[
https://issues.apache.org/jira/browse/RATIS-1598?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17563402#comment-17563402
]
Tsz-wo Sze commented on RATIS-1598:
-----------------------------------
bq. ... Since req#1 is not completely processed, the RetryCache will not have
the entry for #req1. ...
[~William Song], the cache entry is added in queryCache(..); see
-
https://github.com/apache/ratis/blob/dc0b68b4c0b8c187a08f669422a2cd099d7be0b7/ratis-server/src/main/java/org/apache/ratis/server/impl/RetryCacheImpl.java#L202
.
So the entry is put to the cache in the very beginning in
submitClientRequestAsync(..); see
-
https://github.com/apache/ratis/blob/dc0b68b4c0b8c187a08f669422a2cd099d7be0b7/ratis-server/src/main/java/org/apache/ratis/server/impl/RaftServerImpl.java#L768
.
Note that the future inside the entry is not yet completed at that time. When
there is a retry, the retry will wait for the same future. As a conclusion, we
do have exactly-once semantic in the current code.
> Check RetryCache before applyLog
> --------------------------------
>
> Key: RATIS-1598
> URL: https://issues.apache.org/jira/browse/RATIS-1598
> Project: Ratis
> Issue Type: Improvement
> Reporter: Song Ziyang
> Priority: Major
>
> RetryCache can filter redundant retry requests when a client-request arrives
> at RaftServer. This filtering happens before AppendLog. However, to support
> exactly-once semantic, it is necessary to check RetryCache also before
> ApplyLog.
>
> Consider this scenario:
> RetryCache is updated when the request is *successfully committed to
> RaftLog, before replying to client.*
> If the same 2 requests arrive at roughly same time, the req#1 pass the
> RetryCache check and successfully appended to RaftLog, but before updating
> RetryCache and reply, the req#2 arrives. Since req#1 is not completely
> processed, the RetryCache will not have the entry for #req1. In this case,
> req#2 will also pass RetryCache check and start to be appended to log. This
> leads to these 2 same requests both be applied.
>
> If the RetryCache is also checked before applyLog, we can still filter #req2
> before statemachine knows it.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
