[ 
https://issues.apache.org/jira/browse/HDDS-15463?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18145783#comment-18145783
 ] 

Stephen O'Donnell commented on HDDS-15463:
------------------------------------------

>  serve multiple containers and many clients concurrently, similar to RATIS.

[~ssa] 

If these pipelines are intended to live forever as the Ratis ones do (until a 
failure on one of the DNs forces it to close down) I am -1 on any design that 
does not include closing down the pipelines over a relatively short time and 
adjusting for the constantly changing space usage on the cluster. That design 
needs to take into account pipeline limits and how pipelines can be closed 
while allowing new ones to be created that don't just use the same 3 nodes 
where the pipeline was just closed. Ie you cannot close one or two and open 1 
or 2, as you will end up using the same nodes as they are the only ones with 
free capacity.

This is one of the reasons why we designed EC to be the way it is - one 
pipeline serves a container. When that container is full, we create a new 
pipeline for the new container. There is no Ratis or voting overhead so 
creating a "pipeline" is nothing more than a set of nodes associated with a 
container while it is open. This allows us to use placement policies that take 
free space or load into account on a container by container basis. This is what 
HDFS does for every single block - the pipeline lives for the block and then is 
gone when it is closed.

The current Ratis pipeline design is a total disaster and the best thing we 
could do is get rid of it. Hopefully by replacing it with what is designed 
here. The new design has to account for the problems with the old one and long 
(forever) lived pipelines is a major part of that.

I also think that the pipeline concept stems totally from Ratis consensus and 
how expensive it is to establish the pipelines on the DNs. If that expense goes 
away, then what does long lived pipelines get us, except complexity and hard to 
solve problems?

> Streaming Write Pipeline without Raft
> -------------------------------------
>
>                 Key: HDDS-15463
>                 URL: https://issues.apache.org/jira/browse/HDDS-15463
>             Project: Apache Ozone
>          Issue Type: New Feature
>          Components: Ozone Client, Ozone Datanode
>            Reporter: Tsz-wo Sze
>            Assignee: Tsz-wo Sze
>            Priority: Major
>              Labels: pull-request-available
>         Attachments: 20260712HDDS-15463 Streaming Write Pipeline without 
> Raft.pdf
>
>
> - V1) Raft Pipeline: Use Raft for both WriteChunk and PutBlock
> - V2)Streaming Write Pipeline (HDDS-4454): Use Ratis streaming (RATIS-979) 
> for WriteChunk and Raft for PutBlock
> - V3) Streaming Write Pipeline without Raft: Use Ratis streaming for both 
> WriteChunk and PutBlock.
> We implement V3 in this JIRA.  Will post a design and create subtasks.



--
This message was sent by Atlassian Jira
(v10.3.26#10030026)

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

Reply via email to