """Return oldest output file."""
        try:
            mintime = min(
                f.mtime.local_or_remote() for f in self.expanded_output if f.exists
            )
        except ValueError:
            # no existing output
            mintime = None

        if self.benchmark and self.benchmark.exists:
            mintime_benchmark = self.benchmark.mtime.local_or_remote()
            if mintime is not None:
                return min(mintime, mintime_benchmark)
            else:
                return mintime_benchmark

        return mintime

    def missing_output(self, requested):
        def handle_file(f):
            # pipe or service output is always declared as missing
            # (even if it might be present on disk for some reason)
            if is_flagged(f, "pipe") or is_flagged(f, "service") or not f.exists:
                yield f

        if self.dynamic_output:
            for f, f_ in zip(self.output, self.rule.output):
                if f in requested:
                    if f in self.dynamic_output:
                        if not self.expand_dynamic(f_):
                            yield "{} (dynamic)".format(f_)
                    else:
                        yield from handle_file(f)
        else:
            for f in requested:
                yield from handle_file(f)

    @property
    def local_input(self):
        for f in self.input:
            if not f.is_remote:
                yield f

    @property
    def unique_input(self):
        seen = set()

        for element in filterfalse(seen.__contains__, self.input):
            seen.add(element)
            yield element

    @property
    def local_output(self):
        for f in self.output:
            if not f.is_remote:
                yield f

    @property
    def remote_input(self):
        for f in self.input:
            if f.is_remote:
                yield f

    @property
    def remote_output(self):
        for f in self.output:
            if f.is_remote:
                yield f

    @property
    def remote_input_newer_than_local(self):
        files = set()
        for f in self.remote_input:
            if (f.exists_remote and f.exists_local) and (
                f.mtime.remote() > f.mtime.local(follow_symlinks=True)
            ):
                files.add(f)
        return files

    @property
    def remote_input_older_than_local(self):
        files = set()
        for f in self.remote_input:
            if (f.exists_remote and f.exists_local) and (
                f.mtime.remote() < f.mtime.local(follow_symlinks=True)
            ):
                files.add(f)
        return files

    @property
    def remote_output_newer_than_local(self):
        files = set()
        for f in self.remote_output:
            if (f.exists_remote and f.exists_local) and (
                f.mtime.remote() > f.mtime.local(follow_symlinks=True)
            ):
                files.add(f)
        return files

    @property
    def remote_output_older_than_local(self):
        files = set()
        for f in self.remote_output:
            if (f.exists_remote and f.exists_local) and (
                f.mtime.remote() < f.mtime.local(follow_symlinks=True)
            ):
                files.add(f)
        return files

    @property
    def files_to_download(self):
        toDownload = set()

        for f in self.input:
            if f.is_remote:
                if (not f.exists_local and f.exists_remote) and (
                    not self.rule.norun or f.remote_object.keep_local
                ):
                    toDownload.add(f)

        toDownload = toDownload | self.remote_input_newer_than_local
        return toDownload

    @property
    def files_to_upload(self):
        return self.missing_remote_input & self.remote_input_older_than_local

    @property
    def existing_output(self):
        return filter(lambda f: f.exists, self.expanded_output)

    def check_protected_output(self):
        protected = list(filter(lambda f: f.protected, self.expanded_output))
        if protected:
            raise ProtectedOutputException(self, protected)

    def remove_existing_output(self):
        """Clean up both dynamic and regular output before rules actually run"""
        if self.dynamic_output:
            for f, _ in chain(*map(self.expand_dynamic, self.rule.dynamic_output)):
                os.remove(f)

        for f, f_ in zip(self.output, self.rule.output):
            try:
                # remove_non_empty_dir only applies to directories which aren't
                # flagged with directory().
                f.remove(remove_non_empty_dir=False)
            except FileNotFoundError:
                # No file == no problem
                pass

        for f in self.log:
            f.remove(remove_non_empty_dir=False)

    def download_remote_input(self):
        for f in self.files_to_download:
            f.download_from_remote()

    def prepare(self):
        """
        Prepare execution of job.
        This includes creation of directories and deletion of previously
        created dynamic files.
        Creates a shadow directory for the job if specified.
        """

        self.check_protected_output()

        unexpected_output = self.dag.reason(self).missing_output.intersection(
            self.existing_output
        )
        if unexpected_output:
            logger.warning(
                "Warning: the following output files of rule {} were not "
                "present when the DAG was created:\n{}".format(
                    self.rule, unexpected_output
                )
            )

        self.remove_existing_output()

        # Create tmpdir if necessary
        if self.resources.get("tmpdir"):
            os.makedirs(self.resources.tmpdir, exist_ok=True)

        for f, f_ in zip(self.output, self.rule.output):
            f.prepare()

        self.download_remote_input()

        for f in self.log:
            f.prepare()
        if self.benchmark:
            self.benchmark.prepare()

        if not self.is_shadow:
            return

        # Create shadow directory structure
        self.shadow_dir = tempfile.mkdtemp(
            dir=self.rule.workflow.persistence.shadow_path
        )
        cwd = os.getcwd()

        # "minimal" creates symlinks only to the input files in the shadow directory
        # "copy-minimal" creates copies instead
        if (
            self.rule.shadow_depth == "minimal"
            or self.rule.shadow_depth == "copy-minimal"
        ):
            # Re-create the directory structure in the shadow directory
            for (f, d) in set(
                [
                    (item, os.path.dirname(item))
                    for sublist in [self.input, self.output, self.log]
                    if sublist is not None
                    for item in sublist
                ]
            ):
                if d and not os.path.isabs(d):
                    rel_path = os.path.relpath(d)
                    # Only create subdirectories
                    if not rel_path.split(os.path.sep)[0] == "..":
                        os.makedirs(
                            os.path.join(self.shadow_dir, rel_path), exist_ok=True
                        )
                    else:
                        raise RuleException(
                            "The following file name references a parent directory relative to your workdir.\n"
                            'This isn\'t supported for shadow: "{}". Consider using an absolute path instead.\n{}'.format(
                                f, self.rule.shadow_depth
                            ),
                            rule=self.rule,
                        )

            # Symlink or copy the input files
            if self.rule.shadow_depth == "copy-minimal":
                for rel_path in set(
                    [os.path.relpath(f) for f in self.input if not os.path.isabs(f)]
                ):
                    copy = os.path.join(self.shadow_dir, rel_path)
                    shutil.copy(rel_path, copy)
            else:
                for rel_path in set(
                    [os.path.relpath(f) for f in self.input if not os.path.isabs(f)]
                ):
                    link = os.path.join(self.shadow_dir, rel_path)
                    original = os.path.relpath(rel_path, os.path.dirname(link))
                    os.symlink(original, link)

        # Shallow simply symlink everything in the working directory.
        elif self.rule.shadow_depth == "shallow":
            for source in os.listdir(cwd):
                link = os.path.join(self.shadow_dir, source)
                os.symlink(os.path.abspath(source), link)
        elif self.rule.shadow_depth == "full":
            snakemake_dir = os.path.join(cwd, ".snakemake")
            for dirpath, dirnames, filenames in os.walk(cwd):
                # Must exclude .snakemake and its children to avoid infinite
                # loop of symlinks.
                if os.path.commonprefix([snakemake_dir, dirpath]) == snakemake_dir:
                    continue
                for dirname in dirnames:
                    if dirname == ".snakemake":
                        continue
                    relative_source = os.path.relpath(os.path.join(dirpath, dirname))
                    shadow = os.path.join(self.shadow_dir, relative_source)
                    os.mkdir(shadow)

                for filename in filenames:
                    source = os.path.join(dirpath, filename)
                    relative_source = os.path.relpath(source)
                    link = os.path.join(self.shadow_dir, relative_source)
                    os.symlink(source, link)

    def close_remote(self):
        for f in self.input + self.output:
            if f.is_remote:
                f.remote_object.close()

    def cleanup(self):
        """Cleanup output files."""
        to_remove = [f for f in self.expanded_output if f.exists]

        to_remove.extend(
            [
                f
                for f in self.remote_output
                if (
                    f.exists_remote
                    if (f.is_remote and f.should_stay_on_remote)
                    else f.exists_local
                )
            ]
        )
        if to_remove:
            logger.info(
                "Removing output files of failed job {}"
                " since they might be corrupted:\n{}".format(self, ", ".join(to_remove))
            )
            for f in to_remove:
                f.remove()

    def format_wildcards(self, string, **variables):
        """Format a string with variables from the job."""
        _variables = dict()
        _variables.update(self.rule.workflow.globals)
        _variables.update(
            dict(
                input=self.input,
                output=self.output,
                params=self.params,
                wildcards=self._format_wildcards,
                threads=self.threads,
                resources=self.resources,
                log=self.log,
                jobid=self.jobid,
                version=self.rule.version,
                name=self.name,
                rule=self.rule.name,
                rulename=self.rule.name,
                bench_iteration=None,
            )
        )
        _variables.update(variables)
        try:
            return format(string, **_variables)
        except NameError as ex:
            raise RuleException("NameError: " + str(ex), rule=self.rule)
        except IndexError as ex:
            raise RuleException("IndexError: " + str(ex), rule=self.rule)

    def properties(self, omit_resources=["_cores", "_nodes"], **aux_properties):
        resources = {
            name: res
            for name, res in self.resources.items()
            if name not in omit_resources
        }
        params = {name: value for name, value in self.params.items()}
        properties = {
            "type": "single",
            "rule": self.rule.name,
            "local": self.is_local,
            "input": self.input,
            "output": self.output,
            "wildcards": self.wildcards_dict,
            "params": params,
            "log": self.log,
            "threads": self.threads,
            "resources": resources,
            "jobid": self.dag.jobid(self),
        }
        properties.update(aux_properties)

        try:
            return json.dumps(properties)
        except TypeError:
            del properties["params"]
            return json.dumps(properties)

    @property
    def is_local(self):
        return self.dag.workflow.is_local(self.rule)

    def __repr__(self):
        return self.rule.name

    def __lt__(self, other):
        return self.rule.__lt__(other.rule)

    def __gt__(self, other):
        return self.rule.__gt__(other.rule)

    def expand_dynamic(self, pattern):
        """Expand dynamic files."""
        return list(
            listfiles(pattern, restriction=self.wildcards, omit_value=DYNAMIC_FILL)
        )

    def is_group(self):
        return False

    def log_info(self, skip_dynamic=False, indent=False, printshellcmd=True):
        # skip dynamic jobs that will be "executed" only in dryrun mode
        if skip_dynamic and self.dag.dynamic(self):
            return

        priority = self.priority
        logger.job_info(
            jobid=self.dag.jobid(self),
            msg=self.message,
            name=self.rule.name,
            local=self.dag.workflow.is_local(self.rule),
            input=list(format_files(self, self.input, self.dynamic_input)),
            output=list(format_files(self, self.output, self.dynamic_output)),
            log=list(self.log),
            benchmark=self.benchmark,
            wildcards=self.wildcards_dict,
            reason=str(self.dag.reason(self)),
            resources=self.resources,
            priority="highest" if priority == Job.HIGHEST_PRIORITY else priority,
            threads=self.threads,
            indent=indent,
            is_checkpoint=self.rule.is_checkpoint,
            printshellcmd=printshellcmd,
            is_handover=self.rule.is_handover,
        )
        logger.shellcmd(self.shellcmd, indent=indent)

        if self.dynamic_output:
            logger.info(
                "Subsequent jobs will be added dynamically "
                "depending on the output of this job",
                indent=True,
            )

    def log_error(self, msg=None, indent=False, **kwargs):
        logger.job_error(
            name=self.rule.name,
            jobid=self.dag.jobid(self),
            output=list(format_files(self, self.output, self.dynamic_output)),
            log=list(self.log),
            conda_env=self.conda_env.address if self.conda_env else None,
            aux=kwargs,
            indent=indent,
            shellcmd=self.shellcmd,
        )
        if msg is not None:
            logger.error(msg)

    def register(self):
        self.dag.workflow.persistence.started(self)

    def get_wait_for_files(self):
        wait_for_files = []
        wait_for_files.extend(self.local_input)
        wait_for_files.extend(
            f for f in self.remote_input if not f.should_stay_on_remote
        )

        if self.shadow_dir:
            wait_for_files.append(self.shadow_dir)
        if (
            self.dag.workflow.use_conda
            and self.conda_env
            and not self.conda_env.is_named
        ):
            wait_for_files.append(self.conda_env.address)
        return wait_for_files

    @property
    def jobid(self):
        return self.dag.jobid(self)

    def uuid(self):
        return str(
            get_uuid(
                f"{self.rule.name}:{','.join(sorted(f'{w}:{v}' for w, v in self.wildcards_dict.items()))}"
            )
        )

    def postprocess(
        self,
        upload_remote=True,
        handle_log=True,
        handle_touch=True,
        handle_temp=True,
        error=False,
        ignore_missing_output=False,
        assume_shared_fs=True,
        latency_wait=None,
        keep_metadata=True,
    ):
        if self.dag.is_edit_notebook_job(self):
            # No postprocessing necessary, we have just created the skeleton notebook and
            # execution will anyway stop afterwards.
            return
        if assume_shared_fs:
            if not error and handle_touch:
                self.dag.handle_touch(self)
            if handle_log:
                self.dag.handle_log(self)
            if not error:
                self.dag.check_and_touch_output(
                    self, wait=latency_wait, ignore_missing_output=ignore_missing_output
                )
            self.dag.unshadow_output(self, only_log=error)
            if not error:
                self.dag.handle_remote(self, upload=upload_remote)
                self.dag.handle_protected(self)
            self.close_remote()
        else:
            if not error:
                self.dag.check_and_touch_output(
                    self, wait=latency_wait, no_touch=True, force_stay_on_remote=True
                )
        if not error:
            try:
                self.dag.workflow.persistence.finished(
                    self, keep_metadata=keep_metadata
                )
            except IOError as e:
                raise WorkflowError(
                    "Error recording metadata for finished job "
                    "({}). Please ensure write permissions for the "
                    "directory {}".format(e, self.dag.workflow.persistence.path)
                )
            if handle_temp:
                # temp handling has to happen after calling finished(),
                # because we need to access temp output files to record
                # start and end times.
                self.dag.handle_temp(self)

    @property
    def name(self):
        return self.rule.name

    @property
    def priority(self):
        return self.dag.priority(self)

    @property
    def products(self):
        products = list(self.output)
        if self.benchmark:
            products.append(self.benchmark)
        products.extend(self.log)
        return products

    def get_targets(self):
        return self.targetfile or [self.rule.name]

    @property
    def is_branched(self):
        return self.rule.is_branched

    @property
    def rules(self):
        return [self.rule.name]

    @property
    def restart_times(self):
        return self.rule.restart_times

    @property
    def is_checkpoint(self):
        return self.rule.is_checkpoint

    def __len__(self):
        return 1


class GroupJobFactory:
    def __init__(self):
        self.cache = dict()

    def new(self, id, jobs):
        jobs = frozenset(jobs)
        key = (id, jobs)
        try:
            obj = self.cache[key]
        except KeyError:
            obj = GroupJob(id, jobs)
            self.cache[key] = obj
        return obj


class GroupJob(AbstractJob):

    obj_cache = dict()

    __slots__ = [
        "groupid",
        "jobs",
        "_resources",
        "_input",
        "_output",
        "_log",
        "_inputsize",
        "_all_products",
        "_attempt",
        "toposorted",
        "_jobid",
    ]

    def __init__(self, id, jobs):
        self.groupid = id
        self.jobs = jobs
        self.toposorted = None
        self._resources = None
        self._input = None
        self._output = None
        self._log = None
        self._inputsize = None
        self._all_products = None
        self._attempt = self.dag.workflow.attempt
        self._jobid = None

    @property
    def dag(self):
        return next(iter(self.jobs)).dag

    def merge(self, other):
        assert other.groupid == self.groupid
        self.jobs = self.jobs | other.jobs

    def finalize(self):
        from toposort import toposort

        if self.toposorted is None:

            def get_dependencies(job):
                for dep, files in self.dag.dependencies[job].items():
                    if dep in self.jobs:
                        yield dep
                        if any(is_flagged(f, "pipe") for f in files):
                            # In case of a pipe, inherit the dependencies of the producer,
                            # such that the two jobs end up on the same toposort level.
                            # This is important because they are executed simulataneously.
                            yield from get_dependencies(dep)

            dag = {job: set(get_dependencies(job)) for job in self.jobs}

            self.toposorted = list(toposort(dag))

    @property
    def all_products(self):
        if self._all_products is None:
            self._all_products = set(f for job in self.jobs for f in job.products)
        return self._all_products

    def __iter__(self):
        if self.toposorted is None:
            yield from self.jobs
        else:
            yield from chain.from_iterable(self.toposorted)

    def __repr__(self):
        return "JobGroup({},{})".format(self.groupid, repr(self.jobs))

    def __contains__(self, job):
        return job in self.jobs

    def is_group(self):
        return True

    @property
    def is_checkpoint(self):
        return any(job.is_checkpoint for job in self.jobs)

    @property
    def is_updated(self):
        return any(job.is_updated for job in self.jobs)

    def log_info(self, skip_dynamic=False):
        logger.group_info(groupid=self.groupid)
        for job in sorted(self.jobs, key=lambda j: j.rule.name):
            job.log_info(skip_dynamic, indent=True)

    def log_error(self, msg=None, **kwargs):
        logger.group_error(groupid=self.groupid)
        for job in self.jobs:
            job.log_error(msg=msg, indent=True, **kwargs)

    def register(self):
        for job in self.jobs:
            job.register()

    def remove_existing_output(self):
        for job in self.jobs:
            job.remove_existing_output()

    def reset_params_and_resources(self):
        for job in self.jobs:
            job.reset_params_and_resources()

    def download_remote_input(self):
        for job in self.jobs:
            job.download_remote_input()

    def get_wait_for_files(self):
        local_input = [
            f
            for job in self.jobs
            for f in job.local_input
            if f not in self.all_products
        ]
        remote_input = [
            f
            for job in self.jobs
            for f in job.remote_input
            if f not in self.all_products
        ]

        wait_for_files = []
        wait_for_files.extend(local_input)
        wait_for_files.extend(f for f in remote_input if not f.should_stay_on_remote)

        for job in self.jobs:
            if job.shadow_dir:
                wait_for_files.append(job.shadow_dir)
            if (
                self.dag.workflow.use_conda
                and job.conda_env
                and not job.conda_env.is_named
            ):
                wait_for_files.append(job.conda_env.address)
        return wait_for_files

    @property
    def resources(self):
        if self._resources is None:

            def check_string_resource(res, value1, value2):
                if value1 != value2:
                    raise WorkflowError(
                        "Failed to group jobs together. Resource {} "
                        "is a string but not all group jobs require the same value. "
                        "Observed: {} != {}.".format(res, value1, value2)
                    )

            self._resources = defaultdict(int)
            self._resources["_nodes"] = 1
            pipe_or_service_group = any(
                [job.is_pipe or job.is_service for job in self.jobs]
            )
            # iterate over siblings that can be executed in parallel
            for siblings in self.toposorted:
                sibling_resources = defaultdict(int)
                for job in siblings:
                    try:
                        job_resources = job.resources
                    except FileNotFoundError:
                        # Skip job if resource evaluation leads to a file not found error.
                        # This will be caused by an inner job, which needs files created by the same group.
                        # All we can do is to ignore such jobs for now.
                        continue
                    for res, value in job_resources.items():
                        if isinstance(value, int):
                            if res != "_nodes":
                                sibling_resources[res] += value
                        elif isinstance(value, TBDString):
                            # we omit TBDs
                            continue
                        else:
                            # all string resources must be the same for all group jobs
                            if res in sibling_resources:
                                check_string_resource(
                                    res, sibling_resources[res], value
                                )
                            else:
                                sibling_resources[res] = value

                for res, value in sibling_resources.items():
                    if isinstance(value, int):
                        if res != "_nodes":
                            if self.dag.workflow.run_local or pipe_or_service_group:
                                # in case of local execution, this must be a
                                # group of jobs that are connected with pipes
                                # and have to run simultaneously
                                self._resources[res] += value
                            else:
                                # take the maximum with previous values
                                self._resources[res] = max(
                                    self._resources.get(res, 0), value
                                )
                    elif isinstance(value, TBDString):
                        # we omit TBDs
                        continue
                    else:
                        # all string resources must be the same for all group jobs
                        if res in self._resources:
                            check_string_resource(res, self._resources[res], value)
                        else:
                            self._resources[res] = value

        return Resources(fromdict=self._resources)

    @property
    def input(self):
        if self._input is None:
            self._input = [
                f for job in self.jobs for f in job.input if f not in self.all_products
            ]
        return self._input

    @property
    def output(self):
        all_input = set(f for job in self.jobs for f in job.input)
        if self._output is None:
            self._output = [
                f for job in self.jobs for f in job.output if f not in all_input
            ]
        return self._output

    @property
    def log(self):
        if self._log is None:
            self._log = [f for job in self.jobs for f in job.log]
        return self._log

    @property
    def products(self):
        all_input = set(f for job in self.jobs for f in job.input)
        return [f for job in self.jobs for f in job.products if f not in all_input]

    def properties(self, omit_resources=["_cores", "_nodes"], **aux_properties):
        resources = {
            name: res
            for name, res in self.resources.items()
            if name not in omit_resources
        }
        properties = {
            "type": "group",
            "groupid": self.groupid,
            "local": self.is_local,
            "input": self.input,
            "output": self.output,
            "threads": self.threads,
            "resources": resources,
            "jobid": self.jobid,
        }
        properties.update(aux_properties)

        return json.dumps(properties)

    @property
    def jobid(self):
        if not self._jobid:
            # The uuid of the last job is sufficient to uniquely identify the group job.
            # This is true because each job can only occur in one group job.
            # Additionally, this is the most stable id we can get, even if the group
            # changes by adding more upstream jobs, e.g. due to groupid usage in input
            # functions (see Dag.update_incomplete_input_expand_jobs())
            last_job = sorted(self.toposorted[-1])[-1]
            self._jobid = last_job.uuid()
        return self._jobid

    def cleanup(self):
        for job in self.jobs:
            job.cleanup()

    def postprocess(self, error=False, **kwargs):
        for job in self.jobs:
            job.postprocess(handle_temp=False, error=error, **kwargs)
        # Handle temp after per-job postprocess.
        # This is necessary because group jobs are not topologically sorted,
        # and we might otherwise delete a temp input file before it has been
        # postprocessed by the outputting job in the same group.
        if not error:
            for job in self.jobs:
                self.dag.handle_temp(job)
        # remove all pipe and service outputs since all jobs of this group are done and the
        # outputs are no longer needed
        for job in self.jobs:
            for f in job.output:
                if is_flagged(f, "pipe") or is_flagged(f, "service"):
                    f.remove()

    @property
    def name(self):
        return str(self.groupid)

    def check_protected_output(self):
        for job in self.jobs:
            job.check_protected_output()

    @property
    def dynamic_input(self):
        return [
            f
            for job in self.jobs
            for f in job.dynamic_input
            if f not in self.all_products
        ]

    @property
    def inputsize(self):
        if self._inputsize is None:
            self._inputsize = sum(f.size for f in self.input)
        return self._inputsize

    @property
    def priority(self):
        return max(self.dag.priority(job) for job in self.jobs)

    @property
    def is_local(self):
        return all(job.is_local for job in self.jobs)

    def format_wildcards(self, string, **variables):
        """Format a string with variables from the job."""
        _variables = dict()
        _variables.update(self.dag.workflow.globals)
        _variables.update(
            dict(
                input=self.input,
                output=self.output,
                threads=self.threads,
                jobid=self.jobid,
                name=self.name,
                rule="GROUP",
                rulename="GROUP",
                resources=self.resources,
            )
        )
        _variables.update(variables)
        try:
            return format(string, **_variables)
        except NameError as ex:
            raise WorkflowError(
                "NameError with group job {}: {}".format(self.jobid, str(ex))
            )
        except IndexError as ex:
            raise WorkflowError(
                "IndexError with group job {}: {}".format(self.jobid, str(ex))
            )

    @property
    def threads(self):
        return self.resources["_cores"]

    def get_targets(self):
        # jobs without output are targeted by rule name
        targets = [job.rule.name for job in self.jobs if not job.products]
        targets.extend(self.products)
        return targets

    @property
    def attempt(self):
        return self._attempt

    @attempt.setter
    def attempt(self, attempt):
        # reset resources
        self._resources = None
        self._attempt = attempt

    @property
    def is_branched(self):
        return any(job.is_branched for job in self.jobs)

    @property
    def needs_singularity(self):
        return any(job.needs_singularity for job in self.jobs)

    @property
    def rules(self):
        return [job.rule.name for job in self.jobs]

    @property
    def expanded_output(self):
        """Yields the entire expanded output of all jobs"""
        for job in self.jobs:
            yield from job.expanded_output

    @property
    def restart_times(self):
        return max(job.restart_times for job in self.jobs)

    def __len__(self):
        return len(self.jobs)

    def __hash__(self):
        return hash(self.jobs)

    def __eq__(self, other):
        if not isinstance(other, AbstractJob):
            return False
        if other.is_group():
            return self.jobs == other.jobs
        else:
            return False


class Reason:

    __slots__ = [
        "_updated_input",
        "_updated_input_run",
        "_missing_output",
        "_incomplete_output",
        "forced",
        "noio",
        "nooutput",
        "derived",
        "pipe",
        "service",
        "target",
        "finished",
    ]

    def __init__(self):
        self.finished = False
        self._updated_input = None
        self._updated_input_run = None
        self._missing_output = None
        self._incomplete_output = None
        self.forced = False
        self.noio = False
        self.nooutput = False
        self.derived = True
        self.pipe = False
        self.service = False

    @lazy_property
    def updated_input(self):
        return set()

    @lazy_property
    def updated_input_run(self):
        return set()

    @lazy_property
    def missing_output(self):
        return set()

    @lazy_property
    def incomplete_output(self):
        return set()

    def mark_finished(self):
        "called if the job has been run"
        self.finished = True

    def __str__(self):
        s = list()
        if self.forced:
            s.append("Forced execution")
        else:
            if self.noio:
                s.append(
                    "Rules with neither input nor " "output files are always executed."
                )
            elif self.nooutput:
                s.append(
                    "Rules with a run or shell declaration but no output "
                    "are always executed."
                )
            else:
                if self._missing_output:
                    s.append(
                        "Missing output files: {}".format(
                            ", ".join(self.missing_output)
                        )
                    )
                if self._incomplete_output:
                    s.append(
                        "Incomplete output files: {}".format(
                            ", ".join(self.incomplete_output)
                        )
                    )
                if self._updated_input:
                    updated_input = self.updated_input - self.updated_input_run
                    s.append("Updated input files: {}".format(", ".join(updated_input)))
                if self._updated_input_run:
                    s.append(
                        "Input files updated by another job: {}".format(
                            ", ".join(self.updated_input_run)
                        )
                    )
                if self.pipe:
                    s.append(
                        "Output file is a pipe and has to be filled for consuming job."
                    )
                if self.service:
                    s.append(
                        "Job provides a service which has to be kept active until all consumers are finished."
                    )
        s = "; ".join(s)
        if self.finished:
            return "Finished (was: {s})".format(s=s)
        return s

    def __bool__(self):
        v = bool(
            self.updated_input
            or self.missing_output
            or self.forced
            or self.updated_input_run
            or self.noio
            or self.nooutput
            or self.pipe
            or self.service
        )
        return v and not self.finished