scheduling.py

"""
This file contains the database models
"""

import os
import logging
logger = logging.getLogger(__name__)

from datetime import datetime, timedelta

from django.db.models import ForeignKey, CharField, DateTimeField, BooleanField, IntegerField, BigIntegerField, \
    ManyToManyField, CASCADE, SET_NULL, PROTECT, QuerySet
from django.contrib.postgres.fields import ArrayField, JSONField
from django.contrib.auth.models import User
from .specification import AbstractChoice, BasicCommon, Template, NamedCommon, annotate_validate_add_defaults_to_doc_using_template
from enum import Enum
from django.db.models.expressions import RawSQL

from lofar.sas.tmss.tmss.exceptions import SubtaskSchedulingException
from lofar.messaging.messagebus import ToBus, DEFAULT_BROKER, DEFAULT_BUSNAME
from lofar.messaging.messages import EventMessage
from lofar.sas.tmss.client.tmssbuslistener import DEFAULT_TMSS_SUBTASK_NOTIFICATION_PREFIX
from lofar.common.util import single_line_with_single_spaces

#
# I/O
#

#
# Choices
#

class SubtaskState(AbstractChoice):
    """Defines the model and predefined list of possible SubtaskStatusChoice's for Subtask.
    The items in the Choices class below are automagically populated into the database via a data migration."""
    class Choices(Enum):
        DEFINING = "defining"
        DEFINED = "defined"
        SCHEDULING = "scheduling"
        SCHEDULED = "scheduled"
        UNSCHEDULING = "unscheduling"
        QUEUEING = "queueing"
        QUEUED = "queued"
        STARTING = "starting"
        STARTED = "started"
        FINISHING = "finishing"
        FINISHED = "finished"
        CANCELLING = "cancelling"
        CANCELLED = "cancelled"
        ERROR = "error"


class SubtaskType(AbstractChoice):
    """Defines the model and predefined list of possible SubtaskType's for Subtask.
    The items in the Choices class below are automagically populated into the database via a data migration."""
    class Choices(Enum):
        OBSERVATION = "observation"
        PIPELINE = "pipeline"
        COPY = "copy"
        INSPECTION = "inspection"
        QA_FILES = "qa_files" # task which creates "adder" QA h5 file(s) from a MeasurementSet of beamformed data
        QA_PLOTS = "qa_plots" # task which creates "adder" QA plots from an "adder" QA h5 file h5
        DELETION = "deletion"
        MANUAL = 'manual'
        OTHER = 'other'


class StationType(AbstractChoice):
        """Defines the model and predefined list of possible StationType's for AntennaSet.
        The items in the Choices class below are automagically populated into the database via a data migration."""

        class Choices(Enum):
            CORE = "core"
            REMOTE = "remote"
            INTERNATIONAL = "international"


class Algorithm(AbstractChoice):
    """Defines the model and predefined list of possible Algorithm's for DataproductHash.
    The items in the Choices class below are automagically populated into the database via a data migration."""

    class Choices(Enum):
        MD5 = 'md5'
        AES256 = 'aes256'


class ScheduleMethod(AbstractChoice):
    """Defines the model and predefined list of possible Algorithm's for DataproductHash.
    The items in the Choices class below are automagically populated into the database via a data migration."""

    class Choices(Enum):
        MANUAL = 'manual'
        BATCH = 'batch'
        DYNAMIC = 'dynamic'


#
# Templates
#

class SubtaskTemplate(Template):
    type = ForeignKey('SubtaskType', null=False, on_delete=PROTECT)
    queue = BooleanField(default=False)
    realtime = BooleanField(default=False)


class DefaultSubtaskTemplate(BasicCommon):
    name = CharField(max_length=128, unique=True)
    template = ForeignKey('SubtaskTemplate', on_delete=PROTECT)


class DataproductSpecificationsTemplate(Template):
    pass


class DefaultDataproductSpecificationsTemplate(BasicCommon):
    name = CharField(max_length=128, unique=True)
    template = ForeignKey('DataproductSpecificationsTemplate', on_delete=PROTECT)


class DataproductFeedbackTemplate(Template):
    pass

# todo: do we need to specify a default?

#
# Instance Objects
#
class Subtask(BasicCommon):
    """
    Represents a low-level task, which is an atomic unit of execution, such as running an observation, running
    inspection plots on the observed data, etc. Each task has a specific configuration, will have resources allocated
    to it, and represents a single run.
    """
    start_time = DateTimeField(null=True, help_text='Start this subtask at the specified time (NULLable).')
    stop_time = DateTimeField(null=True, help_text='Stop this subtask at the specified time (NULLable).')
    state = ForeignKey('SubtaskState', null=False, on_delete=PROTECT, related_name='task_states', help_text='Subtask state (see Subtask State Machine).')
    specifications_doc = JSONField(help_text='Final specifications, as input for the controller.')
    task_blueprint = ForeignKey('TaskBlueprint', related_name='subtasks', null=True, on_delete=SET_NULL, help_text='Task Blueprint to which this Subtask belongs.')
    specifications_template = ForeignKey('SubtaskTemplate', null=False, on_delete=PROTECT, help_text='Schema used for specifications_doc.')
    do_cancel = DateTimeField(null=True, help_text='Timestamp when the subtask has been ordered to cancel (NULLable).')
    priority = IntegerField(help_text='Absolute priority of this subtask (higher value means more important).')
    schedule_method = ForeignKey('ScheduleMethod', null=False, on_delete=PROTECT, help_text='Which method to use for scheduling this Subtask. One of (MANUAL, BATCH, DYNAMIC).')
    cluster = ForeignKey('Cluster', null=True, on_delete=PROTECT, help_text='Where the Subtask is scheduled to run (NULLable).')
    # resource_claim = ForeignKey("ResourceClaim", null=False, on_delete=PROTECT) # todo <-- how is this external reference supposed to work?
    created_or_updated_by_user = ForeignKey(User, null=True, editable=False, on_delete=PROTECT, help_text='The user who created / updated the subtask.')
    raw_feedback = CharField(null=True, max_length=1048576, help_text='The raw feedback for this Subtask')

    def __init__(self, *args, **kwargs):
        super().__init__(*args, **kwargs)

        # keep original state for logging
        self.__original_state_id = self.state_id

    @property
    def duration(self) -> timedelta:
        '''the duration of this subtask (stop-start), or 0 if start/stop are None'''
        if self.start_time is None or self.stop_time is None:
            return timedelta(seconds=0)
        return self.stop_time - self.start_time

    @property
    def specified_duration(self) -> timedelta:
        '''get the specified (or estimated) duration of this subtask based on the specified task duration and the subtask type'''
        if self.specifications_template.type.value == SubtaskType.Choices.OBSERVATION:
            # observations have a specified duration, so grab it from the spec.
            return timedelta(seconds=self.task_blueprint.specifications_doc.get('duration', 0))

        if self.specifications_template.type.value == SubtaskType.Choices.PIPELINE:
            # pipelines usually do not have a specified duration, so make a guess (half the obs duration?).
            return timedelta(seconds=self.task_blueprint.specifications_doc.get('duration', max(p.specified_duration for p in self.predecessors)/2))

        # other subtasktypes usually depend on cpu/data/network etc. So, make a guess (for now)
        return timedelta(minutes=5)

    @staticmethod
    def _send_state_change_event_message(subtask_id:int, old_state: str, new_state: str):
        with ToBus(exchange=os.environ.get("TMSS_EXCHANGE", DEFAULT_BUSNAME),
                   broker=os.environ.get("TMSS_BROKER", DEFAULT_BROKER)) as tobus: #TODO: do we want to connect to the bus for each new message, or have some global tobus?
            msg = EventMessage(subject="%s.%s" % (DEFAULT_TMSS_SUBTASK_NOTIFICATION_PREFIX, new_state.capitalize()),
                               content={'subtask_id': subtask_id, 'old_state': old_state, 'new_state': new_state})
            address = tobus.remote_address
            logger.info("Sending message with subject '%s' to exchange='%s' on broker=%s:%s content: %s",
                        msg.subject, tobus.exchange, address[0], address[1], single_line_with_single_spaces(msg.content))
            tobus.send(msg)

    @staticmethod
    def independent_subtasks() -> QuerySet:
        '''return a QuerySet of all subtasks with no input (i.e. which are "independent" because they have no predecessors)
           If you want the result, add .all() like so: Subtask.independent_subtasks().all()
        '''
        return Subtask.objects.filter(inputs=None)

    @property
    def successors(self) -> QuerySet:
        '''return the connect successor subtask(s) as queryset (over which you can perform extended queries, or return via the serializers/viewsets)
           If you want the result, add .all() like so: my_subtask.successors.all()
        '''
        # JS, 20200528: I couldn't make django do a "self-reference" query from the subtask table to the subtask table (via input, output), so I used plain SQL.
        return Subtask.objects.filter(id__in=RawSQL("SELECT successor_st.id FROM tmssapp_subtask as successor_st\n"
                                                    "INNER JOIN tmssapp_subtaskinput as st_input on st_input.subtask_id = successor_st.id\n"
                                                    "INNER JOIN tmssapp_subtaskoutput as st_output on st_output.id = st_input.producer_id\n"
                                                    "WHERE st_output.subtask_id = %s", params=[self.id]))

    @property
    def predecessors(self) -> QuerySet:
        '''return the connect predecessor subtask(s) as queryset (over which you can perform extended queries, or return via the serializers/viewsets)
        If you want the result, add .all() like so: my_subtask.predecessors.all()
        '''
        # JS, 20200528: I couldn't make django do a "self-reference" query from the subtask table to the subtask table (via input, output), so I used plain SQL.
        return Subtask.objects.filter(id__in=RawSQL("SELECT predecessor_st.id FROM tmssapp_subtask as predecessor_st\n"
                                                    "INNER JOIN tmssapp_subtaskoutput as st_output on st_output.subtask_id = predecessor_st.id\n"
                                                    "INNER JOIN tmssapp_subtaskinput as st_input on st_input.producer_id = st_output.id\n"
                                                    "WHERE st_input.subtask_id = %s", params=[self.id]))

    def save(self, force_insert=False, force_update=False, using=None, update_fields=None):
        creating = self._state.adding  # True on create, False on update

        annotate_validate_add_defaults_to_doc_using_template(self, 'specifications_doc', 'specifications_template')

        if self.state.value == SubtaskState.Choices.SCHEDULED.value and self.__original_state_id == SubtaskState.Choices.SCHEDULING.value:
            if self.start_time is None:
                if self.predecessors.all().count() == 0:
                    raise SubtaskSchedulingException("Cannot schedule subtask id=%s when start time is 'None'." % (self.pk, ))
                else:
                    self.start_time = datetime.utcnow()

        if self.state.value == SubtaskState.Choices.FINISHING.value:
            self.stop_time = datetime.utcnow()

        super().save(force_insert, force_update, using, update_fields)

        # log if either state update or new entry:
        if self.state_id != self.__original_state_id or creating == True:
            if self.created_or_updated_by_user is None:
                identifier = None
            else:
                identifier = self.created_or_updated_by_user.email
            log_entry = SubtaskStateLog(subtask=self, old_state=SubtaskState.objects.get(value=self.__original_state_id), new_state=self.state,
                                           user=self.created_or_updated_by_user, user_identifier=identifier)
            log_entry.save()

            try:
                self._send_state_change_event_message(self.id, log_entry.old_state.value, log_entry.new_state.value)
            except Exception as e:
                logger.error("Could not send state change to messagebus: %s", e)

            # update the previous state value
            self.__original_state_id = self.state_id

class SubtaskStateLog(BasicCommon):
    """
    History of state changes on subtasks
    This is now a very specific solution and based on what SOS communicated is what they are regularly interested in.
    Maybe one or two additional log tables for other models are benefitial and should be added at some point.

    Note that we could of course also log on the db level and there is also a variety of audit middlewares for Django
    available to keep track of changes more generally: https://djangopackages.org/grids/g/model-audit/
    This seems a bit overkill at the moment and we have to manage access to those logs etc., this needs tbd.
    """
    user = ForeignKey(User, null=True, editable=False, on_delete=PROTECT, help_text='The user who changed the state of the subtask.')
    user_identifier = CharField(null=True, editable=False, max_length=128, help_text='The ID of the user who changed the state of the subtask.')
    subtask = ForeignKey('Subtask', null=False,  editable=False, on_delete=CASCADE, help_text='Subtask to which this state change refers.')
    old_state = ForeignKey('SubtaskState', null=True,  editable=False, on_delete=PROTECT, related_name='is_old_state_of', help_text='Subtask state before update (see Subtask State Machine).')
    new_state = ForeignKey('SubtaskState', null=False, editable=False, on_delete=PROTECT, related_name='is_new_state_of', help_text='Subtask state after update (see Subtask State Machine).')


class SubtaskInput(BasicCommon):
    subtask = ForeignKey('Subtask', null=False, on_delete=CASCADE, related_name='inputs', help_text='Subtask to which this input specification refers.')
    task_relation_blueprint = ForeignKey('TaskRelationBlueprint', null=True, on_delete=SET_NULL, help_text='Task Relation Blueprint which this Subtask Input implements (NULLable).')
    producer = ForeignKey('SubtaskOutput', on_delete=PROTECT, related_name='consumers', help_text='The SubtaskOutput producing the input dataproducts for this SubtaskInput.')
    dataproducts = ManyToManyField('Dataproduct', help_text='The Dataproducts resulting from application of the filter at time of scheduling Although the dataproducts are simply the result of applying the filter on immutable data, the filter application could change over time. We thus store the result of this filtering directly to retain which input was specified for the task..')
    selection_doc = JSONField(help_text='Filter to apply to the dataproducts of the producer, to derive input dataproducts when scheduling.')
    selection_template = ForeignKey('TaskRelationSelectionTemplate', on_delete=PROTECT, help_text='Schema used for selection_doc.')

    def save(self, force_insert=False, force_update=False, using=None, update_fields=None):
        annotate_validate_add_defaults_to_doc_using_template(self, 'selection_doc', 'selection_template')
        super().save(force_insert, force_update, using, update_fields)


class SubtaskOutput(BasicCommon):
    subtask = ForeignKey('Subtask', null=False, on_delete=CASCADE, related_name='outputs', help_text='Subtask to which this output specification refers.')


class Dataproduct(BasicCommon):
    """
    A data product represents an atomic dataset, produced and consumed by subtasks. The consumed dataproducts are those
    resulting from interpreting the Subtask Connector filters of the inputs. These links are explicitly saved, should
    the interpretation of the filter change over time. The produced dataproducts enumerate everything produced by a
    Subtask.
    """
    filename = CharField(max_length=128, help_text='Name of the file (or top-level directory) of the dataproduct. Adheres to a naming convention, but is not meant for parsing.')
    directory = CharField(max_length=1024, help_text='Directory where this dataproduct is (to be) stored.')
    dataformat = ForeignKey('Dataformat', null=False, on_delete=PROTECT)
    datatype = ForeignKey('Datatype', null=False, on_delete=PROTECT)
    deleted_since = DateTimeField(null=True, help_text='When this dataproduct was removed from disk, or NULL if not deleted (NULLable).')
    pinned_since = DateTimeField(null=True, help_text='When this dataproduct was pinned to disk, that is, forbidden to be removed, or NULL if not pinned (NULLable).')
    specifications_doc = JSONField(help_text='Dataproduct properties (f.e. beam, subband), to distinguish them when produced by the same task, and to act as input for selections in the Task Input and Work Request Relation Blueprint objects.')
    specifications_template = ForeignKey('DataproductSpecificationsTemplate', null=False, on_delete=CASCADE, help_text='Schema used for specifications_doc.')
    producer = ForeignKey('SubtaskOutput', on_delete=PROTECT, related_name="dataproducts", help_text='Subtask Output which generates this dataproduct.')
    do_cancel = DateTimeField(null=True, help_text='When this dataproduct was cancelled (NULLable).  Cancelling a dataproduct triggers cleanup if necessary.')
    expected_size = BigIntegerField(null=True, help_text='Expected size of dataproduct size, in bytes. Used for scheduling purposes. NULL if size is unknown (NULLable).')
    size = BigIntegerField(null=True, help_text='Dataproduct size, in bytes. Used for accounting purposes. NULL if size is (yet) unknown (NULLable).')
    feedback_doc = JSONField(help_text='Dataproduct properties, as reported by the producing process.')
    feedback_template = ForeignKey('DataproductFeedbackTemplate', on_delete=PROTECT, help_text='Schema used for feedback_doc.')

    def save(self, force_insert=False, force_update=False, using=None, update_fields=None):
        annotate_validate_add_defaults_to_doc_using_template(self, 'specifications_doc', 'specifications_template')
        annotate_validate_add_defaults_to_doc_using_template(self, 'feedback_doc', 'feedback_template')

        super().save(force_insert, force_update, using, update_fields)

class AntennaSet(NamedCommon):
    station_type = ForeignKey('StationType', null=False, on_delete=PROTECT)
    rcus = ArrayField(IntegerField(), size=128, blank=False)
    inputs = ArrayField(CharField(max_length=128), size=128, blank=True)


class DataproductTransform(BasicCommon):
    """
    Each output dataproduct of a Subtask is linked to the input dataproducts that are used to produce it.
    These transforms encode the provenance information needed when tracking dependencies between dataproducts.
    """
    input = ForeignKey('Dataproduct', related_name='inputs', on_delete=PROTECT, help_text='A dataproduct that was the input of a transformation.')
    output = ForeignKey('Dataproduct',  related_name='outputs', on_delete=PROTECT, help_text='A dataproduct that was produced from the input dataproduct.')
    identity = BooleanField(help_text='TRUE if this transform only copies, tars, or losslessly compresses its input, FALSE if the transform changes the data. Allows for efficient reasoning about data duplication.')


class Filesystem(NamedCommon):
    capacity = BigIntegerField(help_text='Capacity in bytes')
    cluster = ForeignKey('Cluster', on_delete=PROTECT, help_text='Cluster hosting this filesystem.')
    directory = CharField(max_length=1024, help_text='Root directory under which we are allowed to write our data.')

    def save(self, force_insert=False, force_update=False, using=None, update_fields=None):
        if self.directory and not self.directory.endswith('/'):
            raise ValueError('directory value must end with a trailing slash!')  # todo: ...and needs to start with slash?

        super().save(force_insert, force_update, using, update_fields)


class Cluster(NamedCommon):
    location = CharField(max_length=128, help_text='Human-readable location of the cluster.')
    archive_site = BooleanField(default=False, null=False, help_text='TRUE if this cluster is an archive site, FALSE if not (f.e. a local cluster, or user-owned cluster).')


class DataproductArchiveInfo(BasicCommon):
    dataproduct = ForeignKey('Dataproduct', on_delete=PROTECT, help_text='A dataproduct residing in the archive.')
    storage_ticket = CharField(max_length=128, help_text='Archive-system identifier.')
    public_since = DateTimeField(null=True, help_text='Dataproduct is available for public download since this moment, or NULL if dataproduct is not (NULLable).')
    corrupted_since = DateTimeField(null=True, help_text='Earliest timestamp from which this dataproduct is known to be partially or fully corrupt, or NULL if dataproduct is not known to be corrupt (NULLable).')


class DataproductHash(BasicCommon):
    dataproduct = ForeignKey('Dataproduct', on_delete=PROTECT, help_text='The dataproduct to which this hash refers.')
    algorithm = ForeignKey('Algorithm', null=False, on_delete=PROTECT, help_text='Algorithm used (MD5, AES256).')
    hash = CharField(max_length=128, help_text='Hash value.')