Skip to content

prefect.runner

Runners are responsible for managing the execution of deployments created and managed by either flow.serve or the serve utility.

Example
import time
from prefect import flow, serve


@flow
def slow_flow(sleep: int = 60):
    "Sleepy flow - sleeps the provided amount of time (in seconds)."
    time.sleep(sleep)


@flow
def fast_flow():
    "Fastest flow this side of the Mississippi."
    return


if __name__ == "__main__":
    slow_deploy = slow_flow.to_deployment(name="sleeper", interval=45)
    fast_deploy = fast_flow.to_deployment(name="fast")

    # serve generates a Runner instance
    serve(slow_deploy, fast_deploy)

Runner

Source code in /home/runner/work/docs/docs/prefect_source/src/prefect/runner.py
 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
293
294
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
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
class Runner:
    def __init__(
        self,
        name: Optional[str] = None,
        query_seconds: float = 10,
        prefetch_seconds: float = 10,
        limit: Optional[int] = None,
        pause_on_shutdown: bool = True,
    ):
        """
        Responsible for managing the execution of remotely initiated flow runs.

        Args:
            name: The name of the runner. If not provided, a random one
                will be generated. If provided, it cannot contain '/' or '%'.
            query_seconds: The number of seconds to wait between querying for
                scheduled flow runs.
            prefetch_seconds: The number of seconds to prefetch flow runs for.
            limit: The maximum number of flow runs this runner should be running at
            pause_on_shutdown: A boolean for whether or not to automatically pause
                deployment schedules on shutdown; defaults to `True`

        Examples:
            Set up a Runner to manage the execute of scheduled flow runs for two flows:
                ```python
                from prefect import flow, Runner

                @flow
                def hello_flow(name):
                    print(f"hello {name}")

                @flow
                def goodbye_flow(name):
                    print(f"goodbye {name}")

                if __name__ == "__main__"
                    runner = Runner(name="my-runner")

                    # Will be runnable via the API
                    runner.add_flow(hello_flow)

                    # Run on a cron schedule
                    runner.add_flow(goodbye_flow, schedule={"cron": "0 * * * *"})

                    runner.start()
                ```
        """
        if name and ("/" in name or "%" in name):
            raise ValueError("Runner name cannot contain '/' or '%'")
        self.name = Path(name).stem if name is not None else f"runner-{uuid4()}"
        self._logger = get_logger("runner")

        self.started = False
        self.pause_on_shutdown = pause_on_shutdown
        self.limit = limit or PREFECT_RUNNER_PROCESS_LIMIT.value()
        self._query_seconds = query_seconds
        self._prefetch_seconds = prefetch_seconds

        self._runs_task_group: anyio.abc.TaskGroup = anyio.create_task_group()
        self._loops_task_group: anyio.abc.TaskGroup = anyio.create_task_group()

        self._limiter: Optional[anyio.CapacityLimiter] = anyio.CapacityLimiter(
            self.limit
        )
        self._client = get_client()
        self._submitting_flow_run_ids = set()
        self._cancelling_flow_run_ids = set()
        self._scheduled_task_scopes = set()
        self._deployment_ids: Set[UUID] = set()
        self._flow_run_process_map = dict()

    @sync_compatible
    async def add_deployment(
        self,
        deployment: RunnerDeployment,
    ) -> UUID:
        """
        Registers the deployment with the Prefect API and will monitor for work once
        the runner is started.

        Args:
            deployment: A deployment for the runner to register.
        """
        deployment_id = await deployment.apply()
        self._deployment_ids.add(deployment_id)

        return deployment_id

    @sync_compatible
    async def add_flow(
        self,
        flow: Flow,
        name: str = None,
        interval: Optional[Union[int, float, datetime.timedelta]] = None,
        cron: Optional[str] = None,
        rrule: Optional[str] = None,
        schedule: Optional[SCHEDULE_TYPES] = None,
        parameters: Optional[dict] = None,
        triggers: Optional[List[DeploymentTrigger]] = None,
        description: Optional[str] = None,
        tags: Optional[List[str]] = None,
        version: Optional[str] = None,
        enforce_parameter_schema: bool = False,
    ) -> UUID:
        """
        Provides a flow to the runner to be run based on the provided configuration.

        Will create a deployment for the provided flow and register the deployment
        with the runner.

        Args:
            flow: A flow for the runner to run.
            name: The name to give the created deployment. Will default to the name
                of the runner.
            interval: An interval on which to execute the current flow. Accepts either a number
                or a timedelta object. If a number is given, it will be interpreted as seconds.
            cron: A cron schedule of when to execute runs of this flow.
            rrule: An rrule schedule of when to execute runs of this flow.
            schedule: A schedule object of when to execute runs of this flow. Used for
                advanced scheduling options like timezone.
            triggers: A list of triggers that should kick of a run of this flow.
            parameters: A dictionary of default parameter values to pass to runs of this flow.
            description: A description for the created deployment. Defaults to the flow's
                description if not provided.
            tags: A list of tags to associate with the created deployment for organizational
                purposes.
            version: A version for the created deployment. Defaults to the flow's version.
        """
        # TODO: expose a filesystem interface with hot reloading
        # will need to create a separate method for deployment creation
        api = PREFECT_API_URL.value()
        if any([interval, cron, rrule]) and not api:
            self._logger.warning(
                "Cannot schedule flows on an ephemeral server; run `prefect server"
                " start` to start the scheduler."
            )
        name = self.name if name is None else name

        deployment = flow.to_deployment(
            name=name,
            interval=interval,
            cron=cron,
            rrule=rrule,
            schedule=schedule,
            triggers=triggers,
            parameters=parameters,
            description=description,
            tags=tags,
            version=version,
            enforce_parameter_schema=enforce_parameter_schema,
        )
        return await self.add_deployment(deployment)

    @sync_compatible
    async def start(self, run_once: bool = False):
        """
        Starts a runner.

        The runner will begin monitoring for and executing any scheduled work for all added flows.

        Args:
            run_once: If True, the runner will through one query loop and then exit.

        Examples:
            Initialize a Runner, add two flows, and serve them by starting the Runner:

            ```python
            from prefect import flow, Runner

            @flow
            def hello_flow(name):
                print(f"hello {name}")

            @flow
            def goodbye_flow(name):
                print(f"goodbye {name}")

            if __name__ == "__main__"
                runner = Runner(name="my-runner")

                # Will be runnable via the API
                runner.add_flow(hello_flow)

                # Run on a cron schedule
                runner.add_flow(goodbye_flow, schedule={"cron": "0 * * * *"})

                runner.start()
            ```
        """
        async with self as runner:
            async with self._loops_task_group as tg:
                tg.start_soon(
                    partial(
                        critical_service_loop,
                        workload=runner._get_and_submit_flow_runs,
                        interval=self._query_seconds,
                        run_once=run_once,
                        jitter_range=0.3,
                    )
                )
                tg.start_soon(
                    partial(
                        critical_service_loop,
                        workload=runner._check_for_cancelled_flow_runs,
                        interval=self._query_seconds * 2,
                        run_once=run_once,
                        jitter_range=0.3,
                    )
                )

    def stop(self):
        """Stops the runner's polling cycle."""
        if not self.started:
            raise RuntimeError(
                "Runner has not yet started. Please start the runner by calling"
                " .start()"
            )
        self._loops_task_group.cancel_scope.cancel()

    async def execute_flow_run(self, flow_run_id: UUID):
        """
        Executes a single flow run with the given ID.

        Execution will wait to monitor for cancellation requests. Exits once
        the flow run process has exited.
        """
        async with self:
            if not self._acquire_limit_slot(flow_run_id):
                return

            async with anyio.create_task_group() as tg:
                with anyio.CancelScope():
                    self._submitting_flow_run_ids.add(flow_run_id)
                    flow_run = await self._client.read_flow_run(flow_run_id)

                    pid = await self._runs_task_group.start(
                        self._submit_run_and_capture_errors, flow_run
                    )

                    self._flow_run_process_map[flow_run.id] = pid

                    workload = partial(
                        self._check_for_cancelled_flow_runs,
                        on_nothing_to_watch=tg.cancel_scope.cancel,
                    )

                    tg.start_soon(
                        partial(
                            critical_service_loop,
                            workload=workload,
                            interval=self._query_seconds,
                            jitter_range=0.3,
                        )
                    )

    def _get_flow_run_logger(self, flow_run: "FlowRun") -> PrefectLogAdapter:
        return flow_run_logger(flow_run=flow_run).getChild(
            "runner",
            extra={
                "runner_name": self.name,
            },
        )

    async def _run_process(
        self,
        flow_run: "FlowRun",
        task_status: Optional[anyio.abc.TaskStatus] = None,
    ):
        """
        Runs the given flow run in a subprocess.

        Args:
            flow_run: Flow run to execute via process. The ID of this flow run
                is stored in the PREFECT__FLOW_RUN_ID environment variable to
                allow the engine to retrieve the corresponding flow's code and
                begin execution.
            task_status: anyio task status used to send a message to the caller
                than the flow run process has started.
        """
        command = f"{sys.executable} -m prefect.engine"

        flow_run_logger = self._get_flow_run_logger(flow_run)

        # We must add creationflags to a dict so it is only passed as a function
        # parameter on Windows, because the presence of creationflags causes
        # errors on Unix even if set to None
        kwargs: Dict[str, object] = {}
        if sys.platform == "win32":
            kwargs["creationflags"] = subprocess.CREATE_NEW_PROCESS_GROUP

        _use_threaded_child_watcher()
        flow_run_logger.info("Opening process...")

        env = get_current_settings().to_environment_variables(exclude_unset=True)
        env.update({"PREFECT__FLOW_RUN_ID": flow_run.id.hex})
        env.update(**os.environ)  # is this really necessary??

        process = await run_process(
            command.split(" "),
            stream_output=True,
            task_status=task_status,
            env=env,
            **kwargs,
        )

        # Use the pid for display if no name was given
        display_name = f" {process.pid}"

        if process.returncode:
            help_message = None
            if process.returncode == -9:
                help_message = (
                    "This indicates that the process exited due to a SIGKILL signal. "
                    "Typically, this is either caused by manual cancellation or "
                    "high memory usage causing the operating system to "
                    "terminate the process."
                )
            if process.returncode == -15:
                help_message = (
                    "This indicates that the process exited due to a SIGTERM signal. "
                    "Typically, this is caused by manual cancellation."
                )
            elif process.returncode == 247:
                help_message = (
                    "This indicates that the process was terminated due to high "
                    "memory usage."
                )
            elif (
                sys.platform == "win32" and process.returncode == STATUS_CONTROL_C_EXIT
            ):
                help_message = (
                    "Process was terminated due to a Ctrl+C or Ctrl+Break signal. "
                    "Typically, this is caused by manual cancellation."
                )

            flow_run_logger.error(
                f"Process{display_name} exited with status code: {process.returncode}"
                + (f"; {help_message}" if help_message else "")
            )
        else:
            flow_run_logger.info(f"Process{display_name} exited cleanly.")

        return process.returncode

    async def _kill_process(
        self,
        pid: int,
        grace_seconds: int = 30,
    ):
        """
        Kills a given flow run process.

        Args:
            pid: ID of the process to kill
            grace_seconds: Number of seconds to wait for the process to end.
        """
        # In a non-windows environment first send a SIGTERM, then, after
        # `grace_seconds` seconds have passed subsequent send SIGKILL. In
        # Windows we use CTRL_BREAK_EVENT as SIGTERM is useless:
        # https://bugs.python.org/issue26350
        if sys.platform == "win32":
            try:
                os.kill(pid, signal.CTRL_BREAK_EVENT)
            except (ProcessLookupError, WindowsError):
                raise RuntimeError(
                    f"Unable to kill process {pid!r}: The process was not found."
                )
        else:
            try:
                os.kill(pid, signal.SIGTERM)
            except ProcessLookupError:
                raise RuntimeError(
                    f"Unable to kill process {pid!r}: The process was not found."
                )

            # Throttle how often we check if the process is still alive to keep
            # from making too many system calls in a short period of time.
            check_interval = max(grace_seconds / 10, 1)

            with anyio.move_on_after(grace_seconds):
                while True:
                    await anyio.sleep(check_interval)

                    # Detect if the process is still alive. If not do an early
                    # return as the process respected the SIGTERM from above.
                    try:
                        os.kill(pid, 0)
                    except ProcessLookupError:
                        return

            try:
                os.kill(pid, signal.SIGKILL)
            except OSError:
                # We shouldn't ever end up here, but it's possible that the
                # process ended right after the check above.
                return

    async def _pause_schedules(self):
        """
        Pauses all deployment schedules.
        """
        self._logger.info("Pausing schedules for all deployments...")
        for deployment_id in self._deployment_ids:
            self._logger.debug(f"Pausing schedule for deployment '{deployment_id}'")
            await self._client.update_schedule(deployment_id, active=False)
        self._logger.info("All deployment schedules have been paused!")

    async def _get_and_submit_flow_runs(self):
        runs_response = await self._get_scheduled_flow_runs()

        return await self._submit_scheduled_flow_runs(flow_run_response=runs_response)

    async def _check_for_cancelled_flow_runs(
        self, on_nothing_to_watch: Callable = lambda: None
    ):
        if not self.started:
            raise RuntimeError(
                "Runner is not set up. Please make sure you are running this runner "
                "as an async context manager."
            )

        # To stop loop service checking for cancelled runs.
        # Need to find a better way to stop runner spawned by
        # a worker.
        if not self._flow_run_process_map and not self._deployment_ids:
            self._logger.debug(
                "Runner has no active flow runs or deployments. Sending message to loop"
                " service that no further cancellation checks are needed."
            )
            on_nothing_to_watch()

        self._logger.debug("Checking for cancelled flow runs...")

        named_cancelling_flow_runs = await self._client.read_flow_runs(
            flow_run_filter=FlowRunFilter(
                state=FlowRunFilterState(
                    type=FlowRunFilterStateType(any_=[StateType.CANCELLED]),
                    name=FlowRunFilterStateName(any_=["Cancelling"]),
                ),
                # Avoid duplicate cancellation calls
                id=FlowRunFilterId(
                    any_=list(
                        self._flow_run_process_map.keys()
                        - self._cancelling_flow_run_ids
                    )
                ),
            ),
        )

        typed_cancelling_flow_runs = await self._client.read_flow_runs(
            flow_run_filter=FlowRunFilter(
                state=FlowRunFilterState(
                    type=FlowRunFilterStateType(any_=[StateType.CANCELLING]),
                ),
                # Avoid duplicate cancellation calls
                id=FlowRunFilterId(
                    any_=list(
                        self._flow_run_process_map.keys()
                        - self._cancelling_flow_run_ids
                    )
                ),
            ),
        )

        cancelling_flow_runs = named_cancelling_flow_runs + typed_cancelling_flow_runs

        if cancelling_flow_runs:
            self._logger.info(
                f"Found {len(cancelling_flow_runs)} flow runs awaiting cancellation."
            )

        for flow_run in cancelling_flow_runs:
            self._cancelling_flow_run_ids.add(flow_run.id)
            self._runs_task_group.start_soon(self._cancel_run, flow_run)

        return cancelling_flow_runs

    async def _cancel_run(self, flow_run: "FlowRun"):
        run_logger = self._get_flow_run_logger(flow_run)

        pid = self._flow_run_process_map.get(flow_run.id)
        if not pid:
            await self._mark_flow_run_as_cancelled(
                flow_run,
                state_updates={
                    "message": (
                        "Could not find process ID for flow run"
                        " and cancellation cannot be guaranteed."
                    )
                },
            )
            return

        try:
            await self._kill_process(pid)
        except RuntimeError as exc:
            self._logger.warning(f"{exc} Marking flow run as cancelled.")
            await self._mark_flow_run_as_cancelled(flow_run)
        except Exception:
            run_logger.exception(
                "Encountered exception while killing process for flow run "
                f"'{flow_run.id}'. Flow run may not be cancelled."
            )
            # We will try again on generic exceptions
            self._cancelling_flow_run_ids.remove(flow_run.id)
            return
        else:
            await self._mark_flow_run_as_cancelled(flow_run)
            run_logger.info(f"Cancelled flow run '{flow_run.id}'!")

    async def _get_scheduled_flow_runs(
        self,
    ) -> List["FlowRun"]:
        """
        Retrieve scheduled flow runs for this runner.
        """
        scheduled_before = pendulum.now("utc").add(seconds=int(self._prefetch_seconds))
        self._logger.debug(
            f"Querying for flow runs scheduled before {scheduled_before}"
        )

        scheduled_flow_runs = (
            await self._client.get_scheduled_flow_runs_for_deployments(
                deployment_ids=list(self._deployment_ids),
                scheduled_before=scheduled_before,
            )
        )
        self._logger.debug(f"Discovered {len(scheduled_flow_runs)} scheduled_flow_runs")
        return scheduled_flow_runs

    def _acquire_limit_slot(self, flow_run_id: str) -> bool:
        """
        Enforces flow run limit set on runner.

        Returns:
            - bool: True if a slot was acquired, False otherwise.
        """
        try:
            if self._limiter:
                self._limiter.acquire_on_behalf_of_nowait(flow_run_id)
            return True
        except anyio.WouldBlock:
            self._logger.info(
                f"Flow run limit reached; {self._limiter.borrowed_tokens} flow runs"
                " in progress. You can control this limit by adjusting the"
                " PREFECT_RUNNER_PROCESS_LIMIT setting."
            )
            return False

    def _release_limit_slot(self, flow_run_id: str) -> None:
        """
        Frees up a slot taken by the given flow run id.
        """
        if self._limiter:
            self._limiter.release_on_behalf_of(flow_run_id)

    async def _submit_scheduled_flow_runs(
        self, flow_run_response: List["FlowRun"]
    ) -> List["FlowRun"]:
        """
        Takes a list of FlowRuns and submits the referenced flow runs
        for execution by the runner.
        """
        submittable_flow_runs = flow_run_response
        submittable_flow_runs.sort(key=lambda run: run.next_scheduled_start_time)
        for flow_run in submittable_flow_runs:
            if flow_run.id in self._submitting_flow_run_ids:
                continue

            if self._acquire_limit_slot(flow_run.id):
                run_logger = self._get_flow_run_logger(flow_run)
                run_logger.info(
                    f"Runner '{self.name}' submitting flow run '{flow_run.id}'"
                )
                self._submitting_flow_run_ids.add(flow_run.id)
                self._runs_task_group.start_soon(
                    self._submit_run,
                    flow_run,
                )
            else:
                break

        return list(
            filter(
                lambda run: run.id in self._submitting_flow_run_ids,
                submittable_flow_runs,
            )
        )

    async def _submit_run(self, flow_run: "FlowRun") -> None:
        """
        Submits a given flow run for execution by the runner.
        """
        run_logger = self._get_flow_run_logger(flow_run)

        ready_to_submit = await self._propose_pending_state(flow_run)

        if ready_to_submit:
            readiness_result = await self._runs_task_group.start(
                self._submit_run_and_capture_errors, flow_run
            )

            if readiness_result and not isinstance(readiness_result, Exception):
                self._flow_run_process_map[flow_run.id] = readiness_result

            run_logger.info(f"Completed submission of flow run '{flow_run.id}'")
        else:
            # If the run is not ready to submit, release the concurrency slot
            self._release_limit_slot(flow_run.id)

        self._submitting_flow_run_ids.remove(flow_run.id)

    async def _submit_run_and_capture_errors(
        self, flow_run: "FlowRun", task_status: anyio.abc.TaskStatus = None
    ) -> Union[Optional[int], Exception]:
        run_logger = self._get_flow_run_logger(flow_run)

        try:
            status_code = await self._run_process(
                flow_run=flow_run,
                task_status=task_status,
            )
        except Exception as exc:
            if not task_status._future.done():
                # This flow run was being submitted and did not start successfully
                run_logger.exception(
                    f"Failed to start process for flow run '{flow_run.id}'."
                )
                # Mark the task as started to prevent agent crash
                task_status.started(exc)
                await self._propose_crashed_state(
                    flow_run, "Flow run process could not be started"
                )
            else:
                run_logger.exception(
                    f"An error occurred while monitoring flow run '{flow_run.id}'. "
                    "The flow run will not be marked as failed, but an issue may have "
                    "occurred."
                )
            return exc
        finally:
            self._release_limit_slot(flow_run.id)
            self._flow_run_process_map.pop(flow_run.id, None)

        if status_code != 0:
            await self._propose_crashed_state(
                flow_run,
                f"Flow run process exited with non-zero status code {status_code}.",
            )

        return status_code

    async def _propose_pending_state(self, flow_run: "FlowRun") -> bool:
        run_logger = self._get_flow_run_logger(flow_run)
        state = flow_run.state
        try:
            state = await propose_state(
                self._client, Pending(), flow_run_id=flow_run.id
            )
        except Abort as exc:
            run_logger.info(
                (
                    f"Aborted submission of flow run '{flow_run.id}'. "
                    f"Server sent an abort signal: {exc}"
                ),
            )
            return False
        except Exception:
            run_logger.exception(
                f"Failed to update state of flow run '{flow_run.id}'",
            )
            return False

        if not state.is_pending():
            run_logger.info(
                (
                    f"Aborted submission of flow run '{flow_run.id}': "
                    f"Server returned a non-pending state {state.type.value!r}"
                ),
            )
            return False

        return True

    async def _propose_failed_state(self, flow_run: "FlowRun", exc: Exception) -> None:
        run_logger = self._get_flow_run_logger(flow_run)
        try:
            await propose_state(
                self._client,
                await exception_to_failed_state(message="Submission failed.", exc=exc),
                flow_run_id=flow_run.id,
            )
        except Abort:
            # We've already failed, no need to note the abort but we don't want it to
            # raise in the agent process
            pass
        except Exception:
            run_logger.error(
                f"Failed to update state of flow run '{flow_run.id}'",
                exc_info=True,
            )

    async def _propose_crashed_state(self, flow_run: "FlowRun", message: str) -> None:
        run_logger = self._get_flow_run_logger(flow_run)
        try:
            state = await propose_state(
                self._client,
                Crashed(message=message),
                flow_run_id=flow_run.id,
            )
        except Abort:
            # Flow run already marked as failed
            pass
        except Exception:
            run_logger.exception(f"Failed to update state of flow run '{flow_run.id}'")
        else:
            if state.is_crashed():
                run_logger.info(
                    f"Reported flow run '{flow_run.id}' as crashed: {message}"
                )

    async def _mark_flow_run_as_cancelled(
        self, flow_run: "FlowRun", state_updates: Optional[dict] = None
    ) -> None:
        state_updates = state_updates or {}
        state_updates.setdefault("name", "Cancelled")
        state_updates.setdefault("type", StateType.CANCELLED)
        state = flow_run.state.copy(update=state_updates)

        await self._client.set_flow_run_state(flow_run.id, state, force=True)

        # Do not remove the flow run from the cancelling set immediately because
        # the API caches responses for the `read_flow_runs` and we do not want to
        # duplicate cancellations.
        await self._schedule_task(
            60 * 10, self._cancelling_flow_run_ids.remove, flow_run.id
        )

    async def _schedule_task(self, __in_seconds: int, fn, *args, **kwargs):
        """
        Schedule a background task to start after some time.

        These tasks will be run immediately when the runner exits instead of waiting.

        The function may be async or sync. Async functions will be awaited.
        """

        async def wrapper(task_status):
            # If we are shutting down, do not sleep; otherwise sleep until the scheduled
            # time or shutdown
            if self.started:
                with anyio.CancelScope() as scope:
                    self._scheduled_task_scopes.add(scope)
                    task_status.started()
                    await anyio.sleep(__in_seconds)

                self._scheduled_task_scopes.remove(scope)
            else:
                task_status.started()

            result = fn(*args, **kwargs)
            if inspect.iscoroutine(result):
                await result

        await self._runs_task_group.start(wrapper)

    async def __aenter__(self):
        self._logger.debug("Starting runner...")
        self._client = get_client()
        await self._client.__aenter__()
        await self._runs_task_group.__aenter__()

        self.started = True
        return self

    async def __aexit__(self, *exc_info):
        self._logger.debug("Stopping runner...")
        if self.pause_on_shutdown:
            await self._pause_schedules()
        self.started = False
        for scope in self._scheduled_task_scopes:
            scope.cancel()
        if self._runs_task_group:
            await self._runs_task_group.__aexit__(*exc_info)
        if self._client:
            await self._client.__aexit__(*exc_info)

    def __repr__(self):
        return f"Runner(name={self.name!r})"

add_deployment async

Registers the deployment with the Prefect API and will monitor for work once the runner is started.

Parameters:

Name Type Description Default
deployment RunnerDeployment

A deployment for the runner to register.

required
Source code in /home/runner/work/docs/docs/prefect_source/src/prefect/runner.py
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
@sync_compatible
async def add_deployment(
    self,
    deployment: RunnerDeployment,
) -> UUID:
    """
    Registers the deployment with the Prefect API and will monitor for work once
    the runner is started.

    Args:
        deployment: A deployment for the runner to register.
    """
    deployment_id = await deployment.apply()
    self._deployment_ids.add(deployment_id)

    return deployment_id

add_flow async

Provides a flow to the runner to be run based on the provided configuration.

Will create a deployment for the provided flow and register the deployment with the runner.

Parameters:

Name Type Description Default
flow Flow

A flow for the runner to run.

required
name str

The name to give the created deployment. Will default to the name of the runner.

None
interval Optional[Union[int, float, timedelta]]

An interval on which to execute the current flow. Accepts either a number or a timedelta object. If a number is given, it will be interpreted as seconds.

None
cron Optional[str]

A cron schedule of when to execute runs of this flow.

None
rrule Optional[str]

An rrule schedule of when to execute runs of this flow.

None
schedule Optional[SCHEDULE_TYPES]

A schedule object of when to execute runs of this flow. Used for advanced scheduling options like timezone.

None
triggers Optional[List[DeploymentTrigger]]

A list of triggers that should kick of a run of this flow.

None
parameters Optional[dict]

A dictionary of default parameter values to pass to runs of this flow.

None
description Optional[str]

A description for the created deployment. Defaults to the flow's description if not provided.

None
tags Optional[List[str]]

A list of tags to associate with the created deployment for organizational purposes.

None
version Optional[str]

A version for the created deployment. Defaults to the flow's version.

None
Source code in /home/runner/work/docs/docs/prefect_source/src/prefect/runner.py
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
@sync_compatible
async def add_flow(
    self,
    flow: Flow,
    name: str = None,
    interval: Optional[Union[int, float, datetime.timedelta]] = None,
    cron: Optional[str] = None,
    rrule: Optional[str] = None,
    schedule: Optional[SCHEDULE_TYPES] = None,
    parameters: Optional[dict] = None,
    triggers: Optional[List[DeploymentTrigger]] = None,
    description: Optional[str] = None,
    tags: Optional[List[str]] = None,
    version: Optional[str] = None,
    enforce_parameter_schema: bool = False,
) -> UUID:
    """
    Provides a flow to the runner to be run based on the provided configuration.

    Will create a deployment for the provided flow and register the deployment
    with the runner.

    Args:
        flow: A flow for the runner to run.
        name: The name to give the created deployment. Will default to the name
            of the runner.
        interval: An interval on which to execute the current flow. Accepts either a number
            or a timedelta object. If a number is given, it will be interpreted as seconds.
        cron: A cron schedule of when to execute runs of this flow.
        rrule: An rrule schedule of when to execute runs of this flow.
        schedule: A schedule object of when to execute runs of this flow. Used for
            advanced scheduling options like timezone.
        triggers: A list of triggers that should kick of a run of this flow.
        parameters: A dictionary of default parameter values to pass to runs of this flow.
        description: A description for the created deployment. Defaults to the flow's
            description if not provided.
        tags: A list of tags to associate with the created deployment for organizational
            purposes.
        version: A version for the created deployment. Defaults to the flow's version.
    """
    # TODO: expose a filesystem interface with hot reloading
    # will need to create a separate method for deployment creation
    api = PREFECT_API_URL.value()
    if any([interval, cron, rrule]) and not api:
        self._logger.warning(
            "Cannot schedule flows on an ephemeral server; run `prefect server"
            " start` to start the scheduler."
        )
    name = self.name if name is None else name

    deployment = flow.to_deployment(
        name=name,
        interval=interval,
        cron=cron,
        rrule=rrule,
        schedule=schedule,
        triggers=triggers,
        parameters=parameters,
        description=description,
        tags=tags,
        version=version,
        enforce_parameter_schema=enforce_parameter_schema,
    )
    return await self.add_deployment(deployment)

execute_flow_run async

Executes a single flow run with the given ID.

Execution will wait to monitor for cancellation requests. Exits once the flow run process has exited.

Source code in /home/runner/work/docs/docs/prefect_source/src/prefect/runner.py
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
345
async def execute_flow_run(self, flow_run_id: UUID):
    """
    Executes a single flow run with the given ID.

    Execution will wait to monitor for cancellation requests. Exits once
    the flow run process has exited.
    """
    async with self:
        if not self._acquire_limit_slot(flow_run_id):
            return

        async with anyio.create_task_group() as tg:
            with anyio.CancelScope():
                self._submitting_flow_run_ids.add(flow_run_id)
                flow_run = await self._client.read_flow_run(flow_run_id)

                pid = await self._runs_task_group.start(
                    self._submit_run_and_capture_errors, flow_run
                )

                self._flow_run_process_map[flow_run.id] = pid

                workload = partial(
                    self._check_for_cancelled_flow_runs,
                    on_nothing_to_watch=tg.cancel_scope.cancel,
                )

                tg.start_soon(
                    partial(
                        critical_service_loop,
                        workload=workload,
                        interval=self._query_seconds,
                        jitter_range=0.3,
                    )
                )

start async

Starts a runner.

The runner will begin monitoring for and executing any scheduled work for all added flows.

Parameters:

Name Type Description Default
run_once bool

If True, the runner will through one query loop and then exit.

False

Examples:

Initialize a Runner, add two flows, and serve them by starting the Runner:

from prefect import flow, Runner

@flow
def hello_flow(name):
    print(f"hello {name}")

@flow
def goodbye_flow(name):
    print(f"goodbye {name}")

if __name__ == "__main__"
    runner = Runner(name="my-runner")

    # Will be runnable via the API
    runner.add_flow(hello_flow)

    # Run on a cron schedule
    runner.add_flow(goodbye_flow, schedule={"cron": "0 * * * *"})

    runner.start()
Source code in /home/runner/work/docs/docs/prefect_source/src/prefect/runner.py
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
293
294
295
296
297
298
299
300
@sync_compatible
async def start(self, run_once: bool = False):
    """
    Starts a runner.

    The runner will begin monitoring for and executing any scheduled work for all added flows.

    Args:
        run_once: If True, the runner will through one query loop and then exit.

    Examples:
        Initialize a Runner, add two flows, and serve them by starting the Runner:

        ```python
        from prefect import flow, Runner

        @flow
        def hello_flow(name):
            print(f"hello {name}")

        @flow
        def goodbye_flow(name):
            print(f"goodbye {name}")

        if __name__ == "__main__"
            runner = Runner(name="my-runner")

            # Will be runnable via the API
            runner.add_flow(hello_flow)

            # Run on a cron schedule
            runner.add_flow(goodbye_flow, schedule={"cron": "0 * * * *"})

            runner.start()
        ```
    """
    async with self as runner:
        async with self._loops_task_group as tg:
            tg.start_soon(
                partial(
                    critical_service_loop,
                    workload=runner._get_and_submit_flow_runs,
                    interval=self._query_seconds,
                    run_once=run_once,
                    jitter_range=0.3,
                )
            )
            tg.start_soon(
                partial(
                    critical_service_loop,
                    workload=runner._check_for_cancelled_flow_runs,
                    interval=self._query_seconds * 2,
                    run_once=run_once,
                    jitter_range=0.3,
                )
            )

stop

Stops the runner's polling cycle.

Source code in /home/runner/work/docs/docs/prefect_source/src/prefect/runner.py
302
303
304
305
306
307
308
309
def stop(self):
    """Stops the runner's polling cycle."""
    if not self.started:
        raise RuntimeError(
            "Runner has not yet started. Please start the runner by calling"
            " .start()"
        )
    self._loops_task_group.cancel_scope.cancel()

serve async

Serve the provided list of deployments.

Parameters:

Name Type Description Default
*args RunnerDeployment

A list of deployments to serve.

()
pause_on_shutdown bool

A boolean for whether or not to automatically pause deployment schedules on shutdown.

True
**kwargs

Additional keyword arguments to pass to the runner.

{}

Examples:

Prepare two deployments and serve them:

import datetime

from prefect import flow, serve

@flow
def my_flow(name):
    print(f"hello {name}")

@flow
def my_other_flow(name):
    print(f"goodbye {name}")

if __name__ == "__main__":
    # Run once a day
    hello_deploy = my_flow.to_deployment(
        "hello", tags=["dev"], interval=datetime.timedelta(days=1)
    )

    # Run every Sunday at 4:00 AM
    bye_deploy = my_other_flow.to_deployment(
        "goodbye", tags=["dev"], cron="0 4 * * sun"
    )

    serve(hello_deploy, bye_deploy)
Source code in /home/runner/work/docs/docs/prefect_source/src/prefect/runner.py
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
@sync_compatible
async def serve(
    *args: RunnerDeployment,
    pause_on_shutdown: bool = True,
    print_starting_message: bool = True,
    **kwargs,
):
    """
    Serve the provided list of deployments.

    Args:
        *args: A list of deployments to serve.
        pause_on_shutdown: A boolean for whether or not to automatically pause
            deployment schedules on shutdown.
        **kwargs: Additional keyword arguments to pass to the runner.

    Examples:
        Prepare two deployments and serve them:

        ```python
        import datetime

        from prefect import flow, serve

        @flow
        def my_flow(name):
            print(f"hello {name}")

        @flow
        def my_other_flow(name):
            print(f"goodbye {name}")

        if __name__ == "__main__":
            # Run once a day
            hello_deploy = my_flow.to_deployment(
                "hello", tags=["dev"], interval=datetime.timedelta(days=1)
            )

            # Run every Sunday at 4:00 AM
            bye_deploy = my_other_flow.to_deployment(
                "goodbye", tags=["dev"], cron="0 4 * * sun"
            )

            serve(hello_deploy, bye_deploy)
        ```
    """
    runner = Runner(pause_on_shutdown=pause_on_shutdown, **kwargs)
    for deployment in args:
        await runner.add_deployment(deployment)

    if print_starting_message:
        help_message_top = (
            "[green]Your deployments are being served and polling for"
            " scheduled runs!\n[/]"
        )

        table = Table(title="Deployments", show_header=False)

        table.add_column(style="blue", no_wrap=True)

        for deployment in args:
            table.add_row(f"{deployment.flow_name}/{deployment.name}")

        help_message_bottom = (
            "\nTo trigger any of these deployments, use the"
            " following command:\n[blue]\n\t$ prefect deployment run"
            " [DEPLOYMENT_NAME]\n[/]"
        )
        if PREFECT_UI_URL:
            help_message_bottom += (
                "\nYou can also trigger your deployments via the Prefect UI:"
                f" [blue]{PREFECT_UI_URL.value()}/deployments[/]\n"
            )

        console = Console()
        console.print(Panel(Group(help_message_top, table, help_message_bottom)))

    await runner.start()