[
https://issues.apache.org/jira/browse/RATIS-2379?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Ivan Andika updated RATIS-2379:
-------------------------------
Description:
>From the discussion in RATIS-2350, we can return applied index for ReadIndex
>instead of commit index to reduce the wait time for a read index (since
>applied index <= commit index). The main argument is that since writes to the
>leader returns after the leader applies the write transactions, we can still
>see the latest write even if the ReadIndex returns the applied index.
>Therefore, the strong consistency requirement is still met. The other pending
>Raft transactions are considered as "concurrent" writes (since the responses
>have not been generated yet) and linearizability does not enforce any strict
>ordering for concurrent writes in the final total order.
We can make this configurable for now.
was:
>From the discussion in RATIS-2350, we can return applied index for ReadIndex
>instead of commit index to reduce the wait time for a read index (since
>applied index <= commit index). The main argument is that since writes to the
>leader returns after the leader applies the write transactions, we can still
>see the latest write even if the ReadIndex returns the applied index.
>Therefore, the strong consistency requirement is still met. The other pending
>Raft transactions are considered as "concurrent" writes (since the response
>has not been generated yet) and linearizability does not enforce any strict
>ordering for concurrent writes in the final total order.
We can make this configurable for now.
> Support returning applied index for ReadIndex
> ---------------------------------------------
>
> Key: RATIS-2379
> URL: https://issues.apache.org/jira/browse/RATIS-2379
> Project: Ratis
> Issue Type: Task
> Components: Linearizable Read
> Reporter: Ivan Andika
> Assignee: Ivan Andika
> Priority: Major
> Time Spent: 1h 10m
> Remaining Estimate: 0h
>
> From the discussion in RATIS-2350, we can return applied index for ReadIndex
> instead of commit index to reduce the wait time for a read index (since
> applied index <= commit index). The main argument is that since writes to the
> leader returns after the leader applies the write transactions, we can still
> see the latest write even if the ReadIndex returns the applied index.
> Therefore, the strong consistency requirement is still met. The other pending
> Raft transactions are considered as "concurrent" writes (since the responses
> have not been generated yet) and linearizability does not enforce any strict
> ordering for concurrent writes in the final total order.
> We can make this configurable for now.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
