Ozone doesn't work with Kerberos and Spark #4996
Replies: 3 comments 8 replies
-
The setup that doesn't work is the following: Apache Ozone 1.3.0 + Spark 3.2.3 (cluster mode) + Kerberos. All resides in Kubernetes 1.21.6. Here you are Spark driver logs: |
Beta Was this translation helpful? Give feedback.
-
If kerberos authentication in spark is not mandatory, spark can access ozone using s3g which is s3 compatible. For me, after making ozone secure with kerberos, I created access and secret key from s3g.
|
Beta Was this translation helpful? Give feedback.
-
The most relevant log is these two lines 23/06/12 04:35:13 DEBUG OzoneDelegationTokenSelector: Got tokens: null for service 10.246.0.181:9862 Kerberos is working fine. But Spark didn't request a delegation token. I don't see a ozone delegation token returned. Normally this happens when a file system is not in the default hadoop fs (fs.defaultFS). For a default FS, the MapReduce framework invokes FileSystem.addDelegationTokens() API to request its delegation token. Spark doesn't do this itself. Your code may need to call FileSystem.addDelegationTokens() explicitly to make sure ozone dt is returned. |
Beta Was this translation helpful? Give feedback.
-
We planned to go to production with Ozone , but the setup (Ozone 1.3.0 + Kerberos + Spark cluster + Kubernetes) does not work.
We can't go to the enterprise without Kerberos, regardless of having Ozone worked without Kerberos.
Should we go back to HDFS or is there some kind of solution?
Beta Was this translation helpful? Give feedback.
All reactions