yuqian90 opened a new issue #21017:
URL: https://github.com/apache/airflow/issues/21017
### Apache Airflow version
2.2.3 (latest released)
### What happened
`TaskInstance.max_tries` is a cumulative value that's stored in the
database. Whenever a failed `TaskInstance` is cleared, its `max_tries` is
incremented by `task.retries - 1`. However, the increased `max_tries` value of
a `TaskInstance` is reset to the original value (`task.retries`) whenever a
user opens the `Task Instance Details` page in the web UI. This is because
`www/views.py` queries `TaskInstance` and then calls `refresh_from_task()` on
it in a few places, which changes the attributes of the `TaskInstance`. The
`session` used in `www/views.py` is then auto committed and the updated values
such as `max_tries` are saved to the database.
This looks like a bug. So far the only effect I noticed is `max_tries` gets
reset causing sensors in "reschedule" mode to timeout prematurely. It happens
because of the following lines in `sensors/base.py`. When `max_tries` is reset,
this code always gets the `first_try_number` 1 which may have happened a long
time ago and thinks the sensor times out.
```
if self.reschedule:
# If reschedule, use the start date of the first try (first try
can be either the very
# first execution of the task, or the first execution after the
task was cleared.)
first_try_number = context['ti'].max_tries - self.retries + 1
task_reschedules = TaskReschedule.find_for_task_instance(
context['ti'], try_number=first_try_number
)
```
### What you expected to happen
Opening "Task Instance Details" page should not make any changes to the
state of the TaskInstance in the database.
### How to reproduce
Create a DAG like this:
```
from datetime import datetime
from airflow import DAG
from airflow.sensors.python import PythonSensor
with DAG(
dag_id="reschedule_clear",
start_date=datetime(2021, 1, 1),
catchup=False,
schedule_interval="@daily",
) as dag:
task_1 = PythonSensor(
task_id="task_1",
python_callable=lambda: False,
mode="reschedule",
timeout=90,
retries=5,
)
```
```
State of the TaskInstance after it failed for the first time:
airflow=# select max_tries, try_number from task_instance ;
max_tries | try_number
-----------+------------
5 | 1
(1 row)
After the failed TaskInstance was cleared:
airflow=# select max_tries, try_number from task_instance ;
max_tries | try_number
-----------+------------
6 | 1
(1 row)
After the TaskInstance failed for a second time:
airflow=# select max_tries, try_number from task_instance ;
max_tries | try_number
-----------+------------
6 | 2
(1 row)
After the TaskInstance is cleared a second time:
airflow=# select max_tries, try_number from task_instance ;
max_tries | try_number
-----------+------------
7 | 2
(1 row)
After user opens the "Task Instance Details" page in Web UI:
airflow=# select max_tries, try_number from task_instance ;
max_tries | try_number
-----------+------------
5 | 2
(1 row)
```
### Operating System
Any
### Versions of Apache Airflow Providers
Any
### Deployment
Virtualenv installation
### Deployment details
Any
### Anything else
Every time a user opens "Task Instance Details" page.
### Are you willing to submit PR?
- [X] Yes I am willing to submit a PR!
### Code of Conduct
- [X] I agree to follow this project's [Code of
Conduct](https://github.com/apache/airflow/blob/main/CODE_OF_CONDUCT.md)
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
