Skip to content

prefect.server.services.scheduler

The Scheduler service.

RecentDeploymentsScheduler

Bases: Scheduler

A scheduler that only schedules deployments that were updated very recently. This scheduler can run on a tight loop and ensure that runs from newly-created or updated deployments are rapidly scheduled without having to wait for the "main" scheduler to complete its loop.

Note that scheduling is idempotent, so its ok for this scheduler to attempt to schedule the same deployments as the main scheduler. It's purpose is to accelerate scheduling for any deployments that users are interacting with.

Source code in prefect/server/services/scheduler.py
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
class RecentDeploymentsScheduler(Scheduler):
    """
    A scheduler that only schedules deployments that were updated very recently.
    This scheduler can run on a tight loop and ensure that runs from
    newly-created or updated deployments are rapidly scheduled without having to
    wait for the "main" scheduler to complete its loop.

    Note that scheduling is idempotent, so its ok for this scheduler to attempt
    to schedule the same deployments as the main scheduler. It's purpose is to
    accelerate scheduling for any deployments that users are interacting with.
    """

    # this scheduler runs on a tight loop
    loop_seconds = 5

    @inject_db
    def _get_select_deployments_to_schedule_query(self, db: PrefectDBInterface):
        """
        Returns a sqlalchemy query for selecting deployments to schedule
        """
        query = (
            sa.select(db.Deployment.id)
            .where(
                sa.and_(
                    db.Deployment.paused.is_not(True),
                    # use a slightly larger window than the loop interval to pick up
                    # any deployments that were created *while* the scheduler was
                    # last running (assuming the scheduler takes less than one
                    # second to run). Scheduling is idempotent so picking up schedules
                    # multiple times is not a concern.
                    db.Deployment.updated
                    >= pendulum.now("UTC").subtract(seconds=self.loop_seconds + 1),
                    (
                        # Only include deployments that have at least one
                        # active schedule.
                        sa.select(db.DeploymentSchedule.deployment_id)
                        .where(
                            sa.and_(
                                db.DeploymentSchedule.deployment_id == db.Deployment.id,
                                db.DeploymentSchedule.active.is_(True),
                            )
                        )
                        .exists()
                    ),
                )
            )
            .order_by(db.Deployment.id)
            .limit(self.deployment_batch_size)
        )
        return query

Scheduler

Bases: LoopService

A loop service that schedules flow runs from deployments.

Source code in prefect/server/services/scheduler.py
 33
 34
 35
 36
 37
 38
 39
 40
 41
 42
 43
 44
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
class Scheduler(LoopService):
    """
    A loop service that schedules flow runs from deployments.
    """

    # the main scheduler takes its loop interval from
    # PREFECT_API_SERVICES_SCHEDULER_LOOP_SECONDS
    loop_seconds = None

    def __init__(self, loop_seconds: float = None, **kwargs):
        super().__init__(
            loop_seconds=(
                loop_seconds
                or self.loop_seconds
                or PREFECT_API_SERVICES_SCHEDULER_LOOP_SECONDS.value()
            ),
            **kwargs,
        )
        self.deployment_batch_size: int = (
            PREFECT_API_SERVICES_SCHEDULER_DEPLOYMENT_BATCH_SIZE.value()
        )
        self.max_runs: int = PREFECT_API_SERVICES_SCHEDULER_MAX_RUNS.value()
        self.min_runs: int = PREFECT_API_SERVICES_SCHEDULER_MIN_RUNS.value()
        self.max_scheduled_time: datetime.timedelta = (
            PREFECT_API_SERVICES_SCHEDULER_MAX_SCHEDULED_TIME.value()
        )
        self.min_scheduled_time: datetime.timedelta = (
            PREFECT_API_SERVICES_SCHEDULER_MIN_SCHEDULED_TIME.value()
        )
        self.insert_batch_size = (
            PREFECT_API_SERVICES_SCHEDULER_INSERT_BATCH_SIZE.value()
        )

    @inject_db
    async def run_once(self, db: PrefectDBInterface):
        """
        Schedule flow runs by:

        - Querying for deployments with active schedules
        - Generating the next set of flow runs based on each deployments schedule
        - Inserting all scheduled flow runs into the database

        All inserted flow runs are committed to the database at the termination of the
        loop.
        """
        total_inserted_runs = 0

        last_id = None
        while True:
            async with db.session_context(begin_transaction=False) as session:
                query = self._get_select_deployments_to_schedule_query()

                # use cursor based pagination
                if last_id:
                    query = query.where(db.Deployment.id > last_id)

                result = await session.execute(query)
                deployment_ids = result.scalars().unique().all()

                # collect runs across all deployments
                try:
                    runs_to_insert = await self._collect_flow_runs(
                        session=session, deployment_ids=deployment_ids
                    )
                except TryAgain:
                    continue

            # bulk insert the runs based on batch size setting
            for batch in batched_iterable(runs_to_insert, self.insert_batch_size):
                async with db.session_context(begin_transaction=True) as session:
                    inserted_runs = await self._insert_scheduled_flow_runs(
                        session=session, runs=batch
                    )
                    total_inserted_runs += len(inserted_runs)

            # if this is the last page of deployments, exit the loop
            if len(deployment_ids) < self.deployment_batch_size:
                break
            else:
                # record the last deployment ID
                last_id = deployment_ids[-1]

        self.logger.info(f"Scheduled {total_inserted_runs} runs.")

    @inject_db
    def _get_select_deployments_to_schedule_query(self, db: PrefectDBInterface):
        """
        Returns a sqlalchemy query for selecting deployments to schedule.

        The query gets the IDs of any deployments with:

            - an active schedule
            - EITHER:
                - fewer than `min_runs` auto-scheduled runs
                - OR the max scheduled time is less than `max_scheduled_time` in the future
        """
        now = pendulum.now("UTC")
        query = (
            sa.select(db.Deployment.id)
            .select_from(db.Deployment)
            # TODO: on Postgres, this could be replaced with a lateral join that
            # sorts by `next_scheduled_start_time desc` and limits by
            # `self.min_runs` for a ~ 50% speedup. At the time of writing,
            # performance of this universal query appears to be fast enough that
            # this optimization is not worth maintaining db-specific queries
            .join(
                db.FlowRun,
                # join on matching deployments, only picking up future scheduled runs
                sa.and_(
                    db.Deployment.id == db.FlowRun.deployment_id,
                    db.FlowRun.state_type == StateType.SCHEDULED,
                    db.FlowRun.next_scheduled_start_time >= now,
                    db.FlowRun.auto_scheduled.is_(True),
                ),
                isouter=True,
            )
            .where(
                sa.and_(
                    db.Deployment.paused.is_not(True),
                    (
                        # Only include deployments that have at least one
                        # active schedule.
                        sa.select(db.DeploymentSchedule.deployment_id)
                        .where(
                            sa.and_(
                                db.DeploymentSchedule.deployment_id == db.Deployment.id,
                                db.DeploymentSchedule.active.is_(True),
                            )
                        )
                        .exists()
                    ),
                )
            )
            .group_by(db.Deployment.id)
            # having EITHER fewer than three runs OR runs not scheduled far enough out
            .having(
                sa.or_(
                    sa.func.count(db.FlowRun.next_scheduled_start_time) < self.min_runs,
                    sa.func.max(db.FlowRun.next_scheduled_start_time)
                    < now + self.min_scheduled_time,
                )
            )
            .order_by(db.Deployment.id)
            .limit(self.deployment_batch_size)
        )
        return query

    async def _collect_flow_runs(
        self,
        session: sa.orm.Session,
        deployment_ids: List[UUID],
    ) -> List[Dict]:
        runs_to_insert = []
        for deployment_id in deployment_ids:
            now = pendulum.now("UTC")
            # guard against erroneously configured schedules
            try:
                runs_to_insert.extend(
                    await self._generate_scheduled_flow_runs(
                        session=session,
                        deployment_id=deployment_id,
                        start_time=now,
                        end_time=now + self.max_scheduled_time,
                        min_time=self.min_scheduled_time,
                        min_runs=self.min_runs,
                        max_runs=self.max_runs,
                    )
                )
            except Exception:
                self.logger.exception(
                    f"Error scheduling deployment {deployment_id!r}.",
                )
            finally:
                connection = await session.connection()
                if connection.invalidated:
                    # If the error we handled above was the kind of database error that
                    # causes underlying transaction to rollback and the connection to
                    # become invalidated, rollback this session.  Errors that may cause
                    # this are connection drops, database restarts, and things of the
                    # sort.
                    #
                    # This rollback _does not rollback a transaction_, since that has
                    # actually already happened due to the error above.  It brings the
                    # Python session in sync with underlying connection so that when we
                    # exec the outer with block, the context manager will not attempt to
                    # commit the session.
                    #
                    # Then, raise TryAgain to break out of these nested loops, back to
                    # the outer loop, where we'll begin a new transaction with
                    # session.begin() in the next loop iteration.
                    await session.rollback()
                    raise TryAgain()
        return runs_to_insert

    @inject_db
    async def _generate_scheduled_flow_runs(
        self,
        session: sa.orm.Session,
        deployment_id: UUID,
        start_time: datetime.datetime,
        end_time: datetime.datetime,
        min_time: datetime.timedelta,
        min_runs: int,
        max_runs: int,
        db: PrefectDBInterface,
    ) -> List[Dict]:
        """
        Given a `deployment_id` and schedule params, generates a list of flow run
        objects and associated scheduled states that represent scheduled flow runs.

        Pass-through method for overrides.


        Args:
            session: a database session
            deployment_id: the id of the deployment to schedule
            start_time: the time from which to start scheduling runs
            end_time: runs will be scheduled until at most this time
            min_time: runs will be scheduled until at least this far in the future
            min_runs: a minimum amount of runs to schedule
            max_runs: a maximum amount of runs to schedule

        This function will generate the minimum number of runs that satisfy the min
        and max times, and the min and max counts. Specifically, the following order
        will be respected:

            - Runs will be generated starting on or after the `start_time`
            - No more than `max_runs` runs will be generated
            - No runs will be generated after `end_time` is reached
            - At least `min_runs` runs will be generated
            - Runs will be generated until at least `start_time + min_time` is reached

        """
        return await models.deployments._generate_scheduled_flow_runs(
            session=session,
            deployment_id=deployment_id,
            start_time=start_time,
            end_time=end_time,
            min_time=min_time,
            min_runs=min_runs,
            max_runs=max_runs,
        )

    @inject_db
    async def _insert_scheduled_flow_runs(
        self,
        session: sa.orm.Session,
        runs: List[Dict],
        db: PrefectDBInterface,
    ) -> List[UUID]:
        """
        Given a list of flow runs to schedule, as generated by
        `_generate_scheduled_flow_runs`, inserts them into the database. Note this is a
        separate method to facilitate batch operations on many scheduled runs.

        Pass-through method for overrides.
        """
        return await models.deployments._insert_scheduled_flow_runs(
            session=session, runs=runs
        )

run_once async

Schedule flow runs by:

  • Querying for deployments with active schedules
  • Generating the next set of flow runs based on each deployments schedule
  • Inserting all scheduled flow runs into the database

All inserted flow runs are committed to the database at the termination of the loop.

Source code in prefect/server/services/scheduler.py
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
@inject_db
async def run_once(self, db: PrefectDBInterface):
    """
    Schedule flow runs by:

    - Querying for deployments with active schedules
    - Generating the next set of flow runs based on each deployments schedule
    - Inserting all scheduled flow runs into the database

    All inserted flow runs are committed to the database at the termination of the
    loop.
    """
    total_inserted_runs = 0

    last_id = None
    while True:
        async with db.session_context(begin_transaction=False) as session:
            query = self._get_select_deployments_to_schedule_query()

            # use cursor based pagination
            if last_id:
                query = query.where(db.Deployment.id > last_id)

            result = await session.execute(query)
            deployment_ids = result.scalars().unique().all()

            # collect runs across all deployments
            try:
                runs_to_insert = await self._collect_flow_runs(
                    session=session, deployment_ids=deployment_ids
                )
            except TryAgain:
                continue

        # bulk insert the runs based on batch size setting
        for batch in batched_iterable(runs_to_insert, self.insert_batch_size):
            async with db.session_context(begin_transaction=True) as session:
                inserted_runs = await self._insert_scheduled_flow_runs(
                    session=session, runs=batch
                )
                total_inserted_runs += len(inserted_runs)

        # if this is the last page of deployments, exit the loop
        if len(deployment_ids) < self.deployment_batch_size:
            break
        else:
            # record the last deployment ID
            last_id = deployment_ids[-1]

    self.logger.info(f"Scheduled {total_inserted_runs} runs.")

TryAgain

Bases: Exception

Internal control-flow exception used to retry the Scheduler's main loop

Source code in prefect/server/services/scheduler.py
29
30
class TryAgain(Exception):
    """Internal control-flow exception used to retry the Scheduler's main loop"""