Tak-Lon (Stephen) Wu created HADOOP-18310:
---------------------------------------------
Summary: Add option and make 400 bad request retryable
Key: HADOOP-18310
URL: https://issues.apache.org/jira/browse/HADOOP-18310
Project: Hadoop Common
Issue Type: Bug
Components: fs/s3
Affects Versions: 3.3.4
Reporter: Tak-Lon (Stephen) Wu
When one is using a customized credential provider via
fs.s3a.aws.credentials.provider, e.g.
org.apache.hadoop.fs.s3a.TemporaryAWSCredentialsProvider, when the provided
credential by this pluggable provider is expired and return an error code of
400 as bad request exception.
Here, the current S3ARetryPolicy will fail immediately and does not retry on
the S3A level.
Our recent use case in HBase found this use case could lead to a Region Server
got immediate abandoned from this Exception without retry, when the file system
is trying open or S3AInputStream is trying to reopen the file. especially the
S3AInputStream use cases, we cannot find a good way to retry outside of the
file system semantic (because if a ongoing stream is failing currently it's
considered as irreparable state), and thus we come up with this optional flag
for retrying in S3A.
--
This message was sent by Atlassian Jira
(v8.20.7#820007)
---------------------------------------------------------------------
To unsubscribe, e-mail: common-dev-unsubscr...@hadoop.apache.org
For additional commands, e-mail: common-dev-h...@hadoop.apache.org
