metaflow 2.12.36__py2.py3-none-any.whl → 2.12.37__py2.py3-none-any.whl

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (45) hide show
  1. metaflow/__init__.py +3 -0
  2. metaflow/cli.py +84 -697
  3. metaflow/cli_args.py +17 -0
  4. metaflow/cli_components/__init__.py +0 -0
  5. metaflow/cli_components/dump_cmd.py +96 -0
  6. metaflow/cli_components/init_cmd.py +51 -0
  7. metaflow/cli_components/run_cmds.py +358 -0
  8. metaflow/cli_components/step_cmd.py +189 -0
  9. metaflow/cli_components/utils.py +140 -0
  10. metaflow/cmd/develop/stub_generator.py +9 -2
  11. metaflow/decorators.py +54 -2
  12. metaflow/extension_support/plugins.py +41 -27
  13. metaflow/flowspec.py +156 -16
  14. metaflow/includefile.py +50 -22
  15. metaflow/metaflow_config.py +1 -1
  16. metaflow/package.py +17 -3
  17. metaflow/parameters.py +80 -23
  18. metaflow/plugins/__init__.py +4 -0
  19. metaflow/plugins/airflow/airflow_cli.py +1 -0
  20. metaflow/plugins/argo/argo_workflows.py +41 -1
  21. metaflow/plugins/argo/argo_workflows_cli.py +1 -0
  22. metaflow/plugins/aws/batch/batch_decorator.py +2 -2
  23. metaflow/plugins/aws/step_functions/step_functions.py +32 -0
  24. metaflow/plugins/aws/step_functions/step_functions_cli.py +1 -0
  25. metaflow/plugins/datatools/s3/s3op.py +3 -3
  26. metaflow/plugins/kubernetes/kubernetes_cli.py +1 -1
  27. metaflow/plugins/kubernetes/kubernetes_decorator.py +2 -2
  28. metaflow/plugins/pypi/conda_decorator.py +22 -0
  29. metaflow/plugins/pypi/pypi_decorator.py +1 -0
  30. metaflow/plugins/timeout_decorator.py +2 -2
  31. metaflow/runner/click_api.py +73 -19
  32. metaflow/runtime.py +111 -73
  33. metaflow/sidecar/sidecar_worker.py +1 -1
  34. metaflow/user_configs/__init__.py +0 -0
  35. metaflow/user_configs/config_decorators.py +563 -0
  36. metaflow/user_configs/config_options.py +495 -0
  37. metaflow/user_configs/config_parameters.py +386 -0
  38. metaflow/util.py +17 -0
  39. metaflow/version.py +1 -1
  40. {metaflow-2.12.36.dist-info → metaflow-2.12.37.dist-info}/METADATA +3 -2
  41. {metaflow-2.12.36.dist-info → metaflow-2.12.37.dist-info}/RECORD +45 -35
  42. {metaflow-2.12.36.dist-info → metaflow-2.12.37.dist-info}/LICENSE +0 -0
  43. {metaflow-2.12.36.dist-info → metaflow-2.12.37.dist-info}/WHEEL +0 -0
  44. {metaflow-2.12.36.dist-info → metaflow-2.12.37.dist-info}/entry_points.txt +0 -0
  45. {metaflow-2.12.36.dist-info → metaflow-2.12.37.dist-info}/top_level.txt +0 -0
metaflow/cli.py CHANGED
@@ -1,18 +1,18 @@
1
+ import functools
1
2
  import inspect
2
- import json
3
3
  import sys
4
4
  import traceback
5
5
  from datetime import datetime
6
- from functools import wraps
7
6
 
8
7
  import metaflow.tracing as tracing
9
8
  from metaflow._vendor import click
10
- from metaflow.client.core import get_metadata
11
9
 
12
- from . import decorators, lint, metaflow_version, namespace, parameters, plugins
10
+ from . import decorators, lint, metaflow_version, parameters, plugins
13
11
  from .cli_args import cli_args
14
- from .datastore import FlowDataStore, TaskDataStore, TaskDataStoreSet
12
+ from .cli_components.utils import LazyGroup, LazyPluginCommandCollection
13
+ from .datastore import FlowDataStore
15
14
  from .exception import CommandException, MetaflowException
15
+ from .flowspec import _FlowState
16
16
  from .graph import FlowGraph
17
17
  from .metaflow_config import (
18
18
  DECOSPECS,
@@ -26,8 +26,6 @@ from .metaflow_config import (
26
26
  from .metaflow_current import current
27
27
  from metaflow.system import _system_monitor, _system_logger
28
28
  from .metaflow_environment import MetaflowEnvironment
29
- from .mflog import LOG_SOURCES, mflog
30
- from .package import MetaflowPackage
31
29
  from .plugins import (
32
30
  DATASTORES,
33
31
  ENVIRONMENTS,
@@ -37,16 +35,9 @@ from .plugins import (
37
35
  )
38
36
  from .pylint_wrapper import PyLint
39
37
  from .R import metaflow_r_version, use_r
40
- from .runtime import NativeRuntime
41
- from .tagging_util import validate_tags
42
- from .task import MetaflowTask
43
- from .unbounded_foreach import UBF_CONTROL, UBF_TASK
44
- from .util import (
45
- decompress_list,
46
- get_latest_run_id,
47
- resolve_identity,
48
- write_latest_run_id,
49
- )
38
+ from .util import resolve_identity
39
+ from .user_configs.config_options import LocalFileInput, config_options
40
+ from .user_configs.config_parameters import ConfigValue
50
41
 
51
42
  ERASE_TO_EOL = "\033[K"
52
43
  HIGHLIGHT = "red"
@@ -56,13 +47,6 @@ LOGGER_TIMESTAMP = "magenta"
56
47
  LOGGER_COLOR = "green"
57
48
  LOGGER_BAD_COLOR = "red"
58
49
 
59
- try:
60
- # Python 2
61
- import cPickle as pickle
62
- except ImportError:
63
- # Python 3
64
- import pickle
65
-
66
50
 
67
51
  def echo_dev_null(*args, **kwargs):
68
52
  pass
@@ -141,7 +125,16 @@ def config_merge_cb(ctx, param, value):
141
125
  return tuple(list(value) + DECOSPECS.split())
142
126
 
143
127
 
144
- @click.group()
128
+ @click.group(
129
+ cls=LazyGroup,
130
+ lazy_subcommands={
131
+ "init": "metaflow.cli_components.init_cmd.init",
132
+ "dump": "metaflow.cli_components.dump_cmd.dump",
133
+ "step": "metaflow.cli_components.step_cmd.step",
134
+ "run": "metaflow.cli_components.run_cmds.run",
135
+ "resume": "metaflow.cli_components.run_cmds.resume",
136
+ },
137
+ )
145
138
  def cli(ctx):
146
139
  pass
147
140
 
@@ -155,7 +148,13 @@ def cli(ctx):
155
148
  )
156
149
  @click.pass_obj
157
150
  def check(obj, warnings=False):
158
- _check(obj.graph, obj.flow, obj.environment, pylint=obj.pylint, warnings=warnings)
151
+ if obj.is_quiet:
152
+ echo = echo_dev_null
153
+ else:
154
+ echo = echo_always
155
+ _check(
156
+ echo, obj.graph, obj.flow, obj.environment, pylint=obj.pylint, warnings=warnings
157
+ )
159
158
  fname = inspect.getfile(obj.flow.__class__)
160
159
  echo(
161
160
  "\n*'{cmd} show'* shows a description of this flow.\n"
@@ -221,671 +220,32 @@ def output_dot(obj):
221
220
  echo_always(obj.graph.output_dot(), err=False)
222
221
 
223
222
 
224
- @cli.command(
225
- help="Get data artifacts of a task or all tasks in a step. "
226
- "The format for input-path is either <run_id>/<step_name> or "
227
- "<run_id>/<step_name>/<task_id>."
228
- )
229
- @click.argument("input-path")
230
- @click.option(
231
- "--private/--no-private",
232
- default=False,
233
- show_default=True,
234
- help="Show also private attributes.",
235
- )
236
- @click.option(
237
- "--max-value-size",
238
- default=1000,
239
- show_default=True,
240
- type=int,
241
- help="Show only values that are smaller than this number. "
242
- "Set to 0 to see only keys.",
243
- )
244
- @click.option(
245
- "--include",
246
- type=str,
247
- default="",
248
- help="Include only artifacts in the given comma-separated list.",
249
- )
250
- @click.option(
251
- "--file", type=str, default=None, help="Serialize artifacts in the given file."
252
- )
253
- @click.pass_obj
254
- def dump(obj, input_path, private=None, max_value_size=None, include=None, file=None):
255
- output = {}
256
- kwargs = {
257
- "show_private": private,
258
- "max_value_size": max_value_size,
259
- "include": {t for t in include.split(",") if t},
260
- }
261
-
262
- # Pathspec can either be run_id/step_name or run_id/step_name/task_id.
263
- parts = input_path.split("/")
264
- if len(parts) == 2:
265
- run_id, step_name = parts
266
- task_id = None
267
- elif len(parts) == 3:
268
- run_id, step_name, task_id = parts
269
- else:
270
- raise CommandException(
271
- "input_path should either be run_id/step_name or run_id/step_name/task_id"
272
- )
273
-
274
- datastore_set = TaskDataStoreSet(
275
- obj.flow_datastore,
276
- run_id,
277
- steps=[step_name],
278
- prefetch_data_artifacts=kwargs.get("include"),
279
- )
280
- if task_id:
281
- ds_list = [datastore_set.get_with_pathspec(input_path)]
282
- else:
283
- ds_list = list(datastore_set) # get all tasks
284
-
285
- for ds in ds_list:
286
- echo(
287
- "Dumping output of run_id=*{run_id}* "
288
- "step=*{step}* task_id=*{task_id}*".format(
289
- run_id=ds.run_id, step=ds.step_name, task_id=ds.task_id
290
- ),
291
- fg="magenta",
292
- )
293
-
294
- if file is None:
295
- echo_always(
296
- ds.format(**kwargs), highlight="green", highlight_bold=False, err=False
297
- )
298
- else:
299
- output[ds.pathspec] = ds.to_dict(**kwargs)
300
-
301
- if file is not None:
302
- with open(file, "wb") as f:
303
- pickle.dump(output, f, protocol=pickle.HIGHEST_PROTOCOL)
304
- echo("Artifacts written to *%s*" % file)
305
-
306
-
307
- # TODO - move step and init under a separate 'internal' subcommand
308
-
309
-
310
- @cli.command(help="Internal command to execute a single task.", hidden=True)
311
- @click.argument("step-name")
312
- @click.option(
313
- "--run-id",
314
- default=None,
315
- required=True,
316
- help="ID for one execution of all steps in the flow.",
317
- )
318
- @click.option(
319
- "--task-id",
320
- default=None,
321
- required=True,
322
- show_default=True,
323
- help="ID for this instance of the step.",
324
- )
325
- @click.option(
326
- "--input-paths",
327
- help="A comma-separated list of pathspecs specifying inputs for this step.",
328
- )
329
- @click.option(
330
- "--input-paths-filename",
331
- type=click.Path(exists=True, readable=True, dir_okay=False, resolve_path=True),
332
- help="A filename containing the argument typically passed to `input-paths`",
333
- hidden=True,
334
- )
335
- @click.option(
336
- "--split-index",
337
- type=int,
338
- default=None,
339
- show_default=True,
340
- help="Index of this foreach split.",
341
- )
342
- @click.option(
343
- "--tag",
344
- "opt_tag",
345
- multiple=True,
346
- default=None,
347
- help="Annotate this run with the given tag. You can specify "
348
- "this option multiple times to attach multiple tags in "
349
- "the task.",
350
- )
351
- @click.option(
352
- "--namespace",
353
- "opt_namespace",
354
- default=None,
355
- help="Change namespace from the default (your username) to the specified tag.",
356
- )
357
- @click.option(
358
- "--retry-count",
359
- default=0,
360
- help="How many times we have attempted to run this task.",
361
- )
362
- @click.option(
363
- "--max-user-code-retries",
364
- default=0,
365
- help="How many times we should attempt running the user code.",
366
- )
367
- @click.option(
368
- "--clone-only",
369
- default=None,
370
- help="Pathspec of the origin task for this task to clone. Do "
371
- "not execute anything.",
372
- )
373
- @click.option(
374
- "--clone-run-id",
375
- default=None,
376
- help="Run id of the origin flow, if this task is part of a flow being resumed.",
377
- )
378
- @click.option(
379
- "--with",
380
- "decospecs",
381
- multiple=True,
382
- help="Add a decorator to this task. You can specify this "
383
- "option multiple times to attach multiple decorators "
384
- "to this task.",
385
- )
386
- @click.option(
387
- "--ubf-context",
388
- default="none",
389
- type=click.Choice(["none", UBF_CONTROL, UBF_TASK]),
390
- help="Provides additional context if this task is of type unbounded foreach.",
391
- )
392
- @click.option(
393
- "--num-parallel",
394
- default=0,
395
- type=int,
396
- help="Number of parallel instances of a step. Ignored in local mode (see parallel decorator code).",
397
- )
398
- @click.pass_context
399
- def step(
400
- ctx,
401
- step_name,
402
- opt_tag=None,
403
- run_id=None,
404
- task_id=None,
405
- input_paths=None,
406
- input_paths_filename=None,
407
- split_index=None,
408
- opt_namespace=None,
409
- retry_count=None,
410
- max_user_code_retries=None,
411
- clone_only=None,
412
- clone_run_id=None,
413
- decospecs=None,
414
- ubf_context="none",
415
- num_parallel=None,
416
- ):
417
- if ubf_context == "none":
418
- ubf_context = None
419
- if opt_namespace is not None:
420
- namespace(opt_namespace or None)
421
-
422
- func = None
423
- try:
424
- func = getattr(ctx.obj.flow, step_name)
425
- except:
426
- raise CommandException("Step *%s* doesn't exist." % step_name)
427
- if not func.is_step:
428
- raise CommandException("Function *%s* is not a step." % step_name)
429
- echo("Executing a step, *%s*" % step_name, fg="magenta", bold=False)
430
-
431
- if decospecs:
432
- decorators._attach_decorators_to_step(func, decospecs)
433
-
434
- step_kwargs = ctx.params
435
- # Remove argument `step_name` from `step_kwargs`.
436
- step_kwargs.pop("step_name", None)
437
- # Remove `opt_*` prefix from (some) option keys.
438
- step_kwargs = dict(
439
- [(k[4:], v) if k.startswith("opt_") else (k, v) for k, v in step_kwargs.items()]
440
- )
441
- cli_args._set_step_kwargs(step_kwargs)
442
-
443
- ctx.obj.metadata.add_sticky_tags(tags=opt_tag)
444
- if not input_paths and input_paths_filename:
445
- with open(input_paths_filename, mode="r", encoding="utf-8") as f:
446
- input_paths = f.read().strip(" \n\"'")
447
-
448
- paths = decompress_list(input_paths) if input_paths else []
449
-
450
- task = MetaflowTask(
451
- ctx.obj.flow,
452
- ctx.obj.flow_datastore,
453
- ctx.obj.metadata,
454
- ctx.obj.environment,
455
- ctx.obj.echo,
456
- ctx.obj.event_logger,
457
- ctx.obj.monitor,
458
- ubf_context,
459
- )
460
- if clone_only:
461
- task.clone_only(
462
- step_name,
463
- run_id,
464
- task_id,
465
- clone_only,
466
- retry_count,
467
- )
468
- else:
469
- task.run_step(
470
- step_name,
471
- run_id,
472
- task_id,
473
- clone_run_id,
474
- paths,
475
- split_index,
476
- retry_count,
477
- max_user_code_retries,
478
- )
479
-
480
- echo("Success", fg="green", bold=True, indent=True)
481
-
482
-
483
- @parameters.add_custom_parameters(deploy_mode=False)
484
- @cli.command(help="Internal command to initialize a run.", hidden=True)
485
- @click.option(
486
- "--run-id",
487
- default=None,
488
- required=True,
489
- help="ID for one execution of all steps in the flow.",
490
- )
491
- @click.option(
492
- "--task-id", default=None, required=True, help="ID for this instance of the step."
493
- )
494
- @click.option(
495
- "--tag",
496
- "tags",
497
- multiple=True,
498
- default=None,
499
- help="Tags for this instance of the step.",
500
- )
501
- @click.pass_obj
502
- def init(obj, run_id=None, task_id=None, tags=None, **kwargs):
503
- # init is a separate command instead of an option in 'step'
504
- # since we need to capture user-specified parameters with
505
- # @add_custom_parameters. Adding custom parameters to 'step'
506
- # is not desirable due to the possibility of name clashes between
507
- # user-specified parameters and our internal options. Note that
508
- # user-specified parameters are often defined as environment
509
- # variables.
510
-
511
- obj.metadata.add_sticky_tags(tags=tags)
512
-
513
- runtime = NativeRuntime(
514
- obj.flow,
515
- obj.graph,
516
- obj.flow_datastore,
517
- obj.metadata,
518
- obj.environment,
519
- obj.package,
520
- obj.logger,
521
- obj.entrypoint,
522
- obj.event_logger,
523
- obj.monitor,
524
- run_id=run_id,
525
- )
526
- obj.flow._set_constants(obj.graph, kwargs)
527
- runtime.persist_constants(task_id=task_id)
528
-
529
-
530
- def common_run_options(func):
531
- @click.option(
532
- "--tag",
533
- "tags",
534
- multiple=True,
535
- default=None,
536
- help="Annotate this run with the given tag. You can specify "
537
- "this option multiple times to attach multiple tags in "
538
- "the run.",
539
- )
540
- @click.option(
541
- "--max-workers",
542
- default=16,
543
- show_default=True,
544
- help="Maximum number of parallel processes.",
545
- )
546
- @click.option(
547
- "--max-num-splits",
548
- default=100,
549
- show_default=True,
550
- help="Maximum number of splits allowed in a foreach. This "
551
- "is a safety check preventing bugs from triggering "
552
- "thousands of steps inadvertently.",
553
- )
554
- @click.option(
555
- "--max-log-size",
556
- default=10,
557
- show_default=True,
558
- help="Maximum size of stdout and stderr captured in "
559
- "megabytes. If a step outputs more than this to "
560
- "stdout/stderr, its output will be truncated.",
561
- )
562
- @click.option(
563
- "--with",
564
- "decospecs",
565
- multiple=True,
566
- help="Add a decorator to all steps. You can specify this "
567
- "option multiple times to attach multiple decorators "
568
- "in steps.",
569
- )
570
- @click.option(
571
- "--run-id-file",
572
- default=None,
573
- show_default=True,
574
- type=str,
575
- help="Write the ID of this run to the file specified.",
576
- )
577
- @click.option(
578
- "--runner-attribute-file",
579
- default=None,
580
- show_default=True,
581
- type=str,
582
- help="Write the metadata and pathspec of this run to the file specified. Used internally for Metaflow's Runner API.",
583
- )
584
- @wraps(func)
585
- def wrapper(*args, **kwargs):
586
- return func(*args, **kwargs)
587
-
588
- return wrapper
589
-
590
-
591
- @click.option(
592
- "--origin-run-id",
593
- default=None,
594
- help="ID of the run that should be resumed. By default, the "
595
- "last run executed locally.",
596
- )
597
- @click.option(
598
- "--run-id",
599
- default=None,
600
- help="Run ID for the new run. By default, a new run-id will be generated",
601
- hidden=True,
602
- )
603
- @click.option(
604
- "--clone-only/--no-clone-only",
605
- default=False,
606
- show_default=True,
607
- help="Only clone tasks without continuing execution",
608
- hidden=True,
609
- )
610
- @click.option(
611
- "--reentrant/--no-reentrant",
612
- default=False,
613
- show_default=True,
614
- hidden=True,
615
- help="If specified, allows this call to be called in parallel",
616
- )
617
- @click.option(
618
- "--resume-identifier",
619
- default=None,
620
- show_default=True,
621
- hidden=True,
622
- help="If specified, it identifies the task that started this resume call. It is in the form of {step_name}-{task_id}",
623
- )
624
- @click.argument("step-to-rerun", required=False)
625
- @cli.command(help="Resume execution of a previous run of this flow.")
626
- @common_run_options
627
- @click.pass_obj
628
- def resume(
629
- obj,
630
- tags=None,
631
- step_to_rerun=None,
632
- origin_run_id=None,
633
- run_id=None,
634
- clone_only=False,
635
- reentrant=False,
636
- max_workers=None,
637
- max_num_splits=None,
638
- max_log_size=None,
639
- decospecs=None,
640
- run_id_file=None,
641
- resume_identifier=None,
642
- runner_attribute_file=None,
643
- ):
644
- before_run(obj, tags, decospecs)
645
-
646
- if origin_run_id is None:
647
- origin_run_id = get_latest_run_id(obj.echo, obj.flow.name)
648
- if origin_run_id is None:
649
- raise CommandException(
650
- "A previous run id was not found. Specify --origin-run-id."
651
- )
652
-
653
- if step_to_rerun is None:
654
- steps_to_rerun = set()
655
- else:
656
- # validate step name
657
- if step_to_rerun not in obj.graph.nodes:
658
- raise CommandException(
659
- "invalid step name {0} specified, must be step present in "
660
- "current form of execution graph. Valid step names include: {1}".format(
661
- step_to_rerun, ",".join(list(obj.graph.nodes.keys()))
662
- )
663
- )
664
- steps_to_rerun = {step_to_rerun}
665
-
666
- if run_id:
667
- # Run-ids that are provided by the metadata service are always integers.
668
- # External providers or run-ids (like external schedulers) always need to
669
- # be non-integers to avoid any clashes. This condition ensures this.
670
- try:
671
- int(run_id)
672
- except:
673
- pass
674
- else:
675
- raise CommandException("run-id %s cannot be an integer" % run_id)
676
-
677
- runtime = NativeRuntime(
678
- obj.flow,
679
- obj.graph,
680
- obj.flow_datastore,
681
- obj.metadata,
682
- obj.environment,
683
- obj.package,
684
- obj.logger,
685
- obj.entrypoint,
686
- obj.event_logger,
687
- obj.monitor,
688
- run_id=run_id,
689
- clone_run_id=origin_run_id,
690
- clone_only=clone_only,
691
- reentrant=reentrant,
692
- steps_to_rerun=steps_to_rerun,
693
- max_workers=max_workers,
694
- max_num_splits=max_num_splits,
695
- max_log_size=max_log_size * 1024 * 1024,
696
- resume_identifier=resume_identifier,
697
- )
698
- write_file(run_id_file, runtime.run_id)
699
- runtime.print_workflow_info()
700
-
701
- runtime.persist_constants()
702
-
703
- if runner_attribute_file:
704
- with open(runner_attribute_file, "w", encoding="utf-8") as f:
705
- json.dump(
706
- {
707
- "run_id": runtime.run_id,
708
- "flow_name": obj.flow.name,
709
- "metadata": obj.metadata.metadata_str(),
710
- },
711
- f,
712
- )
713
-
714
- # We may skip clone-only resume if this is not a resume leader,
715
- # and clone is already complete.
716
- if runtime.should_skip_clone_only_execution():
717
- return
718
-
719
- current._update_env(
720
- {
721
- "run_id": runtime.run_id,
722
- }
723
- )
724
- _system_logger.log_event(
725
- level="info",
726
- module="metaflow.resume",
727
- name="start",
728
- payload={
729
- "msg": "Resuming run",
730
- },
731
- )
732
-
733
- with runtime.run_heartbeat():
734
- if clone_only:
735
- runtime.clone_original_run()
736
- else:
737
- runtime.clone_original_run(generate_task_obj=True, verbose=False)
738
- runtime.execute()
739
-
740
-
741
- @tracing.cli_entrypoint("cli/run")
742
- @parameters.add_custom_parameters(deploy_mode=True)
743
- @cli.command(help="Run the workflow locally.")
744
- @common_run_options
745
- @click.option(
746
- "--namespace",
747
- "user_namespace",
748
- default=None,
749
- help="Change namespace from the default (your username) to "
750
- "the specified tag. Note that this option does not alter "
751
- "tags assigned to the objects produced by this run, just "
752
- "what existing objects are visible in the client API. You "
753
- "can enable the global namespace with an empty string."
754
- "--namespace=",
755
- )
756
- @click.pass_obj
757
- def run(
758
- obj,
759
- tags=None,
760
- max_workers=None,
761
- max_num_splits=None,
762
- max_log_size=None,
763
- decospecs=None,
764
- run_id_file=None,
765
- runner_attribute_file=None,
766
- user_namespace=None,
767
- **kwargs
768
- ):
769
- if user_namespace is not None:
770
- namespace(user_namespace or None)
771
- before_run(obj, tags, decospecs)
772
-
773
- runtime = NativeRuntime(
774
- obj.flow,
775
- obj.graph,
776
- obj.flow_datastore,
777
- obj.metadata,
778
- obj.environment,
779
- obj.package,
780
- obj.logger,
781
- obj.entrypoint,
782
- obj.event_logger,
783
- obj.monitor,
784
- max_workers=max_workers,
785
- max_num_splits=max_num_splits,
786
- max_log_size=max_log_size * 1024 * 1024,
787
- )
788
- write_latest_run_id(obj, runtime.run_id)
789
- write_file(run_id_file, runtime.run_id)
790
-
791
- obj.flow._set_constants(obj.graph, kwargs)
792
- current._update_env(
793
- {
794
- "run_id": runtime.run_id,
795
- }
796
- )
797
- _system_logger.log_event(
798
- level="info",
799
- module="metaflow.run",
800
- name="start",
801
- payload={
802
- "msg": "Starting run",
803
- },
804
- )
805
- with runtime.run_heartbeat():
806
- runtime.print_workflow_info()
807
- runtime.persist_constants()
808
-
809
- if runner_attribute_file:
810
- with open(runner_attribute_file, "w", encoding="utf-8") as f:
811
- json.dump(
812
- {
813
- "run_id": runtime.run_id,
814
- "flow_name": obj.flow.name,
815
- "metadata": obj.metadata.metadata_str(),
816
- },
817
- f,
818
- )
819
- runtime.execute()
820
-
821
-
822
- def write_file(file_path, content):
823
- if file_path is not None:
824
- with open(file_path, "w") as f:
825
- f.write(str(content))
826
-
827
-
828
- def before_run(obj, tags, decospecs):
829
- validate_tags(tags)
830
-
831
- # There's a --with option both at the top-level and for the run
832
- # subcommand. Why?
833
- #
834
- # "run --with shoes" looks so much better than "--with shoes run".
835
- # This is a very common use case of --with.
836
- #
837
- # A downside is that we need to have the following decorators handling
838
- # in two places in this module and make sure _init_step_decorators
839
- # doesn't get called twice.
840
-
841
- # We want the order to be the following:
842
- # - run level decospecs
843
- # - top level decospecs
844
- # - environment decospecs
845
- all_decospecs = (
846
- list(decospecs or [])
847
- + obj.tl_decospecs
848
- + list(obj.environment.decospecs() or [])
849
- )
850
- if all_decospecs:
851
- decorators._attach_decorators(obj.flow, all_decospecs)
852
- obj.graph = FlowGraph(obj.flow.__class__)
853
-
854
- obj.check(obj.graph, obj.flow, obj.environment, pylint=obj.pylint)
855
- # obj.environment.init_environment(obj.logger)
856
-
857
- decorators._init_step_decorators(
858
- obj.flow, obj.graph, obj.environment, obj.flow_datastore, obj.logger
859
- )
860
-
861
- obj.metadata.add_sticky_tags(tags=tags)
862
-
863
- # Package working directory only once per run.
864
- # We explicitly avoid doing this in `start` since it is invoked for every
865
- # step in the run.
866
- obj.package = MetaflowPackage(
867
- obj.flow, obj.environment, obj.echo, obj.package_suffixes
868
- )
869
-
870
-
871
223
  @cli.command(help="Print the Metaflow version")
872
224
  @click.pass_obj
873
225
  def version(obj):
874
226
  echo_always(obj.version)
875
227
 
876
228
 
877
- @tracing.cli_entrypoint("cli/start")
229
+ # NOTE: add_decorator_options should be TL because it checks to make sure
230
+ # that no option conflict with the ones below
878
231
  @decorators.add_decorator_options
232
+ @config_options
879
233
  @click.command(
880
- cls=click.CommandCollection,
881
- sources=[cli] + plugins.get_plugin_cli(),
234
+ cls=LazyPluginCommandCollection,
235
+ sources=[cli],
236
+ lazy_sources=plugins.get_plugin_cli_path(),
882
237
  invoke_without_command=True,
883
238
  )
239
+ @tracing.cli_entrypoint("cli/start")
240
+ # Quiet is eager to make sure it is available when processing --config options since
241
+ # we need it to construct a context to pass to any DeployTimeField for the default
242
+ # value.
884
243
  @click.option(
885
244
  "--quiet/--not-quiet",
886
245
  show_default=True,
887
246
  default=False,
888
247
  help="Suppress unnecessary messages",
248
+ is_eager=True,
889
249
  )
890
250
  @click.option(
891
251
  "--metadata",
@@ -901,12 +261,14 @@ def version(obj):
901
261
  type=click.Choice(["local"] + [m.TYPE for m in ENVIRONMENTS]),
902
262
  help="Execution environment type",
903
263
  )
264
+ # See comment for --quiet
904
265
  @click.option(
905
266
  "--datastore",
906
267
  default=DEFAULT_DATASTORE,
907
268
  show_default=True,
908
269
  type=click.Choice([d.TYPE for d in DATASTORES]),
909
270
  help="Data backend type",
271
+ is_eager=True,
910
272
  )
911
273
  @click.option("--datastore-root", help="Root path for datastore")
912
274
  @click.option(
@@ -943,6 +305,15 @@ def version(obj):
943
305
  type=click.Choice(MONITOR_SIDECARS),
944
306
  help="Monitoring backend type",
945
307
  )
308
+ @click.option(
309
+ "--local-config-file",
310
+ type=LocalFileInput(exists=True, readable=True, dir_okay=False, resolve_path=True),
311
+ required=False,
312
+ default=None,
313
+ help="A filename containing the dumped configuration values. Internal use only.",
314
+ hidden=True,
315
+ is_eager=True,
316
+ )
946
317
  @click.pass_context
947
318
  def start(
948
319
  ctx,
@@ -956,9 +327,11 @@ def start(
956
327
  pylint=None,
957
328
  event_logger=None,
958
329
  monitor=None,
330
+ local_config_file=None,
331
+ config_file_options=None,
332
+ config_value_options=None,
959
333
  **deco_options
960
334
  ):
961
- global echo
962
335
  if quiet:
963
336
  echo = echo_dev_null
964
337
  else:
@@ -973,17 +346,27 @@ def start(
973
346
  echo(" executing *%s*" % ctx.obj.flow.name, fg="magenta", nl=False)
974
347
  echo(" for *%s*" % resolve_identity(), fg="magenta")
975
348
 
349
+ # At this point, we are able to resolve the user-configuration options so we can
350
+ # process all those decorators that the user added that will modify the flow based
351
+ # on those configurations. It is important to do this as early as possible since it
352
+ # actually modifies the flow itself
353
+
354
+ # When we process the options, the first one processed will return None and the
355
+ # second one processed will return the actual options. The order of processing
356
+ # depends on what (and in what order) the user specifies on the command line.
357
+ config_options = config_file_options or config_value_options
358
+ ctx.obj.flow = ctx.obj.flow._process_config_decorators(config_options)
359
+
976
360
  cli_args._set_top_kwargs(ctx.params)
977
361
  ctx.obj.echo = echo
978
362
  ctx.obj.echo_always = echo_always
979
363
  ctx.obj.is_quiet = quiet
980
- ctx.obj.graph = FlowGraph(ctx.obj.flow.__class__)
364
+ ctx.obj.graph = ctx.obj.flow._graph
981
365
  ctx.obj.logger = logger
982
- ctx.obj.check = _check
983
366
  ctx.obj.pylint = pylint
367
+ ctx.obj.check = functools.partial(_check, echo)
984
368
  ctx.obj.top_cli = cli
985
369
  ctx.obj.package_suffixes = package_suffixes.split(",")
986
- ctx.obj.reconstruct_cli = _reconstruct_cli
987
370
 
988
371
  ctx.obj.environment = [
989
372
  e for e in ENVIRONMENTS + [MetaflowEnvironment] if e.TYPE == environment
@@ -1029,6 +412,10 @@ def start(
1029
412
  ctx.obj.monitor,
1030
413
  )
1031
414
 
415
+ ctx.obj.config_options = config_options
416
+
417
+ decorators._init(ctx.obj.flow)
418
+
1032
419
  # It is important to initialize flow decorators early as some of the
1033
420
  # things they provide may be used by some of the objects initialized after.
1034
421
  decorators._init_flow_decorators(
@@ -1051,10 +438,22 @@ def start(
1051
438
  # initialize current and parameter context for deploy-time parameters
1052
439
  current._set_env(flow=ctx.obj.flow, is_running=False)
1053
440
  parameters.set_parameter_context(
1054
- ctx.obj.flow.name, ctx.obj.echo, ctx.obj.flow_datastore
441
+ ctx.obj.flow.name,
442
+ ctx.obj.echo,
443
+ ctx.obj.flow_datastore,
444
+ {
445
+ k: ConfigValue(v)
446
+ for k, v in ctx.obj.flow.__class__._flow_state.get(
447
+ _FlowState.CONFIGS, {}
448
+ ).items()
449
+ },
1055
450
  )
1056
451
 
1057
- if ctx.invoked_subcommand not in ("run", "resume"):
452
+ if (
453
+ hasattr(ctx, "saved_args")
454
+ and ctx.saved_args
455
+ and ctx.saved_args[0] not in ("run", "resume")
456
+ ):
1058
457
  # run/resume are special cases because they can add more decorators with --with,
1059
458
  # so they have to take care of themselves.
1060
459
  all_decospecs = ctx.obj.tl_decospecs + list(
@@ -1062,6 +461,7 @@ def start(
1062
461
  )
1063
462
  if all_decospecs:
1064
463
  decorators._attach_decorators(ctx.obj.flow, all_decospecs)
464
+ decorators._init(ctx.obj.flow, only_non_static=True)
1065
465
  # Regenerate graph if we attached more decorators
1066
466
  ctx.obj.graph = FlowGraph(ctx.obj.flow.__class__)
1067
467
 
@@ -1075,25 +475,12 @@ def start(
1075
475
 
1076
476
  # TODO (savin): Enable lazy instantiation of package
1077
477
  ctx.obj.package = None
478
+
1078
479
  if ctx.invoked_subcommand is None:
1079
480
  ctx.invoke(check)
1080
481
 
1081
482
 
1082
- def _reconstruct_cli(params):
1083
- for k, v in params.items():
1084
- if v:
1085
- if k == "decospecs":
1086
- k = "with"
1087
- k = k.replace("_", "-")
1088
- if not isinstance(v, tuple):
1089
- v = [v]
1090
- for value in v:
1091
- yield "--%s" % k
1092
- if not isinstance(value, bool):
1093
- yield str(value)
1094
-
1095
-
1096
- def _check(graph, flow, environment, pylint=True, warnings=False, **kwargs):
483
+ def _check(echo, graph, flow, environment, pylint=True, warnings=False, **kwargs):
1097
484
  echo("Validating your flow...", fg="magenta", bold=False)
1098
485
  linter = lint.linter
1099
486
  # TODO set linter settings