Hi All, Why is the flink-s3-fs-presto recommended ...
# troubleshooting
c
Hi All, Why is the flink-s3-fs-presto recommended over the hadoop plugin while working with the checkpoint files in S3? I am running into 403 access issues (TM pods can not read the chk file) after the scaling events (flink-autoscaler functionality with the flink-k8-operator 1.8) with the presto plugin but if I switch to the hadoop plugin I don't. So I am thinking about switching to the hadoop plugin. Note: Not all, only some TM pods run into 403.
b
I don't know why is presto recommended. Though, try
Copy code
presto.s3.use-instance-credentials: false
in the flink conf, if this helps with the 403 issue. Setting this key to false, allows presto to use most generic AWS credentials provider.
c
@Barisa Obradovic Thanks for the response. I have tried
Copy code
presto.s3.use-instance-credentials: false
in the flink-config but ran into the same 403 issue.
b
odd, if you have flink 1.14 or later; which has needed presto code; I'd expect it to work. You can try enabling debug logs for AWS packages to see what's wrong
πŸ‘ 1