[
https://issues.apache.org/jira/browse/SOLR-16924?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17844085#comment-17844085
]
David Smiley commented on SOLR-16924:
-------------------------------------
[~gerlowskija] I could use your insights on something here, please. I am
looking back at this months later and got very confused for a while until I
realized we have two RestoreCore classes, one in
{{org.apache.solr.handler.admin.api}} and the other in
{{{}org.apache.solr.handler{}}}. Wow; ok! The first implements the V2 API and
calls the second, the V1 API. The change in this PR was placed at the end of
the V2 API without consideration of the ambiguity. Uh oh! Thus it would seem
the change will not take effect for the V1 API {*}but{*}, I see in SOLR-16490
that RestoreCoreOp (yet another layer below the CoreAdminHandler but above the
real impl) calls the V2 RestoreCore. Wow again; I didn't expect that! I
wonder then, is V1 RestoreCore invoked in any other code path? I see two –
{{ReplicationHandler.restore()}} and {{{}InstallCoreData.installCoreData(){}}}.
Are there any problems or bugs here? Like, _should_ the change in this PR be
placed at the end of V1 instead? Isn't it wrong for CoreAdminHandler to be
calling v2 stuff? On second thought, I could rationalize that as we transition
the migration. Should the RestoreCore classes be merged?
> Restore: Have RESTORECORE set the UpdateLog state
> --------------------------------------------------
>
> Key: SOLR-16924
> URL: https://issues.apache.org/jira/browse/SOLR-16924
> Project: Solr
> Issue Type: Improvement
> Reporter: David Smiley
> Priority: Minor
> Fix For: 9.5
>
> Time Spent: 50m
> Remaining Estimate: 0h
>
> This is a refactoring improvement designed to simplify & clarify a step in
> collection restores. One of the final phases of RestoreCmd (collection
> restore) is to call REQUESTAPPLYUPDATES on each newly restored replica in
> order to transition the state of the UpdateLog to ACTIVE (not actually to
> apply updates). The underlying call on the UpdateLog could instead be done
> inside RESTORECORE at the end with explanatory comments as to the intent. I
> think this makes more sense that RESTORECORE finish with its UpdateLog ready.
> And it's strange/curious to see requests in the cluster to apply updates
> from an updateLog when there is none to do! Adding clarifying comments is
> important.
> See my comment:
> https://issues.apache.org/jira/browse/SOLR-12065?focusedCommentId=17751792&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-17751792
> I think there isn't any back-compat concern.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@solr.apache.org
For additional commands, e-mail: issues-h...@solr.apache.org
