Berislav Lopac created AIRFLOW-2222:
---------------------------------------
Summary: GoogleCloudStorageHook.copy fails for large files between
locations
Key: AIRFLOW-2222
URL: https://issues.apache.org/jira/browse/AIRFLOW-2222
Project: Apache Airflow
Issue Type: Bug
Reporter: Berislav Lopac
When copying large files (confirmed for around 3GB) between buckets in
different projects, the operation fails and the Google API returns error
[413—Payload Too
Large|https://cloud.google.com/storage/docs/json_api/v1/status-codes#413_Payload_Too_Large].
The documentation for the error says:
{quote}The Cloud Storage JSON API supports up to 5 TB objects.
This error may, alternatively, arise if copying objects between locations
and/or storage classes can not complete within 30 seconds. In this case, use
the [Rewrite|https://cloud.google.com/storage/docs/json_api/v1/objects/rewrite]
method instead.{quote}
The reason seems to be that the {{GoogleCloudStorageHook.copy}} is using the
API {{copy}} method.
h3. Proposed Solution
There are two potential solutions:
1. Implement {{GoogleCloudStorageHook.rewrite}} method which can be called from
operators and other objects to ensure successful execution. This method is more
flexible but requires changes both in the {{GoogleCloudStorageHook}} class and
any other classes that use it for copying files to ensure that they explicitly
call {{rewrite}} when needed.
bq. 2. Modify {{GoogleCloudStorageHook.copy}} to determine when to use
{{rewrite}} instead of {{copy}} underneath. This requires updating only the
{{GoogleCloudStorageHook}} class, but the logic might not cover all the edge
cases and might be difficult to define.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
