[
https://issues.apache.org/jira/browse/HDDS-6589?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Jie Yao updated HDDS-6589:
--------------------------
Description:
# Rename the existing RM class to LegacyReplicationManager, and expose the
processContainer() method as protected or public. Remove the scheduling from
this RM.
# Create a new replication Manager, called ReplicationManager, then have it
scheduled periodically like the current one.
# When the new RM processes a container, if it is EC, it goes down the new
path. If it is not EC, we call LegacyReplicationManager.processContainer(...).
was:
After HDDS-3816 is merged into master branch, we need to process EC container
in replication manager, which is different to ratis container . . the current
code of replication manager is complex enough as it stands, i think we need to
refactor it and split it into several separate modules as below:
1 the main class, that is a replication manager class, which will act as a
`SCMService` and run `processContainer` periodicly.
2 an `ContainerProcessor` interface , which has a function of
`processContainer` , and two implementation : `RatisContainerProcessor` and
`ECContainerProcessor`. because most of the logic in the two implementation
will be the same , we need to add a abstract class , let say
`AbstractContainerProcessor` , and in that two implementations we just
implement the abstract functions according to the specified container type.
3 add an interface `moveScheduler` and a implementation `moveSchdulerImpl` ,
which are used to support container move in replication manager.
4 if there are some common util functions are called among different class in
replication manager, we can add a utils class.
i think after this refactor, the code of relication manager will be more
modular and easily extendable
> Add replication manager V2
> --------------------------
>
> Key: HDDS-6589
> URL: https://issues.apache.org/jira/browse/HDDS-6589
> Project: Apache Ozone
> Issue Type: Improvement
> Reporter: Jie Yao
> Assignee: Jie Yao
> Priority: Major
>
> # Rename the existing RM class to LegacyReplicationManager, and expose the
> processContainer() method as protected or public. Remove the scheduling from
> this RM.
> # Create a new replication Manager, called ReplicationManager, then have it
> scheduled periodically like the current one.
> # When the new RM processes a container, if it is EC, it goes down the new
> path. If it is not EC, we call LegacyReplicationManager.processContainer(...).
--
This message was sent by Atlassian Jira
(v8.20.7#820007)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
