Versions Compared


  • This line was added.
  • This line was removed.
  • Formatting was changed.


  • RC0 is out, please vote when you get a chance.
 2Refactoring Interfacespifta, marton, xiaoyu, anu , bharat
  • We discussed the interface renaming and refactoring issue at great depth. 
  • Please see the link to understand the decisions made, we will file JIRAs based on these
  • We can continue discussion in JIRAs, where more finer feedback will be possible.
      • OM Client - The Ozone Manager client – The APIs from RPC client.
      • OM Admin - If there are Admin only commands supported by OM then it will be part of this interface.
      • SCM Client - We really have no users for this API today. But will help other services if they decide to use HDDS.
      • SCM Admin - The Admin API for SCM.
      • SCM Server - This interface will be used by OzoneManager, Recon and see if we can move the SCMSecurity service also into this end point.
      • Datanode Client - The RPC client will use this interface to communicate with Data node
      • Datanode Server - This interface will be used by data node to data node communication like copying blocks and containers for replication.We decided to create the following interfaces.
    • SCM HB protocol - This is a special case, since the HBs and lots of command/info flows through this interface.
    • We have decided to lose the word protocol from the proto files.
    • We will try to have a service (that is a network port corresponding) to these services. This will make it easier for network admins to manage firewall configuration for admin commands.
3URI Refactoring ChangeAnu, pifta, Clay, Xiaoyu

We discussed the proposal from Sanjay in depth. There are few things that Clay mentioned.

He likes the like losing the bucket and volume for the following reasons.

  1. If the system is purely DNS driven,  and not config driven, then creates a dependency that deployments need Dynamic DNS.
  2. His more important point was that this makes write code against FileSystem object much easier. You can create one FileSystem object and write into any bucket. That is a huge  win from the programming perspective.
  3. Xiaoyu asked if we are planning to support the .trash feature with OzoneFS at all? We think it is simpler and easier to have a server side delete policy under Ozone.
  4. pifta wanted to make sure that Hive and other applications can work correctly.
  5. He pointed out that one of the issues is that user will get an error if they try to write a file right under the volume.
  6. Clay pointed out that it might be a "good" feature to have for a file system with Ozone's scale. The fact that you cannot write to a volume should make managing the FS easier.
4Discuss Error handling in RatisClay

Clay wanted to bring in FailSafe like API into Ratis. We discussed how Ozone handles failures when a Ratis pipeline hangs. Ozone just discards those pipelines, and open up other pipelines in the cluster. But we would like to have failsafe like API.

5Ratis Fails with 5 ServersClayClay was trying to run Ratis with 5 server instances, even the basic math server example does not work. He will file a issue and work on root causing the issue.