Can't read S3 remote logs when using gevent/eventlent webserver workers.

[webster-chainalysis]

Hey everyone. I’ve upgraded to 1.10.9. It appears that the logging changes broke the functionality for reading S3 remote logs in the Web UI (writing is ok). In the change log it mentions that Airflow's logging mechanism has been refactored to uses Python’s builtin logging module:

[AIRFLOW-1611] Customize logging

I followed the directions in the changelog and created the following log config:

import os
import six

from airflow import AirflowException
from airflow.configuration import conf
from airflow.utils.file import mkdirs
from typing import Dict, Any

LOG_LEVEL = conf.get('core', 'LOGGING_LEVEL').upper()

FAB_LOG_LEVEL = conf.get('core', 'FAB_LOGGING_LEVEL').upper()

LOG_FORMAT = conf.get('core', 'LOG_FORMAT')

COLORED_LOG_FORMAT = conf.get('core', 'COLORED_LOG_FORMAT')

COLORED_LOG = conf.getboolean('core', 'COLORED_CONSOLE_LOG')

COLORED_FORMATTER_CLASS = conf.get('core', 'COLORED_FORMATTER_CLASS')

BASE_LOG_FOLDER = conf.get('core', 'BASE_LOG_FOLDER')

PROCESSOR_LOG_FOLDER = conf.get('scheduler', 'CHILD_PROCESS_LOG_DIRECTORY')

DAG_PROCESSOR_MANAGER_LOG_LOCATION = \
    conf.get('core', 'DAG_PROCESSOR_MANAGER_LOG_LOCATION')

FILENAME_TEMPLATE = conf.get('core', 'LOG_FILENAME_TEMPLATE')

PROCESSOR_FILENAME_TEMPLATE = conf.get('core', 'LOG_PROCESSOR_FILENAME_TEMPLATE')

FORMATTER_CLASS_KEY = '()' if six.PY2 else 'class'


#
# Getting this from environment because the changelog for 1.10.9 says to set
# the path of `REMOTE_BASE_LOG_FOLDER` explicitly in the config. The
# `REMOTE_BASE_LOG_FOLDER` key is not used anymore.
#
REMOTE_BASE_LOG_FOLDER = os.environ.get('AIRFLOW__CORE__REMOTE_BASE_LOG_FOLDER')

LOGGING_CONFIG = {
    'version': 1,
    'disable_existing_loggers': False,
    'formatters': {
        'airflow.task': {
            'format': LOG_FORMAT,
        },
        'airflow.processor': {
            'format': LOG_FORMAT,
        },
        'airflow_coloured': {
            'format': COLORED_LOG_FORMAT if COLORED_LOG else LOG_FORMAT,
            FORMATTER_CLASS_KEY: COLORED_FORMATTER_CLASS if COLORED_LOG else 'logging.Formatter'
        },
    },
    'handlers': {
        'console': {
            'class': 'logging.StreamHandler',
            'formatter': 'airflow.task',
            'stream': 'ext://sys.stdout'
        },
        'file.task': {
            'class': 'airflow.utils.log.file_task_handler.FileTaskHandler',
            'formatter': 'airflow.task',
            'base_log_folder': os.path.expanduser(BASE_LOG_FOLDER),
            'filename_template': FILENAME_TEMPLATE,
        },
        'file.processor': {
            'class': 'airflow.utils.log.file_processor_handler.FileProcessorHandler',
            'formatter': 'airflow.processor',
            'base_log_folder': os.path.expanduser(PROCESSOR_LOG_FOLDER),
            'filename_template': PROCESSOR_FILENAME_TEMPLATE,
        },
        's3.task': {
            'class': 'airflow.utils.log.s3_task_handler.S3TaskHandler',
            'formatter': 'airflow.task',
            'base_log_folder': os.path.expanduser(BASE_LOG_FOLDER),
            's3_log_folder': REMOTE_BASE_LOG_FOLDER,
            'filename_template': FILENAME_TEMPLATE,
        },
    },
    'loggers': {
        '': {
            'handlers': ['console'],
            'level': LOG_LEVEL
        },
        'airflow': {
            'handlers': ['console'],
            'level': LOG_LEVEL,
            'propagate': False,
        },
        'airflow.processor': {
            'handlers': ['file.processor'],
            'level': LOG_LEVEL,
            'propagate': True,
        },
        'airflow.task': {
            'handlers': ['s3.task'],
            'level': LOG_LEVEL,
            'propagate': False,
        },
        'airflow.task_runner': {
            'handlers': ['s3.task'],
            'level': LOG_LEVEL,
            'propagate': True,
        },
    }
}  # type: Dict[str, Any]

However, the task log reader is always defaulting to using the FileTaskHandler. This should not occur because I have the following settings in airflow.cfg:

remote_logging = True
remote_base_log_folder = s3://my-bucket-name
remote_log_conn_id = aws_default
task_log_reader = s3.task

The s3.task handler passed to the task_log_reader setting should be creating an instance of the S3TaskHandler class to read the task logs to from S3. This occurs when rendering the get_logs_with_metadata view in www/views.py.

---

Apache Airflow version: 1.10.9
Kubernetes version: 1.15

Environment:

• Cloud provider or hardware configuration: AWS
• OS (e.g. from /etc/os-release): python:3.7.6-buster docker image

What happened: Logs did not appear in the Airflow Web UI. The FileTaskHandler tries to fetch the file locally or from the worker on port 8793. However, the logs do not exist in either location since we are using the Kubernetes Executor. This produces the following errors messages:

*** Log file does not exist: /usr/local/airflow/logs/MY_DAG_NAME/MY_TASK_NAME/2020-04-07T20:59:19.312402+00:00/6.log
*** Fetching from: http://MY_DAG_NAME-0dde5ff5a786437cb14234:8793/log/MY_DAG_NAME/MY_TASK_NAME/2020-04-07T20:59:19.312402+00:00/6.log
*** Failed to fetch log file from worker. HTTPConnectionPool(host='MY_DAG_NAME-0dde5ff5a786437cb14234', port=8793): Max retries exceeded with url: /log/MY_DAG_NAME/MY_TASK_NAME/2020-04-07T20:59:19.312402+00:00/6.log (Caused by NewConnectionError('<urllib3.connection.HTTPConnection object at 0x7f708332fc90>: Failed to establish a new connection: [Errno -2] Name or service not known'))

What you expected to happen:

The logs should be rendered in the Web UI using the S3TaskHandler class.

[boring-cyborg]

Thanks for opening your first issue here! Be sure to follow the issue template!

[dimon222]

@webster-chainalysis
I noticed such issue on 1.10.10. Do you also have same behaviour on this version?

[webster-chainalysis]

@dimon222 - Yes.

[dimon222]

@webster-chainalysis
Worker class switch in airflow.cfg from gevent/eventlet to "sync" seem to resolve the issue for me.

[webster-chainalysis]

@dimon222 - That makes sense. I'm currently using gevent behind an Amazon elastic load balancer. When using the sync worker class with the ELB I had unacceptable performance.

[TRReeve]

I have had this issue as well on Airflow 10.10 and been tearing my hear out at it a couple of days now. I'm running using the KubernetesOperator and the logging seems to upload to S3 fine it is only the reading of the log back from the web ui that is always in local mode.

[TRReeve]

Having gotten remote logging working on 1.10.10 I've noticed there seems to be a difference in how the workers handle the log upload versus the webserver component. The Connection you define needs to point to the same folder as the remote_base_logs_folder. E.g if your remote base logs folder is data/airflow/logs then your connection used for remote logging also needs to point to it e.g s3://access_key:secret@data/airflow/logs. You cannot just have a generic "AWS connection" and then have it figure out which folder it needs to point to on its own using the filepath from the worker.

[marclamberti]

@TRReeve I didn’t need to do that. Actually, if you want to have the remote logging working with the kubernetes executor, you have to define additional kubernetes environment variables so that your PODs are in sync with the scheduler/web server. Eg:
AIRFLOW__CORE__REMOTE_LOGGING=True
AIRFLOW__CORE__REMOTE_BASE_LOGS_FOLDER=s3://my-bucket/my-key
AIRFLOW__CORE__REMOTE_LOG_CONN_ID=myawsconn
AIRFLOW__CORE__FERNET_KEY=myfernetkey

AIRFLOW__KUBERNETES_ENVIRONMENT_VARIABALES__AIRFLOW__CORE__REMOTE_LOGGING=True
AIRFLOW__KUBERNETES_ENVIRONMENT_VARIABALES__AIRFLOW__CORE__REMOTE_BASE_LOGS_FOLDER=s3://my-bucket/my-key
AIRFLOW__KUBERNETES_ENVIRONMENT_VARIABALES__AIRFLOW__CORE__REMOTE_LOG_CONN_ID=myawsconn
AIRFLOW__KUBERNETES_ENVIRONMENT_VARIABALES__AIRFLOW__CORE__FERNET_KEY=myfernetkey

The connection myawsconn with type S3 and extra field

• aws_access_key_id
• aws_secret_acess_key_id
• region_name

Also I didn’t need to change the task handler as it is automatically changed when remote logging is set to True, and I didn’t have to define a custom logging class.

Hope it helps :)

[TRReeve]

@marclamberti Your answer is how I understood it would work as well and it half worked for me I would get the logs uploading fine into the S3 bucket but then when i went to "view logs" in the UI it would give the "logs not found" error with no output in the logs to indicate it was using the s3 connection or the read_key function to retrieve anything.

It would be really nice if I could just define AIRFLOW_CONN_S3_URI = s3://user:pass@S3 then have REMOTE_BASE_LOGS_FOLDER=s3://airflow-stuff/logs and the UI would build the path but I could only get logs uploading. My working helm template for airflow on k8s builds the connection s3://access_key:secret_key@{{ mys3path }} and then remote_log_path is s3://{{ mys3path }}. Aside that it's exactly the same as you defined above with the same variables defined under AIRFLOW__KUBERNETES__ENVIRONMENT_VARIABLES.

And yes I can confirm for anyone reading there was no need for any custom logging classes or task handlers.

[marclamberti]

Well, in my case it works for both reading and writing logs from S3 with Airflow 1.10.10 and the Kubernetes executor
I agreed it could be simpler :)

[ashb]

@webster-chainalysis I've just opened #9118 to improve the debugging of S3 log handler.

Could you also try the debug steps I put in that description there -- lets rule out that sort of problem.

[prithvisathiya]

Im also experiencing the exact same issue when I upgraded to 1.10.9, but Im still using LocalExecutor. I can clearly see from the S3 console that the logs are getting uploaded, but the Airflow UI is unable to read it back. It attempts to read the local folder and then just gives up with the following error:

*** Log file does not exist: /app/logs/dag_name/task_name/2020-07-05T19:21:09.715128+00:00/1.log
*** Fetching from: http://airflow-test-scheduler-5f8ccc76df-tc8j8:8793/log/dag_name/task_name/2020-07-05T19:21:09.715128+00:00/1.log
*** Failed to fetch log file from worker. HTTPConnectionPool(host='airflow-test-scheduler-5f8ccc76df-tc8j8', port=8793): Max retries exceeded with url: /log/dag_name/task_name/2020-07-05T19:21:09.715128+00:00/1.log 

My configurations below:

AIRFLOW__CORE__EXECUTOR=LocalExecutor
AIRFLOW__CORE__REMOTE_BASE_LOG_FOLDER=s3://{bucket}/logs/
AIRFLOW__CORE__REMOTE_LOGGING=True
AIRFLOW__CORE__REMOTE_LOG_CONN_ID=aws_default
AIRFLOW_CONN_AWS_DEFAULT=aws://?region_name=us-west-2

update: Also when I switched from gevent to sync worker class, it seemed to read fine again

[alanbrent]

Can confirm that using a threaded gunicorn worker (gevent in our case) breaks the web component's ability to show task logs that are in S3. Moving back to sync works for us, and since we're running on Kubernetes a few more web pods to absorb the ALB/ELB health checks isn't (so far) a performance concern.

Would love to see this fixed, however.

Using Airflow default aws_default for the S3 connection. Relevant configuration shared among web, scheduler, and worker pods below the fold

  - AIRFLOW__CORE__REMOTE_LOGGING=True
  - AIRFLOW__CORE__REMOTE_LOG_CONN_ID=aws_default
  - AIRFLOW__CORE__REMOTE_BASE_LOG_FOLDER=s3://${OUR-BUCKET}/logs
  - AIRFLOW__WEBSERVER__WORKER_CLASS=sync
  - AIRFLOW__WEBSERVER__WORKER_REFRESH_INTERVAL=3600 # the default 30s worker lifetime is *way* too short
  - AWS_DEFAULT_REGION=us-east-1
  - AWS_ROLE_ARN=arn:aws:iam::${OUR_AWS_ACCOUNT_ID}:role/${OUR_AIRFLOW_APP_ROLE}
  - AWS_WEB_IDENTITY_TOKEN_FILE=/var/run/secrets/eks.amazonaws.com/serviceaccount/token # This is how EKS does IAM<>Pod stuff

[cmlad]

The issue seems to be infinite recursion due to an interaction between gevent's monkey patching and the botocore library used by S3TaskHandler:

Traceback (most recent call last):
  File "/usr/local/Caskroom/miniconda/base/envs/airflow/lib/python3.7/site-packages/airflow/utils/log/s3_task_handler.py", line 131, in s3_log_exists
    return self.hook.get_key(remote_log_location) is not None
  File "/usr/local/Caskroom/miniconda/base/envs/airflow/lib/python3.7/site-packages/airflow/hooks/S3_hook.py", line 224, in get_key
    obj = self.get_resource_type('s3').Object(bucket_name, key)
  File "/usr/local/Caskroom/miniconda/base/envs/airflow/lib/python3.7/site-packages/airflow/contrib/hooks/aws_hook.py", line 186, in get_resource_type
    config=config, verify=self.verify)
  File "/usr/local/Caskroom/miniconda/base/envs/airflow/lib/python3.7/site-packages/boto3/session.py", line 389, in resource
    aws_session_token=aws_session_token, config=config)
  File "/usr/local/Caskroom/miniconda/base/envs/airflow/lib/python3.7/site-packages/boto3/session.py", line 263, in client
    aws_session_token=aws_session_token, config=config)
  File "/usr/local/Caskroom/miniconda/base/envs/airflow/lib/python3.7/site-packages/botocore/session.py", line 823, in create_client
    credentials = self.get_credentials()
  File "/usr/local/Caskroom/miniconda/base/envs/airflow/lib/python3.7/site-packages/botocore/session.py", line 428, in get_credentials
    'credential_provider').load_credentials()
  File "/usr/local/Caskroom/miniconda/base/envs/airflow/lib/python3.7/site-packages/botocore/session.py", line 919, in get_component
    self._components[name] = factory()
  File "/usr/local/Caskroom/miniconda/base/envs/airflow/lib/python3.7/site-packages/botocore/session.py", line 149, in _create_credential_resolver
    self, region_name=self._last_client_region_used
  File "/usr/local/Caskroom/miniconda/base/envs/airflow/lib/python3.7/site-packages/botocore/credentials.py", line 70, in create_credential_resolver
    container_provider = ContainerProvider()
  File "/usr/local/Caskroom/miniconda/base/envs/airflow/lib/python3.7/site-packages/botocore/credentials.py", line 1803, in __init__
    fetcher = ContainerMetadataFetcher()
  File "/usr/local/Caskroom/miniconda/base/envs/airflow/lib/python3.7/site-packages/botocore/utils.py", line 1578, in __init__
    timeout=self.TIMEOUT_SECONDS
  File "/usr/local/Caskroom/miniconda/base/envs/airflow/lib/python3.7/site-packages/botocore/httpsession.py", line 180, in __init__
    self._manager = PoolManager(**self._get_pool_manager_kwargs())
  File "/usr/local/Caskroom/miniconda/base/envs/airflow/lib/python3.7/site-packages/botocore/httpsession.py", line 188, in _get_pool_manager_kwargs
    'ssl_context': self._get_ssl_context(),
  File "/usr/local/Caskroom/miniconda/base/envs/airflow/lib/python3.7/site-packages/botocore/httpsession.py", line 197, in _get_ssl_context
    return create_urllib3_context()
  File "/usr/local/Caskroom/miniconda/base/envs/airflow/lib/python3.7/site-packages/botocore/httpsession.py", line 72, in create_urllib3_context
    context.options |= options
  File "/usr/local/Caskroom/miniconda/base/envs/airflow/lib/python3.7/ssl.py", line 507, in options
    super(SSLContext, SSLContext).options.__set__(self, value)
  File "/usr/local/Caskroom/miniconda/base/envs/airflow/lib/python3.7/ssl.py", line 507, in options
    super(SSLContext, SSLContext).options.__set__(self, value)
  File "/usr/local/Caskroom/miniconda/base/envs/airflow/lib/python3.7/ssl.py", line 507, in options
    super(SSLContext, SSLContext).options.__set__(self, value)
  [Previous line repeated 409 more times]
RecursionError: maximum recursion depth exceeded

I still need to dive deeper into this

[dimon222]

@cmlad I will make very bad suggestion, but would it work with regular HTTP endpoint instead (without SSL)? Just something to try out.

[dimon222]

@cmlad yes its the reason for sure. And HTTP vs HTTPS no difference.

I randomly found potential fix in one of previously mentioned issues
#8164 (comment)
This works, prepending monkey patching in dagbag.py. What is invoking dagbag.py so early ? Not sure.

[Siddharthk]

How can I read logs from worker pods? This is important since we need to see logs in real time to see whats happening. S3 logs are available only when task gets completed. I am getting below error currently:

Log file does not exist: /opt/airflow/logs/mydag/mytask/2020-07-21T11:58:55.019748+00:00/2.log
*** Fetching from: http://taskpod-49ccd964791a4740b199:8793/log/mydag/mytask/2020-07-21T11:58:55.019748+00:00/2.log
*** Failed to fetch log file from worker. HTTPConnectionPool(host='taskpod-49ccd964791a4740b199', port=8793): Max retries exceeded with url: /log/mydag/mytask/2020-07-21T11:58:55.019748+00:00/2.log (Caused by NewConnectionError('<urllib3.connection.HTTPConnection object at 0x7f486c9b8be0>: Failed to establish a new connection: [Errno -3] Temporary failure in name resolution',))

[TRReeve]

@Siddharthk If you're using any kind of distributed execution (including kubernetes) you wont be able to read logs from the airflow UI while the executor is running using airflow "out of the box". Airflow pushes the log to remote storage when the task is completed and then the UI reads them from there.

I would guess if you do need to get real time logs from batch tasks while they are running maybe you can set up some sort of worker image with fluentd installed or something similar to scrape stdout and push it somewhere else while airflow is running on it? Either way there'll have to be some sort of external integration

[dimon222]

Solution mentioned above by me

@cmlad yes its the reason for sure. And HTTP vs HTTPS no difference.

I randomly found potential fix in one of previously mentioned issues
#8164 (comment)
This works, prepending monkey patching in dagbag.py. What is invoking dagbag.py so early ? Not sure.

This no longer works in 1.10.12. Perhaps, something gets loaded earlier again and messes up urllib3.
The issue is still present.

[dimon222]

Update - new solution is adding gevent monkey patching to top of config_templates/airflow_local_settings.py. It works.

[peycho]

We're facing the same recursion error when using gevent workers.

RecursionError: maximum recursion depth exceeded while calling a Python object

Airflow version 1.10.9

[toderesa97]

We are experiencing the same issue here. Airflow does sucessfully write logs to S3, but we are getting:

Log file does not exist: /opt/airflow/logs/mydag/mytask/2020-07-21T11:58:55.019748+00:00/2.log
*** Fetching from: http://taskpod-49ccd964791a4740b199:8793/log/mydag/mytask/2020-07-21T11:58:55.019748+00:00/2.log
*** Failed to fetch log file from worker. HTTPConnectionPool(host='taskpod-49ccd964791a4740b199', port=8793): Max retries exceeded with url: /log/mydag/mytask/2020-07-21T11:58:55.019748+00:00/2.log (Caused by NewConnectionError('<urllib3.connection.HTTPConnection object at 0x7f486c9b8be0>: Failed to establish a new connection: [Errno -3] Connection refused',))

I have configured my airflow.cfg as follows:

remote_logging = True
remote_log_conn_id = S3Conn
remote_base_log_folder = s3://dlakeufd-dev-audit/logs/airflow
encrypt_s3_logs = False
logging_config_class = airflow.utils.log.s3_task_handler.S3TaskHandler
task_log_reader = s3.task

But we are getting this exception when the docker image is being provisioned into an ECS container (ie, nothing to do with ECS):

...
File "/usr/local/lib/python3.7/site-packages/airflow/logging_config.py", line 53, in configure_logging
2020-09-03 21:16:15.format(logging_class_path, err)
2020-09-03 21:16:15ImportError: Unable to load custom logging from airflow.utils.log.s3_task_handler.S3TaskHandler due to
...

(There is no further explanation after the due to)

I don't get to understand why after extracting the attribute of the class it is expected to be a dict type, which clearly fails. See this class where the exception is triggered in line 43.

[potiuk]

I only handed gevent as this was the only one I knew about. But once you confirm it works for you for gevent I see that there is a similar issue #15116 and seems we can use similar technique by adding eventlet configuration.

    import eventlet
    eventlet.monkey_patch()

But this should be a separate PR once we get confirmation that the approach works for you. Since you have s3 loigging configured and very interested, I will hold on until I get confirmation.

[potiuk]

Did you try gevent with my fixes?

[potiuk]

The PR is green @aa3pankaj @dimon222 -> would love some confirmation that it solves the problem - if we do it quickly then it will get a chance to get into 2.4.3 (together with corresponding eventlet change) and will solve the long standing problem.

[aa3pankaj]

@potiuk will test this out

[potiuk]

Did you also add --preload flag ? The solution will only work if --preload is enabled (either via env variable as described above) or by applying also #27297 from @ephraimbuddy

[aa3pankaj]

let me try with that also

[aa3pankaj]

export GUNICORN_CMD_ARGS='--preload'

/opt/airflow_venv/venv/lib/python3.7/site-packages/airflow/www/gunicorn_config_gevent.py:22 MonkeyPatchWarning: Monkey-patching ssl after ssl has already been imported may lead to errors, including RecursionError on Python 3.6. It may also silently lead to incorrect behaviour on Python 3.7. Please monkey-patch earlier. See gevent/gevent#1016. Modules that had direct imports (NOT patched): ['botocore.httpsession (/opt/airflow_venv/venv/lib/python3.7/site-packages/botocore/httpsession.py)', 'urllib3.contrib.pyopenssl (/opt/airflow_venv/venv/lib/python3.7/site-packages/urllib3/contrib/pyopenssl.py)'].

@potiuk getting this warning now on webserver bootup.
And, same RecursionError while fetching airflow logs from S3.

[potiuk]

Too bad. I will try other things soon.

[EinavDanielDX]

Any news on this?
We're facing the same problem right now

[potiuk]

For now monkey patching is the only solution that works (and seems it works) so i suggest you do not wait for any other solution and apply this. We might or might not find a better one. But there is a workaround that works, so use it above all.

[potiuk]

Does monkeypatching work for you @EinavDanielDX ?

[dimon222]

Only got to test patches now. I'm on 2.4.3

1. Out of the box still doesn't work
2. --preload doesn't help
3. Monkey patching of default settings.py still works with new minor adjustment - new annotation import have to come before monkey patching, otherwise webserver will fail to start.

I didn't test the new draft PR from @potiuk

[potiuk]

Hey @dimon222 @EinavDanielDX @aa3pankaj - can you please try my (new attempt to) fix from #28283 - it's generally a "brute-force" kinda fix. It requires you to set the _AIRFLOW_PATCH_GEVENT environment variable at your webserver though - independently from configuring gevent in the configuration (precisely BEACAUSE reading configuration might trigger 3rd-party libraries to be imported and run their monkeypatching before.

Not the nicest solution, but if it works, you might have a stable way of gevent patching without doing any weird and unpredictable monkeypatching yourself.

[potiuk]

Anyone managed to test my fix yet :) ?

[PApostol]

Hi @potiuk, we've also had the same issue with S3 remote logs and I've tried out your fix. While the fix succeeds in fetching the logs from S3 on the UI, it seems to break the scheduler upon restart. The error is reproducible every time I try to start the scheduler after the fix is applied (note that webserver works fine). Removing the fix from airflow/__init__.py resolves the error. I'm attaching the traceback below. Using Airflow: 2.3.4, Python 3.7.3, LocalExecutor.

Traceback (most recent call last):
  File "/home/ec2-user/venv/bin/airflow", line 8, in <module>
    sys.exit(main())
  File "/home/ec2-user/venv/lib/python3.7/site-packages/airflow/__main__.py", line 38, in main
    args.func(args)
  File "/home/ec2-user/venv/lib/python3.7/site-packages/airflow/cli/cli_parser.py", line 51, in command
    return func(*args, **kwargs)
  File "/home/ec2-user/venv/lib/python3.7/site-packages/airflow/utils/cli.py", line 99, in wrapper
    return f(*args, **kwargs)
  File "/home/ec2-user/venv/lib/python3.7/site-packages/airflow/cli/commands/scheduler_command.py", line 76, in scheduler
    _run_scheduler_job(args=args)
  File "/home/ec2-user/venv/lib/python3.7/site-packages/airflow/cli/commands/scheduler_command.py", line 46, in _run_scheduler_job
    job.run()
  File "/home/ec2-user/venv/lib/python3.7/site-packages/airflow/jobs/base_job.py", line 244, in run
    self._execute()
  File "/home/ec2-user/venv/lib/python3.7/site-packages/airflow/jobs/scheduler_job.py", line 743, in _execute
    self.executor.start()
  File "/home/ec2-user/venv/lib/python3.7/site-packages/airflow/executors/local_executor.py", line 348, in start
    self.result_queue = self.manager.Queue()
  File "/usr/lib64/python3.7/multiprocessing/managers.py", line 724, in temp
    token, exp = self._create(typeid, *args, **kwds)
  File "/usr/lib64/python3.7/multiprocessing/managers.py", line 607, in _create
    conn = self._Client(self._address, authkey=self._authkey)
  File "/usr/lib64/python3.7/multiprocessing/connection.py", line 498, in Client
    answer_challenge(c, authkey)
  File "/usr/lib64/python3.7/multiprocessing/connection.py", line 742, in answer_challenge
    message = connection.recv_bytes(256)         # reject large message
  File "/usr/lib64/python3.7/multiprocessing/connection.py", line 216, in recv_bytes
    buf = self._recv_bytes(maxlength)
  File "/usr/lib64/python3.7/multiprocessing/connection.py", line 407, in _recv_bytes
    buf = self._recv(4)
  File "/usr/lib64/python3.7/multiprocessing/connection.py", line 379, in _recv
    chunk = read(handle, remaining)
BlockingIOError: [Errno 11] Resource temporarily unavailable

UPDATE:
The above has been resolved by using patch_all(thread=False, socket=False) instead of patch_all(), as explained here.

[dimon222]

Just wanted to post my final note that 945fc072fa903c8b96c70d34749bf98dabf8a654 did indeed solve it, and functions properly in 2.5.1 without any extra voodoo.

[potiuk]

Just wanted to post my final note that 945fc07 did indeed solve it, and functions properly in 2.5.1 without any extra voodoo.

Cool thanks for that confirmation!