How to choose how often Apache Airflow scheduler updates a DAG? - airflow

As stated in Apache Airflow documentation, I can control how often a DAG is updated by setting configuration variable min_file_process_interval in your airflow.cfg file:
min_file_process_interval
Number of seconds after which a DAG file is parsed. The DAG file is parsed every min_file_process_interval number of seconds. Updates to DAGs are reflected after this interval. Keeping this number low will increase CPU usage.
However, I didn't find any clue or best practice about which value should I set for min_file_process_interval.
Example
My DAG changes once a day. By default min_file_process_interval is set to 30 seconds. It means most of the time updating DAG is useless: as long as DAG doesn't change, updated DAG and previous DAG are the same. It consumes resources and generates logs. But if I update DAG only once a day, do I risk to run wrong DAG if DAG changes after the daily DAG update or the DAG is also updated just before run ?
What value for min_file_process_interval should I set in this case ?
EDIT: As stated in Elad's answer responding to a previous version of this question, dynamic DAGs should be avoided. However, If I have dynamic DAGs, how to choose min_file_process_interval?

You are mixing two different things.
min_file_process_interval means how often Airflow scan the .py files and update the DAG within Airflow. Consider that when you deploy new .py file Airflow needs to read it and create it in the metastore database - so the setting is about how often it happens.
For your use case the DAG code should not be updated every day - In fact it should not be updated at all. It should just run everyday. Your dag just need to be able to handle the correct file per date. You code can be something like:
from airflow.providers.ftp.sensors.ftp import FTPSensor
with DAG(dag_id='stackoverflow',
default_args=default_args,
schedule_interval="#daily",
catchup=False
) as dag:
# Waits for a file or directory to be present on FTP.
sensor_op = FTPSensor(
task_id='sensor_task',
path='/my_folder/{{ ds }}/file.csv', #path to your file in the server
fail_on_transient_errors=False,
ftp_conn_id='ftp_default'
)
# Operator to process the file
operator_op = SomeOperator()
sensor_op >> operator_op
In that DAG it will start a run everyday - the first operator is sensor thus if the file for that day isn't present the workflow will wait until it appears only once it appear the workflow will continue to the 2nd operator which should process the file.
Note that the path parameter of FTPSensor is templated. This means you can use macros like {{ ds }} this will give you a path that contains each day date like:
/my_folder/2021-05-01/file.csv
/my_folder/2021-05-02/file.csv
/my_folder/2021-05-03/file.csv
You can also do path='/my_folder/{{ ds }}.csv' which will give:
/my_folder/2021-05-01.csv
/my_folder/2021-05-02.csv
/my_folder/2021-05-03.csv

Related

Is it a good practice to use airflow metadatabase to control pipelines?

Recently I'm developing an airflow pipeline that will be running for multi tenants. This DAG will be triggered via API, and separated by batches, which is controlled by a metadabase in SQL following some business rules.
Each batch has a batch_id in order to controll the batches, and it is passed to conf DAG via API. The batch id has the timestamp of creation combined with tenant and filetype. Example: tenant1_20221120123323 ... tenant2_20221120123323. These batches can contain two filetypes ( for example purpouses) and for each filetype a DAG is triggered (DAG1 for filetype 1 and DAG2 for filetype 2) and then from the file perspective, it is combined with the filetype in some stages tenant1_20221120123323_filetype1, tenant1_20221120123323_filetype2 ...
For illustrate this, imagine that the first dag the following pipeline process_data_on_spark >> check_new_files_on_statingstorage >> [filetype2_exists, write_new_data_to_warehouse] filetype2_exists >> read_data_from_filetype2 >> merge_filetype2_filetype2 >> write_new_data_to_warehouse . Where the filetype2_exists is a BranchPythonOperator, that verify if DAG_2 was triggered, and if it was, it will merge the resulted data form DAG2 with the DAG1 before execute write_new_data_to_warehouse.
Based on this DAG model, there will be one DAG run for each tenant. So, the DAG can have multiple DAG runs running in parallel if we trigger more than one DAG run (one per tenant). Here is my first question:
Is a good practice work with multiple DAG runs in the same DAG instead of working with Dynamic DAGs ? In this case, I would end withprocess_data_on_spark _tenant1,
process_data_on_spark _tenant2, ...process_data_on_spark _tenantN. It worth mention that the number of tenants can reach hundreads.
Now, considering that the filetype2 can or not be present in the batch, and, considering that I would use the model mentioned above (on single DAG with multiples DAG run runnning in parallel - one for each tenant). The only idead that I have for check if DAG2 was triggered for the current batch (ie., filetype2 was present in the batch) was modify the DAG_run_id to include the batch_id, combined with the filetype:
The default dag_run_id: manual__2022-11-19T00:00:00+00:00
The new dag_run_id: manual__tenant1_20221120123323_filetype2__2022-11-19T00:00:00+00:00
And from then, I would be able to query the airflow metadatabse and check if there was an dag_run_id that contains the current batch_id and the filetype2 running, and, with a sensor, wait for the dag_status be success. Then, I could run the read_data_from_filetype2 task. Otherwise, if there is no dag_run_id with batch_id and filetype2 registed in airflow metadatabase, I can follow the write_new_data_to_warehouse directly.
Here's the other question:
Is a good practice to modify dag_run_id and use it combined with airflow metadatabase to control pipelines?
Considering this scenario, It would be better to create dynamic DAGs, even if there would be result in hundeads DAGs or working with dag_run_id and airflow_metadabase and keep parallel DAG runs in one single DAG?
Or, there would be a better approach for this problem?
Thank You.

Airflow scheduler wont queue some tasks that have been cleared in the UI

I have a bunch of DAG runs in the ui. If I clear some task across all the DAG runs, some of the tasks are correctly triggered, whereas others are stuck with a cleared state.
At the moment I am simply using airflow CLI to backfill these tasks. This works, but it is unfortunate that I need an unbroken CLI session to complete a clearing/reprocessing scenario.
The reason for this is the naming (and thus type) of your DAG runs.
If you go into your airflow meta data db and open table "dag_runs", you will see run_id. The scheduler identifies the runs it creates with "scheduled__" followed by a datetime. If you backfill the run_id will be named "backfil_" followed by a datetime.
The scheduler will only check and queue tasks for run_ids that starts with "scheduled__", denoting a DagRunType of "scheduled".
If you rename the run_id from backfill_ to scheduled__, the scheduler will identify the dag runs and schedule the cleared task underneath.
This SQL query will change bacfill_ to schelduled__:
UPDATE dag_run
SET run_id = Replace(run_id, 'backfill_', 'scheduled__')
where id in (
select id from dag_run where ("run_id"::TEXT LIKE '%backfill_%'));
-- note that backfill_ is a single underscore, and scheduled__ is two.
-- This is not a mistake in my case. But please review the values in your tabel.

Airflow DAG Multiple Runs

I have a DAG that I want to run multiple times after each successful completion. For an example I want to run it 10 times and stop. Is there a way to accomplish this? I tried looking into scheduling with CRON but it doesn't seem clean nor triggering the DAG via UI multiple times doesn't work (runs in parallel).
I found a solution to my use case. It incorporated using depends_on_past=True (mentioned by #Hitesh Gupta) and setting your airflow.cfg file below:
# The maximum number of active DAG runs per DAG
max_active_runs_per_dag = 1
This allowed us to only have one active DAG run at a time and also to not continue the next DAG run if there were failure in the previous run. This is for Airflow version 1.10.1 that I tested on.
You can, in addition to supplying a start_date, provide your DAG an end_date
Quoting the docstring
:param start_date: The timestamp from which the scheduler will attempt
to backfill
:type start_date: datetime.datetime
:param end_date: A date beyond which your DAG won't run, leave to None for open ended scheduling
:type end_date: datetime.datetime
While unrelated, also have a look at following scheduler settings in airflow.cfg as mentioned in this article
run_duration
num_runs
UPDATE-1
In his article Use apache airflow to run task exactly once, #Andreas P has described a clever technique, which I believe can be adapted to your use-case. While even that won't be a very-tidy solution, it would at-least allow you to specify beforehand the number of runs (integer) for DAG instead of end_date.
Alternatively (assuming you implement the above approach) rather than rather than baking this skipping-dag-after max-runs functionality within each DAG, you can create a separate orchestrator DAG that disables a given DAG after its max runs have passed.
You have to set property depends_on_past. This is set under DAG's default arguments section and it refers to previous instance dag instance. This is fix your problem.

Apache Airflow tasks are stuck in a 'up_for_retry' state

I've been setting up an airflow cluster on our system and previously it has been working. I'm not sure what I may have done to change this.
I have a DAG which I want to run on a schedule. To make sure it's working I'd also like to trigger it manually. Neither of these seem to be working at the moment and no logs seem to be being written for the task instances. The only logs available are the airflow scheduler logs which generally look healthy.
I am just constantly met with this message:
Task is not ready for retry yet but will be retried automatically. Current date is 2018-12-12T11:34:46.978355+00:00 and task will be retried at 2018-12-12T11:35:08.093313+00:00.
However, if I wait a little the exact same message is presented again except the times have moved forward a little. Therefore, it seems the task is never actually being retried.
I am using a LocalExecutor and the task is an SSHOperator. Simplified code below. All it does is ssh onto another machine and start a bunch of application with a pre-determined directory structure.:
DB_INFO_FILE = 'info.json'
START_SCRIPT = '/bin/start.sh'
TIME_IN_PAST = timezone.convert_to_utc(datetime.today() -
timedelta(days=1))
DEFAULT_ARGS = {
'owner': 'airflow',
'depends_on_past': False,
'start_date': TIME_IN_PAST,
'email': ['some_email#blah.com'],
'email_on_failure': False,
'email_on_retry': False,
'retries': 1,
'retry_delay': timedelta(minutes=1),
}
def _extract_instance_id(instance_string):
return re.findall(r'\d+', instance_string)[0]
def _read_file_as_json(file_name):
with open(file_name) as open_file:
return json.load(open_file)
DB_INFO = _read_file_as_json(DB_INFO_FILE)
CONFIG_CLIENT = ConfigDbClient(**DB_INFO)
APP_DIRS = CONFIG_CLIENT.get_values('%my-app-info%')
INSTANCE_START_SCRIPT_PATHS = {
_extract_instance_id(instance_string): directory+START_SCRIPT
for instance_string, directory in APP_DIRS.items()
}
# Create an ssh hook which refers to pre-existing connection information
# setup and stored by airflow
SSH_HOOK = SSHHook(ssh_conn_id='my-conn-id')
# Create a DAG object to add tasks to
DAG = DAG('my-dag-id',
default_args=DEFAULT_ARGS)
# Create a task for each app instance.
for instance_id, start_script in INSTANCE_START_SCRIPT_PATHS.items():
task = SSHOperator(
task_id='run-script-{0}'.format(instance_id),
command='bash {0}'.format(start_script),
ssh_hook=SSH_HOOK,
dag=DAG)
It works when I run the tasks individually, via the command line but not via the UI. It seems I can run tasks but I simply cannot trigger a DAG to run. I've tried many combinations of start_date s and interval schedules just to sanity check also.
Any ideas?
And yes, I am aware this question has been asked before and I have looked at all of them but not of the solutions have helped me.
Oh. Your start_date is changing at the same rate or faster than the schedule interval period.
Here's what the scheduler is seeing every couple of seconds:
start_date: 2018-12-11T12:12:12.000Z # E.G. IFF now is 2018-12-12T12:12:12.000Z, a day ago
schedule_interval: timedelta(days=1) # The default
Here's what the scheduler needs for a DAG to run: The last time a run occurred was more than one schedule interval ago. If no scheduled run has occurred, the first scheduled run could start now if one full schedule interval has passed since the start_date as that is the earliest allowable date for execution_date. In which case the dag_run with the execution_date set to the start of that interval period should be created. Then task_instances can be created for any tasks in the DAG whose dependencies are met as long as the task_instance execution_date is after the start_date of the DAG (this is not stored on the dag_run object but recomputed by loading the DAG file just while inspecting the dag's state).
So it's not getting scheduled automatically for the reason that the start date keeps changing just as the interval is satisfied. However if it were -2d at least one run would get scheduled and then any further runs would have to wait until it's 1d after that to be scheduled. It's easier though if you just set a fixed datetime on your start_date.
But what about those odd retries on your manual runs…
You did start a manual run or two. These runs took the current time as the execution_date unless you specified something else. This should be after the start_date, at least until tomorrow, which should clear them to run. But then it seems in your logs you're seeing that they're failing and being marked for retry and also not decrementing your retries. I'm not sure why that would be but could it be that something isn't right with the SSHOperator.
Did you install airflow with the [ssh] extra so that SSHOperator's dependencies are met (specifically paramiko and sshtunnel) on both the webserver and scheduler? One of them is working because I assume it's parsing and showing up in the UI based on being added to the DB.
What do you get if you execute:
airflow test my-dag-id run-script-an_instance_id 2018-12-12T12:12:12
You know that the scheduler and webserver are looping over refilling their DAG bag and so rerunning this DAG file a few 1000 times a day, reloading that json (it's local access, so similar to importing a module), and recreating that SSHHook with a DB lookup. I don't see anything fancy done setting up this hook, why not just remove ssh_hook from the SSHOperator and replace it with ssh_conn_id='my-conn-id' so it can be created once at execution time?
I doubt that's the issue that's causing the retries that just roll forward though.
I had a task stuck in up_for_retry for almost 24 hours before I noticed it, and it had nothing to do with the start_date, end_date, or any other classic beginner's problem.
I resorted to reading the source code, and found that Airflow treats up_for_retry tasks differently if they are part of a backfill DAG run.
So I connected to the metadata DB and changed backfill_ to scheduled__ in the dag_run row corresponding to the stuck task. Airflow started running the stuck task immediately.

Determining if a DAG is executing

I am using Airflow 1.9.0 with a custom SFTPOperator. I have code in my DAGs that poll an SFTP site to find new files. If any are found, then I create custom task id's for the dynamically created task and retrieve/delete the files.
directory_list = sftp_handler('sftp-site', None, '/', None, SFTPToS3Operation.LIST)
for file_path in directory_list:
... SFTP code that GET's the remote files
That part works fine. It seems both the airflow webserver and airflow scheduler are iterating through all the DAGs once a second and actually running the code that retrieves the directory_list. This means I'm hitting the SFTP site ~2 x a second to authenticate and pull a list of files. I'd like to have some conditional code that only executes if the DAG is actually being run.
When an SFTP site uses password authentication, the # of times I connect really isn't an issue. One site requires key authentication and if there are too many authentication failures in a short timespan, the account is locked. During my testing, this seems to happen occasionally for reasons I'm still trying to track down.
However, if I were authenticating only when the DAG was scheduled to execute, or executing manually, this would not be an issue. It also seems wasteful to spend so much time connecting to an SFTP site when it's not scheduled to do so.
I've seen a post that can check to see if a task is executing, but that's not ideal as I'd have to create a long-running task, using up resources I shouldn't require, just to perform that test. Any thoughts on how to accomplish this?
You have a very good use case for Airflow (SFTP to _____ batch jobs), but Airflow is not meant for dynamic DAGs as you are attempting to use them.
Top-Level DAG Code and the Scheduler Loop
As you noticed, any top-level code in a DAG is executed with each scheduler loop. Or put another way, every time the scheduler loop processes the files in your DAG directory it is interpreting all the code in your DAG files. Anything not in a task or operator is interpreted/executed immediately. This puts undue strain on the scheduler as well as any external systems you are making calls to.
Dynamic DAGs and the Airflow UI
Airflow does not handle dynamic DAGs through the UI well. This is mostly the result of the Airflow DAG state not being stored in the database. DAG views and history are rendered based on what exist in the interpreted DAG file at any given moment. I personally hope to see this change in the future with some form of DAG versioning.
In a dynamic DAG you can both add and remove tasks from a DAG.
Adding Tasks Dynamically
When adding tasks for a DAG run will make it appear (in the UI) that all DAG
runs before when that task never ran that task all. The will have a None state
and the DAG run will be set to success or failed depending on the outcome
of the DAG run.
Removing Tasks Dynamically
If your dynamic DAG ever removes tasks you will lose the ability to review history of the DAG. For example, if you run a DAG with task_x in the first 20 DAG runs but remove it after that, it will fail to show up in the UI until it is added back into the DAG.
Idempotency and Airflow
Airflow works best when the DAG runs are idempotent. This means that re-running any DAG Run should have the same affect no matter when you run it or how many times you run it. Dynamic DAGs in Airflow break idempotency by adding and removing tasks to previous DAG runs so that the results of re-running are not the same.
Solution Options
You have at least two options moving forward
1.) Continue to build your SFTP DAG dynamically, but create another DAG that writes the available SFTP files to a local file (if not using distributed executor) or an Airflow Variable (this will result in more reads to the Airflow DB) and build your DAG dynamically from that.
2.) Overload the SFTPOperator to take a list of files so that every file that exist is processed within a single task run. This will make the DAGs idempotent and you will maintain accurate history through the logs.
I apologize for the extended explanation, but you're touching on one of the rough spots of Airflow and I felt it was appropriate to give an overview of the problem at hand.

Resources