elek commented on pull request #2331:
URL: https://github.com/apache/ozone/pull/2331#issuecomment-862142935


   Thanks for the question/suggestion @arp7 and @umamaheswararao. I also 
considered throwing an exception instead of printing out a warning, but 
finally, I decided to put only a warning here. Let me share my view here:
   
    1. First of all it's a client-side change. Any serious validation should be 
placed on the server/OM side. (from this point of view even warning can be 
removed)
   
    2. It's hard to predict all the possible use cases. I agree that we should 
introduce a configuration option to make it possible for vendors to limit the 
available ReplicationConfig/ECReplicationConfig set. But the Apache Ozone code 
should be as much flexible as possible (IMHO). Example: earlier we hard-coded 
support of `ONE`/`THREE` replication scheme for closed containers, but it 
turned out that there is a demand to use other values (like `TWO`)
   
   3. There or other meaningful combinations: for example if dozens of nodes 
are available and the key sizes are bigger, it may be reasonable to use `6-4` 
instead of `3-2`. The write amplification is the same, but `6-4` can have 
better durability (in exchange for more padding + memory usage).
   
   4. I can imagine even 10-10 as a useful option with very specific/smart 
placement policies (one may require more parity to place them to other racks). 
But I admit it would be a very rare case.
    
   My suggestion:
   
   Remove the warning from here and put a server-side validation based on the 
configuration value. I am fine to restrict it to 3-2,6-3,10-4 by default, but 
power users can modify the defaults if they require it.
   
   Will create this follow-up jira if no objection.


-- 
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.

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