Stephen O'Donnell created HDDS-8187:
---------------------------------------
Summary: ReplicationManager: Datanode commands should be sent to
nodeManager directly
Key: HDDS-8187
URL: https://issues.apache.org/jira/browse/HDDS-8187
Project: Apache Ozone
Issue Type: Sub-task
Components: SCM
Reporter: Stephen O'Donnell
Assignee: Stephen O'Donnell
In order to keep a consistent count of the number of pending replicate and
delete commands, ReplicationManager should send commands to the datanode via
the nodeManager.addDatanodeCommand() method rather than publishing them to the
event queue.
At the moment, the commands goto the event queue, where they are picked up by
NodeManager and added to the datanode command queue. As we are using the counts
in the command queue to determine how many commands are pending, the
indirection provided by the event queue makes it difficult to keep a consistent
count without excessive effort.
There is also no further processing done on the commands in the event queue,
they simply go RM -> Event Queue -> Node Manager -> Command Queue.
This change makes RM send the commands to nodeManager directly.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
