ivandika3 commented on PR #4922:
URL: https://github.com/apache/ozone/pull/4922#issuecomment-1608905042

   From the code comment regarding Container Owners in `ContainerStateMap`
   ```
    * 2. Owners - Each instance of Name service, for example, Namenode of HDFS 
or
    * Ozone Manager (OM) of Ozone or CBlockServer --  is an owner. It is
    * possible to have many OMs for a Ozone cluster and only one SCM. But SCM
    * keeps the data from each OM in separate bucket, never mixing them. To
    * write data, often we have to find all open containers for a specific 
owner.
   ```
   
   It seems the intent is to have one owner per Name service, which is 
comparable to OM Service ID, instead of OM Node ID. We can change it in OM 
Service ID instead. In our cluster, we finally decided to use OM Service ID 
instead of uuid OM ID for a few reasons: 
   
   - For each container, SCM creates a minimum number of containers first for 
each allocateBlock call (i.e. 
ContainerManagerImpl#getOpenContainerCountPerPipeline). If we use OM ID, when 
the OM leader change, it will create these initial containers as well. This 
will increase the number of containers, which might affect SCM performance. 
Instead, if we use Service ID, we only create the initial containers once.
   
   - Using OM ID (uuid) will make the OM container names very confusing since 
from the UUID it's hard to know which OM cluster the container comes from. 
Using Service ID is easy since we know which service comes from (e.g. service1, 
service2 etc)
   
   - As long as we ensure that Service IDs of OM services using the same SCM is 
unique, it should be safe to share SCM. Duplicates of Service ID is harder to 
do since for Ozone client configuration, Service IDs are supposed to be unique.
   
   


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]


---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to