[
https://issues.apache.org/jira/browse/RATIS-2712?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18118177#comment-18118177
]
Ivan Andika commented on RATIS-2712:
------------------------------------
> The Hadoop RPC ThreadLocal should be valid only at the Hadoop RPC layer. It
> should not force other components supporting it. For example, Hadoop RPC
> internally use ThreadLocal to store a RemoteUser. When the thread receives
> the RPC call, it should get the RemoteUser (and clean the ThreadLocal). Then,
> pass it to another component (say Ratis), which can use any thread model
> processing it.
Thanks for the discussion. My thought is that ThreadLocal context propagation
is more implicit / transparent to the implementation of StateMachine#query
without requiring explicitly passing the ThreadLocal variable to a context
object or in a method. I haven't really worked out the inner details, but it's
either we make the StateMachine implementation to allow registration for
ThreadLocal variables that can be passed in a specific point of Ratis when
there are thread change or we somehow copy the internal ThreadLocalMap for
every thread change.
However, I understand that it might not be appropriate for Ratis just to change
to only support specific Hadoop RPC ThreadLocal.
> Hadoop RPC internally use ThreadLocal to store a RemoteUser. When the thread
> receives the RPC call, it should get the RemoteUser (and clean the
> ThreadLocal). Then, pass it to another component (say Ratis), which can use
> any thread model processing it.
Yes, currently that is the approach done in HDDS-16497.
> Support ThreadLocal propagation for Ratis
> -----------------------------------------
>
> Key: RATIS-2712
> URL: https://issues.apache.org/jira/browse/RATIS-2712
> Project: Ratis
> Issue Type: New Feature
> Reporter: Ivan Andika
> Assignee: Ivan Andika
> Priority: Major
>
> See the issue in https://issues.apache.org/jira/browse/HDDS-16497 .
> Since the StateMachine#applyTransactions and StateMachine#query are executed
> in another threads, ThreadLocal mechanisms (like in Hadoop RPC) will not work
> (or requires workaround).
> We can discuss whether we should support ThreadLocal propagation as a first
> class feature of Ratis. This can be similar to
> [https://github.com/alibaba/transmittable-thread-local]
>
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
