Micheal Ascah created AIRFLOW-2771:
--------------------------------------
Summary: S3Hook Broad Exception Silent Failure
Key: AIRFLOW-2771
URL: https://issues.apache.org/jira/browse/AIRFLOW-2771
Project: Apache Airflow
Issue Type: Bug
Components: hooks
Affects Versions: Airflow 2.0, 1.9.0
Reporter: Micheal Ascah
Assignee: Micheal Ascah
h2. Scenario
S3KeySensor is passed an invalid S3/AWS connection id name (doesn't exist or
bad permissions). When poking for the key, it creates an S3Hook and calls
`check_for_key` on the hook. Currently, the call is caught by a generic except
clause that catches all exceptions, rather than the expected
botocore.exceptions.ClientError when an object is not found.
h2. Problem
This causes the sensor to return False and report no issue with the task
instance until it times out, rather than intuitively failing immediately if the
connection is incorrectly configured. The current logging output gives no
insight as to why the key is not being found.
h4. Current code
{code:python}
try:
self.get_conn().head_object(Bucket=bucket_name, Key=key)
return True
except: # <- This catches credential and connection exceptions that should be
raised
return False
{code}
{code:python}
from airflow.hooks.S3_hook import S3Hook
hook = S3Hook(aws_conn_id="conn_that_doesnt_exist")
hook.check_for_key(key="test", bucket="test")
False
{code}
h4. Expected
h5. No credentials
{code:python}
from airflow.hooks.S3_hook import S3Hook
hook = S3Hook(aws_conn_id="conn_that_doesnt_exist")
hook.check_for_key(key="test", bucket="test")
Traceback (most recent call last):
...
botocore.exceptions.NoCredentialsError: Unable to locate credentials
{code}
h5. Good credentials
{code:python}
from airflow.hooks.S3_hook import S3Hook
hook = S3Hook(aws_conn_id="conn_that_does_exist")
hook.check_for_key(key="test", bucket="test")
False
{code}
h4. Proposed Change
Add a type to the except clause for botocore.exceptions.ClientError and log the
message.
{code:python}
try:
self.get_conn().head_object(Bucket=bucket_name, Key=key)
return True
except ClientError as e:
self.log.info(e.response["Error"]["Message"])
return False
{code}
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
