43
votes

I am using docker-compose to set up a scalable airflow cluster. I based my approach off of this Dockerfile https://hub.docker.com/r/puckel/docker-airflow/

My problem is getting the logs set up to write/read from s3. When a dag has completed I get an error like this

*** Log file isn't local.
*** Fetching here: http://ea43d4d49f35:8793/log/xxxxxxx/2017-06-26T11:00:00
*** Failed to fetch log file from worker.

*** Reading remote logs...
Could not read logs from s3://buckets/xxxxxxx/airflow/logs/xxxxxxx/2017-06-
26T11:00:00

I set up a new section in the airflow.cfg file like this

[MyS3Conn]
aws_access_key_id = xxxxxxx
aws_secret_access_key = xxxxxxx
aws_default_region = xxxxxxx

And then specified the s3 path in the remote logs section in airflow.cfg

remote_base_log_folder = s3://buckets/xxxx/airflow/logs
remote_log_conn_id = MyS3Conn

Did I set this up properly and there is a bug? Is there a recipe for success here that I am missing?

-- Update

I tried exporting in URI and JSON formats and neither seemed to work. I then exported the aws_access_key_id and aws_secret_access_key and then airflow started picking it up. Now I get his error in the worker logs

6/30/2017 6:05:59 PMINFO:root:Using connection to: s3
6/30/2017 6:06:00 PMERROR:root:Could not read logs from s3://buckets/xxxxxx/airflow/logs/xxxxx/2017-06-30T23:45:00
6/30/2017 6:06:00 PMERROR:root:Could not write logs to s3://buckets/xxxxxx/airflow/logs/xxxxx/2017-06-30T23:45:00
6/30/2017 6:06:00 PMLogging into: /usr/local/airflow/logs/xxxxx/2017-06-30T23:45:00

-- Update

I found this link as well https://www.mail-archive.com/[email protected]/msg00462.html

I then shelled into one of my worker machines (separate from the webserver and scheduler) and ran this bit of code in python

import airflow
s3 = airflow.hooks.S3Hook('s3_conn')
s3.load_string('test', airflow.conf.get('core', 'remote_base_log_folder'))

I receive this error.

boto.exception.S3ResponseError: S3ResponseError: 403 Forbidden

I tried exporting several different types of AIRFLOW_CONN_ envs as explained here in the connections section https://airflow.incubator.apache.org/concepts.html and by other answers to this question.

s3://<AWS_ACCESS_KEY_ID>:<AWS_SECRET_ACCESS_KEY>@S3

{"aws_account_id":"<xxxxx>","role_arn":"arn:aws:iam::<xxxx>:role/<xxxxx>"}

{"aws_access_key_id":"<xxxxx>","aws_secret_access_key":"<xxxxx>"}

I have also exported AWS_ACCESS_KEY_ID and AWS_SECRET_ACCESS_KEY with no success.

These credentials are being stored in a database so once I add them in the UI they should be picked up by the workers but they are not able to write/read logs for some reason.

7
At this point I will take any strategy to get logging to work. I can't get them locally, on s3, or using rfsJackStat
Does the folder 'logs' exist at the path? At least the local logs should work without any problems if the folder exists. If they don't work even locally, the only other reason I can think of is incorrect permissions on the airflow folder.Him
We might have something here.github.com/puckel/docker-airflow/pull/100JackStat

7 Answers

29
votes

You need to set up the S3 connection through Airflow UI. For this, you need to go to the Admin -> Connections tab on airflow UI and create a new row for your S3 connection.

An example configuration would be:

Conn Id: my_conn_S3

Conn Type: S3

Extra: {"aws_access_key_id":"your_aws_key_id", "aws_secret_access_key": "your_aws_secret_key"}
40
votes

UPDATE Airflow 1.10 makes logging a lot easier.

For s3 logging, set up the connection hook as per the above answer

and then simply add the following to airflow.cfg

    [core]
    # Airflow can store logs remotely in AWS S3. Users must supply a remote
    # location URL (starting with either 's3://...') and an Airflow connection
    # id that provides access to the storage location.
    remote_base_log_folder = s3://my-bucket/path/to/logs
    remote_log_conn_id = MyS3Conn
    # Use server-side encryption for logs stored in S3
    encrypt_s3_logs = False

For gcs logging,

  1. Install the gcp_api package first, like so: pip install apache-airflow[gcp_api].

  2. Set up the connection hook as per the above answer

  3. Add the following to airflow.cfg

    [core]
    # Airflow can store logs remotely in AWS S3. Users must supply a remote
    # location URL (starting with either 's3://...') and an Airflow connection
    # id that provides access to the storage location.
    remote_logging = True
    remote_base_log_folder = gs://my-bucket/path/to/logs
    remote_log_conn_id = MyGCSConn
    

NOTE: As of Airflow 1.9 remote logging has been significantly altered. If you are using 1.9, read on.

Reference here

Complete Instructions:

  1. Create a directory to store configs and place this so that it can be found in PYTHONPATH. One example is $AIRFLOW_HOME/config

  2. Create empty files called $AIRFLOW_HOME/config/log_config.py and $AIRFLOW_HOME/config/__init__.py

  3. Copy the contents of airflow/config_templates/airflow_local_settings.py into the log_config.py file that was just created in the step above.

  4. Customize the following portions of the template:

    #Add this variable to the top of the file. Note the trailing slash.
    S3_LOG_FOLDER = 's3://<bucket where logs should be persisted>/'
    
    Rename DEFAULT_LOGGING_CONFIG to LOGGING CONFIG
    LOGGING_CONFIG = ...
    
    Add a S3TaskHandler to the 'handlers' block of the LOGGING_CONFIG variable
    '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': S3_LOG_FOLDER,
        'filename_template': FILENAME_TEMPLATE,
    },
    
     Update the airflow.task and airflow.task_runner blocks to be 's3.task' instead >of 'file.task'.
    'loggers': {
        'airflow.task': {
            'handlers': ['s3.task'],
            ...
        },
        'airflow.task_runner': {
            'handlers': ['s3.task'],
            ...
        },
        'airflow': {
            'handlers': ['console'],
            ...
        },
    }
    
  5. Make sure a s3 connection hook has been defined in Airflow, as per the above answer. The hook should have read and write access to the s3 bucket defined above in S3_LOG_FOLDER.

  6. Update $AIRFLOW_HOME/airflow.cfg to contain:

    task_log_reader = s3.task
    logging_config_class = log_config.LOGGING_CONFIG
    remote_log_conn_id = <name of the s3 platform hook>
    
  7. Restart the Airflow webserver and scheduler, and trigger (or wait for) a new task execution.

  8. Verify that logs are showing up for newly executed tasks in the bucket you’ve defined.

  9. Verify that the s3 storage viewer is working in the UI. Pull up a newly executed task, and verify that you see something like:

    *** Reading remote log from gs://<bucket where logs should be persisted>/example_bash_operator/run_this_last/2017-10-03T00:00:00/16.log.
    [2017-10-03 21:57:50,056] {cli.py:377} INFO - Running on host chrisr-00532
    [2017-10-03 21:57:50,093] {base_task_runner.py:115} INFO - Running: ['bash', '-c', u'airflow run example_bash_operator run_this_last 2017-10-03T00:00:00 --job_id 47 --raw -sd DAGS_FOLDER/example_dags/example_bash_operator.py']
    [2017-10-03 21:57:51,264] {base_task_runner.py:98} INFO - Subtask: [2017-10-03 21:57:51,263] {__init__.py:45} INFO - Using executor SequentialExecutor
    [2017-10-03 21:57:51,306] {base_task_runner.py:98} INFO - Subtask: [2017-10-03 21:57:51,306] {models.py:186} INFO - Filling up the DagBag from /airflow/dags/example_dags/example_bash_operator.py
    
13
votes

(Updated as of Airflow 1.10.2)

Here's a solution if you don't use the admin UI.

My Airflow doesn't run on a persistent server ... (It gets launched afresh every day in a Docker container, on Heroku.) I know I'm missing out on a lot of great features, but in my minimal setup, I never touch the admin UI or the cfg file. Instead, I have to set Airflow-specific environment variables in a bash script, which overrides the .cfg file.

apache-airflow[s3]

First of all, you need the s3 subpackage installed to write your Airflow logs to S3. (boto3 works fine for the Python jobs within your DAGs, but the S3Hook depends on the s3 subpackage.)

One more side note: conda install doesn't handle this yet, so I have to do pip install apache-airflow[s3].

Environment variables

In a bash script, I set these core variables. Starting from these instructions but using the naming convention AIRFLOW__{SECTION}__{KEY} for environment variables, I do:

export AIRFLOW__CORE__REMOTE_LOGGING=True
export AIRFLOW__CORE__REMOTE_BASE_LOG_FOLDER=s3://bucket/key
export AIRFLOW__CORE__REMOTE_LOG_CONN_ID=s3_uri
export AIRFLOW__CORE__ENCRYPT_S3_LOGS=False

S3 connection ID

The s3_uri above is a connection ID that I made up. In Airflow, it corresponds to another environment variable, AIRFLOW_CONN_S3_URI. The value of that is your S3 path, which has to be in URI form. That's

s3://access_key:secret_key@bucket/key

Store this however you handle other sensitive environment variables.

With this configuration, Airflow will be able to write your logs to S3. They will follow the path of s3://bucket/key/dag/task_id/timestamp/1.log.


Appendix on upgrading from Airflow 1.8 to Airflow 1.10

I recently upgraded my production pipeline from Airflow 1.8 to 1.9, and then 1.10. Good news is that the changes are pretty tiny; the rest of the work was just figuring out nuances with the package installations (unrelated to the original question about S3 logs).

(1) First of all, I needed to upgrade to Python 3.6 with Airflow 1.9.

(2) The package name changed from airflow to apache-airflow with 1.9. You also might run into this in your pip install.

(3) The package psutil has to be in a specific version range for Airflow. You might encounter this when you're doing pip install apache-airflow.

(4) python3-dev headers are needed with Airflow 1.9+.

(5) Here are the substantive changes: export AIRFLOW__CORE__REMOTE_LOGGING=True is now required. And

(6) The logs have a slightly different path in S3, which I updated in the answer: s3://bucket/key/dag/task_id/timestamp/1.log.

But that's it! The logs did not work in 1.9, so I recommend just going straight to 1.10, now that it's available.

2
votes

To complete Arne's answer with the recent Airflow updates, you do not need to set task_log_reader to another value than the default one : task

As if you follow the default logging template airflow/config_templates/airflow_local_settings.py you can see since this commit (note the handler's name changed to's3': {'task'... instead of s3.task) that's the value on the remote folder(REMOTE_BASE_LOG_FOLDER) will replace the handler with the right one:

REMOTE_LOGGING = conf.get('core', 'remote_logging')

if REMOTE_LOGGING and REMOTE_BASE_LOG_FOLDER.startswith('s3://'):
        DEFAULT_LOGGING_CONFIG['handlers'].update(REMOTE_HANDLERS['s3'])
elif REMOTE_LOGGING and REMOTE_BASE_LOG_FOLDER.startswith('gs://'):
        DEFAULT_LOGGING_CONFIG['handlers'].update(REMOTE_HANDLERS['gcs'])
elif REMOTE_LOGGING and REMOTE_BASE_LOG_FOLDER.startswith('wasb'):
        DEFAULT_LOGGING_CONFIG['handlers'].update(REMOTE_HANDLERS['wasb'])
elif REMOTE_LOGGING and ELASTICSEARCH_HOST:
        DEFAULT_LOGGING_CONFIG['handlers'].update(REMOTE_HANDLERS['elasticsearch'])

More details on how to log to/read from S3 : https://github.com/apache/incubator-airflow/blob/master/docs/howto/write-logs.rst#writing-logs-to-amazon-s3

1
votes

Just a side note to anyone following the very useful instructions in the above answer: If you stumble upon this issue: "ModuleNotFoundError: No module named 'airflow.utils.log.logging_mixin.RedirectStdHandler'" as referenced here (which happens when using airflow 1.9), the fix is simple - use rather this base template: https://github.com/apache/incubator-airflow/blob/v1-9-stable/airflow/config_templates/airflow_local_settings.py (and follow all other instructions in the above answer)

The current template incubator-airflow/airflow/config_templates/airflow_local_settings.py present in master branch contains a reference to the class "airflow.utils.log.s3_task_handler.S3TaskHandler", which is not present in apache-airflow==1.9.0 python package. Hope this helps!

1
votes

Have it working with Airflow 1.10 in kube. I have the following env var sets:

AIRFLOW_CONN_LOGS_S3=s3://id:secret_uri_encoded@S3
AIRFLOW__CORE__REMOTE_LOGGING=True
AIRFLOW__CORE__REMOTE_BASE_LOG_FOLDER=s3://xxxx/logs
AIRFLOW__CORE__REMOTE_LOG_CONN_ID=logs_s3
1
votes

Phew! Motivation to keep nipping the airflow bugs in the bud is to confront this as a bunch of python files XD here's my experience on this with apache-airflow==1.9.0.

First of all, there's simply no need trying airflow connections .......... --conn_extra etc, etc.

Just set your airflow.cfg as:

remote_logging = True
remote_base_log_folder = s3://dev-s3-main-ew2-dmg-immutable-potns/logs/airflow-logs/
encrypt_s3_logs = False

# Logging level
logging_level = INFO
fab_logging_level = WARN

# Logging class
# Specify the class that will specify the logging configuration
# This class has to be on the python classpath
# logging_config_class = my.path.default_local_settings.LOGGING_CONFIG
logging_config_class = log_config.LOGGING_CONFIG
remote_log_conn_id = s3://<ACCESS-KEY>:<SECRET-ID>@<MY-S3-BUCKET>/<MY>/<SUB>/<FOLDER>/

keep the $AIRFLOW_HOME/config/__ init __.py and $AIRFLOW_HOME/config/log_config.py file as above.

The problem with me as a missing "boto3" package, which I could get to by:

vi /usr/lib/python3.6/site-packages/airflow/utils/log/s3_task_handler.py
then >> import traceback

and in the line containing:

Could not create an S3Hook with connection id "%s". ' 'Please make sure that airflow[s3] is installed and ' 'the S3 connection exists.

doing a traceback.print_exc() and well it started cribbing about missing boto3!

Installed it and Life was beautiful back again!