Uploaded image for project: 'Apache Airflow'
  1. Apache Airflow
  2. AIRFLOW-7071

LocalExecutor queuing behaviour

    XMLWordPrintableJSON

    Details

    • Type: Improvement
    • Status: Open
    • Priority: Minor
    • Resolution: Unresolved
    • Affects Version/s: 1.10.8
    • Fix Version/s: None
    • Component/s: scheduler
    • Labels:
      None

      Description

      This is about question about possibility to improvement of queuing behaviour related to LocalExecutor.

       

      Here is example log from our test system.

      14:22:45 scheduler.1 | [2020-02-20 14:22:45,015] \{base_executor.py:122} DEBUG - 8 running task instances
14:22:45 scheduler.1 | [2020-02-20 14:22:45,015] \{base_executor.py:123} DEBUG - 0 in queue
14:22:45 scheduler.1 | [2020-02-20 14:22:45,015] \{base_executor.py:124} DEBUG - 0 open slots
14:22:45 scheduler.1 | [2020-02-20 14:22:45,016] \{base_executor.py:133} DEBUG - Calling the <class 'airflow.executors.local_executor.LocalExecutor'> sync method
14:22:47 scheduler.1 | [2020-02-20 14:22:47,200] \{base_executor.py:58} INFO - Adding to queue: ['airflow', 'run', 'DAG_TEST_6_5_20', 'dev.DAG_TEST_6_ENTITY_17_1', '2020-02-20T14:18:01.743479+00:00', '--local', '--pool', 'default_pool', '-sd', '/airflow/dags/dag_DAG_TEST_6_5_20.py']
14:22:47 scheduler.1 | [2020-02-20 14:22:47,200] \{base_executor.py:58} INFO - Adding to queue: ['airflow', 'run', 'DAG_TEST_6_5_20', 'dev.DAG_TEST_6_ENTITY_16_1', '2020-02-20T14:18:01.743479+00:00', '--local', '--pool', 'default_pool', '-sd', '/airflow/dags/dag_DAG_TEST_6_5_20.py']
14:22:47 scheduler.1 | [2020-02-20 14:22:47,201] \{base_executor.py:58} INFO - Adding to queue: ['airflow', 'run', 'DAG_TEST_6_5_20', 'dev.DAG_TEST_6_ENTITY_15_1', '2020-02-20T14:18:01.743479+00:00', '--local', '--pool', 'default_pool', '-sd', '/airflow/dags/dag_DAG_TEST_6_5_20.py']
14:22:47 scheduler.1 | [2020-02-20 14:22:47,202] \{base_executor.py:58} INFO - Adding to queue: ['airflow', 'run', 'DAG_TEST_6_5_20', 'dev.DAG_TEST_6_ENTITY_14_1', '2020-02-20T14:18:01.743479+00:00', '--local', '--pool', 'default_pool', '-sd', '/airflow/dags/dag_DAG_TEST_6_5_20.py']
14:22:47 scheduler.1 | [2020-02-20 14:22:47,202] \{base_executor.py:58} INFO - Adding to queue: ['airflow', 'run', 'DAG_TEST_6_5_20', 'dev.DAG_TEST_6_ENTITY_13_1', '2020-02-20T14:18:01.743479+00:00', '--local', '--pool', 'default_pool', '-sd', '/airflow/dags/dag_DAG_TEST_6_5_20.py']
14:22:47 scheduler.1 | [2020-02-20 14:22:47,203] \{base_executor.py:122} DEBUG - 8 running task instances
14:22:47 scheduler.1 | [2020-02-20 14:22:47,203] \{base_executor.py:123} DEBUG - 5 in queue
14:22:47 scheduler.1 | [2020-02-20 14:22:47,203] \{base_executor.py:124} DEBUG - 0 open slots
14:22:47 scheduler.1 | [2020-02-20 14:22:47,203] \{base_executor.py:133} DEBUG - Calling the <class 'airflow.executors.local_executor.LocalExecutor'> sync method
14:22:47 scheduler.1 | [2020-02-20 14:22:47,204] \{base_executor.py:157} DEBUG - Changing state: ('DAG_TEST_6_5_20', 'dev.DAG_TEST_6_ENTITY_4_1', datetime.datetime(2020, 2, 20, 14, 18, 1, 743479, tzinfo=<TimezoneInfo [UTC, GMT, +00:00:00, STD]>), 1)
14:22:47 scheduler.1 | [2020-02-20 14:22:47,205] \{base_executor.py:157} DEBUG - Changing state: ('DAG_TEST_6_5_20', 'dev.DAG_TEST_6_ENTITY_19_1', datetime.datetime(2020, 2, 20, 14, 18, 1, 743479, tzinfo=<TimezoneInfo [UTC, GMT, +00:00:00, STD]>), 1)
14:22:49 scheduler.1 | [2020-02-20 14:22:49,128] \{base_executor.py:122} DEBUG - 6 running task instances
14:22:49 scheduler.1 | [2020-02-20 14:22:49,128] \{base_executor.py:123} DEBUG - 0 in queue
14:22:49 scheduler.1 | [2020-02-20 14:22:49,128] \{base_executor.py:124} DEBUG - 2 open slots
14:22:49 scheduler.1 | [2020-02-20 14:22:49,128] \{base_executor.py:133} DEBUG - Calling the <class 'airflow.executors.local_executor.LocalExecutor'> sync method
14:22:49 scheduler.1 | [2020-02-20 14:22:49,129] \{base_executor.py:157} DEBUG - Changing state: ('DAG_TEST_6_5_20', 'dev.DAG_TEST_6_ENTITY_1_1', datetime.datetime(2020, 2, 20, 14, 18, 1, 743479, tzinfo=<TimezoneInfo [UTC, GMT, +00:00:00, STD]>), 1)
14:22:49 scheduler.1 | [2020-02-20 14:22:49,130] \{base_executor.py:157} DEBUG - Changing state: ('DAG_TEST_6_5_20', 'dev.DAG_TEST_6_ENTITY_3_1', datetime.datetime(2020, 2, 20, 14, 18, 1, 743479, tzinfo=<TimezoneInfo [UTC, GMT, +00:00:00, STD]>), 1)
14:22:49 scheduler.1 | [2020-02-20 14:22:49,131] \{base_executor.py:157} DEBUG - Changing state: ('DAG_TEST_6_5_20', 'dev.DAG_TEST_6_ENTITY_6_1', datetime.datetime(2020, 2, 20, 14, 18, 1, 743479, tzinfo=<TimezoneInfo [UTC, GMT, +00:00:00, STD]>), 1)
14:22:51 scheduler.1 | [2020-02-20 14:22:51,136] \{base_executor.py:122} DEBUG - 3 running task instances
14:22:51 scheduler.1 | [2020-02-20 14:22:51,136] \{base_executor.py:123} DEBUG - 0 in queue
14:22:51 scheduler.1 | [2020-02-20 14:22:51,136] \{base_executor.py:124} DEBUG - 5 open slots
14:22:51 scheduler.1 | [2020-02-20 14:22:51,137] \{base_executor.py:133} DEBUG - Calling the <class 'airflow.executors.local_executor.LocalExecutor'> sync method
14:22:53 scheduler.1 | [2020-02-20 14:22:53,104] \{base_executor.py:122} DEBUG - 3 running task instances
14:22:53 scheduler.1 | [2020-02-20 14:22:53,104] \{base_executor.py:123} DEBUG - 0 in queue
14:22:53 scheduler.1 | [2020-02-20 14:22:53,105] \{base_executor.py:124} DEBUG - 5 open slots
14:22:53 scheduler.1 | [2020-02-20 14:22:53,105] \{base_executor.py:133} DEBUG - Calling the <class 'airflow.executors.local_executor.LocalExecutor'> sync method
14:22:53 scheduler.1 | [2020-02-20 14:22:53,105] \{base_executor.py:157} DEBUG - Changing state: ('DAG_TEST_6_5_20', 'dev.DAG_TEST_6_ENTITY_5_1', datetime.datetime(2020, 2, 20, 14, 18, 1, 743479, tzinfo=<TimezoneInfo [UTC, GMT, +00:00:00, STD]>), 1)
14:22:55 scheduler.1 | [2020-02-20 14:22:55,111] \{base_executor.py:122} DEBUG - 2 running task instances
14:22:55 scheduler.1 | [2020-02-20 14:22:55,111] \{base_executor.py:123} DEBUG - 0 in queue
14:22:55 scheduler.1 | [2020-02-20 14:22:55,111] \{base_executor.py:124} DEBUG - 6 open slots
14:22:55 scheduler.1 | [2020-02-20 14:22:55,111] \{base_executor.py:133} DEBUG - Calling the <class 'airflow.executors.local_executor.LocalExecutor'> sync method
14:22:55 scheduler.1 | [2020-02-20 14:22:55,112] \{base_executor.py:157} DEBUG - Changing state: ('DAG_TEST_6_5_20', 'dev.DAG_TEST_6_ENTITY_10_1', datetime.datetime(2020, 2, 20, 14, 18, 1, 743479, tzinfo=<TimezoneInfo [UTC, GMT, +00:00:00, STD]>), 1)
14:22:57 scheduler.1 | [2020-02-20 14:22:57,263] \{base_executor.py:58} INFO - Adding to queue: ['airflow', 'run', 'DAG_TEST_6_5_20', 'load_dag_test_6_entity_5_2_45', '2020-02-20T14:18:01.743479+00:00', '--local', '--pool', 'default_pool', '-sd', '/airflow/dags/dag_DAG_TEST_6_5_20.py']
14:22:57 scheduler.1 | [2020-02-20 14:22:57,264] \{base_executor.py:58} INFO - Adding to queue: ['airflow', 'run', 'DAG_TEST_6_5_20', 'dev.DAG_TEST_6_ENTITY_17_1', '2020-02-20T14:18:01.743479+00:00', '--local', '--pool', 'default_pool', '-sd', '/airflow/dags/dag_DAG_TEST_6_5_20.py']
14:22:57 scheduler.1 | [2020-02-20 14:22:57,265] \{base_executor.py:58} INFO - Adding to queue: ['airflow', 'run', 'DAG_TEST_6_5_20', 'dev.DAG_TEST_6_ENTITY_16_1', '2020-02-20T14:18:01.743479+00:00', '--local', '--pool', 'default_pool', '-sd', '/airflow/dags/dag_DAG_TEST_6_5_20.py']
14:22:57 scheduler.1 | [2020-02-20 14:22:57,265] \{base_executor.py:58} INFO - Adding to queue: ['airflow', 'run', 'DAG_TEST_6_5_20', 'dev.DAG_TEST_6_ENTITY_15_1', '2020-02-20T14:18:01.743479+00:00', '--local', '--pool', 'default_pool', '-sd', '/airflow/dags/dag_DAG_TEST_6_5_20.py']
14:22:57 scheduler.1 | [2020-02-20 14:22:57,266] \{base_executor.py:58} INFO - Adding to queue: ['airflow', 'run', 'DAG_TEST_6_5_20', 'dev.DAG_TEST_6_ENTITY_14_1', '2020-02-20T14:18:01.743479+00:00', '--local', '--pool', 'default_pool', '-sd', '/airflow/dags/dag_DAG_TEST_6_5_20.py']
14:22:57 scheduler.1 | [2020-02-20 14:22:57,266] \{base_executor.py:58} INFO - Adding to queue: ['airflow', 'run', 'DAG_TEST_6_5_20', 'dev.DAG_TEST_6_ENTITY_13_1', '2020-02-20T14:18:01.743479+00:00', '--local', '--pool', 'default_pool', '-sd', '/airflow/dags/dag_DAG_TEST_6_5_20.py']
14:22:57 scheduler.1 | [2020-02-20 14:22:57,267] \{base_executor.py:58} INFO - Adding to queue: ['airflow', 'run', 'DAG_TEST_6_5_20', 'dev.DAG_TEST_6_ENTITY_12_1', '2020-02-20T14:18:01.743479+00:00', '--local', '--pool', 'default_pool', '-sd', '/airflow/dags/dag_DAG_TEST_6_5_20.py']
14:22:57 scheduler.1 | [2020-02-20 14:22:57,267] \{base_executor.py:58} INFO - Adding to queue: ['airflow', 'run', 'DAG_TEST_6_5_20', 'dev.DAG_TEST_6_ENTITY_11_1', '2020-02-20T14:18:01.743479+00:00', '--local', '--pool', 'default_pool', '-sd', '/airflow/dags/dag_DAG_TEST_6_5_20.py']
14:22:57 scheduler.1 | [2020-02-20 14:22:57,268] \{base_executor.py:122} DEBUG - 1 running task instances
14:22:57 scheduler.1 | [2020-02-20 14:22:57,268] \{base_executor.py:123} DEBUG - 8 in queue
14:22:57 scheduler.1 | [2020-02-20 14:22:57,268] \{base_executor.py:124} DEBUG - 7 open slots
14:22:57 scheduler.1 | [[2020-02-20 14:22:57,2812020-02-20 14:22:57,281] \{] {local_executor.py:base_executor.py:85} INFO133} DEBUG - - Calling the QueuedLocalWorker<class 'airflow.executors.local_executor.LocalExecutor'> running sync method
14:22:57 scheduler.1 | [2020-02-20 14:22:57,282] \{base_executor.py:157} DEBUG - Changing state: ('DAG_TEST_6_5_20', 'dev.DAG_TEST_6_ENTITY_18_1', datetime.datetime(2020, 2, 20, 14, 18, 1, 743479, tzinfo=<TimezoneInfo [UTC, GMT, +00:00:00, STD]>), 1)
14:22:59 scheduler.1 | [2020-02-20 14:22:59,124] \{base_executor.py:122} DEBUG - 7 running task instances
14:22:59 scheduler.1 | [2020-02-20 14:22:59,124] \{base_executor.py:123} DEBUG - 0 in queue
14:22:59 scheduler.1 | [2020-02-20 14:22:59,124] \{base_executor.py:124} DEBUG - 1 open slots
14:22:59 scheduler.1 | [2020-02-20 14:22:59,124] \{base_executor.py:133} DEBUG - Calling the <class 'airflow.executors.local_executor.LocalExecutor'> sync method

      The issue is that implementation clears queue and ordering is even so that LocalExecutor may been kept idle for no reason.

       

      E.g. 

      1. 8 running task instances 0 in queue 0 open slots
      2. 5 new tasks pushed to queue
      3. 2 tasks get ready
      4. 6 running task instances 0 in queue 2 slots open
      5. 3 tasks get ready
      6. 3 running task instances 0 in queue 5 open slots
      7. 1 tasks get ready
      8. 2 running task instances 0 in queue 6 open slots
      9. 8 new tasks pushed to queue
      10. 1 running task instances 8 in queue 7 open slots
      11. 1 task get ready
      12. 7 running task instances 0 in queue 1 slots open

      So depending how often dag get rescheduled it might take long before we get more tasks for LocalExecutor. Also it feels that now some other dags may get scheduled just at the time when some tasks get ready and then kind of take place of tasks queued before because those are always removed at then end of every scheduler loop.

      So would it be beneficial to keep queued tasks in local executor queue so that LocalExecutor can start processing next task as soon as slot is available instead of waiting until dag get scheduled next time. 

      Also as seen in 1,2,3,4 and 9,10,11,12 it seems that ordering of operations in LocalExecutor heartbeating cause that actually after scheduler loop slots got available are not filled all. Only those slots get used which are available before current heartbeating or at least it looks like it.

       

        Attachments

          Issue Links

          is caused by

          Bug - A problem which impairs or prevents the functions of the product. AIRFLOW-6175 Executor changes queued tasks to scheduled without removing from task queue

          • Blocker - Blocks development and/or testing work, production could not run
          • Resolved

          Improvement - An improvement or enhancement to an existing feature or task. AIRFLOW-2761 Parallelize Celery Executor enqueuing

          • Major - Major loss of function.
          • Resolved

            Activity

              People

              • Assignee:
                Unassigned
                Reporter:
                kattelus Marko Kattelus
              • Votes:
                0 Vote for this issue
                Watchers:
                1 Start watching this issue

                Dates

                • Created:
                  Updated: