Search code examples
authenticationhdfsdcos

How to secure HDFS on DC/OS without Enterprise


I'm trying to secure HDFS cluster on open source DC/OS but it seems it's not an easy thing. The problem I see in HDFS is the fact that it uses username of current system user so without any form of authentication anyone can just create user with certain username and get superuser permissions on cluster.

So I need any form of authentication. IP auth would be fine(clients with certain IPs can only connect to HDFS) but I couldn't find if there's an option to enable it.

Creating Kerberos for HDFS is not an option because running another service just to run another service to run another service etc. will only give tons of work.

If enabling any form of viable security is impossible, is there any other DC/OS HDFS-like service I can use? I need some HA storage to fetch config files and sometimes jars from Artifact Uris to run services. I also need a place to store parquet files from spark streaming.

Version of DC/OS HDFS is 2.6.x.


Solution

  • Unfortunately it seems that Kerberos is the only real form of authentication in HDFS. Without this, HDFS will trust every user.