[
https://issues.apache.org/jira/browse/IGNITE-23376?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Alexander Lapin updated IGNITE-23376:
-------------------------------------
Epic Link: IGNITE-23931
> Possibility to stop replica despite the lease state
> ---------------------------------------------------
>
> Key: IGNITE-23376
> URL: https://issues.apache.org/jira/browse/IGNITE-23376
> Project: Ignite
> Issue Type: Improvement
> Reporter: Vladislav Pyatkov
> Priority: Major
> Labels: ignite-3
>
> h3. Motivation
> In some cases, for example, recovery or manual user intervention, we ought to
> stop replica as soon as possible. It could be a primary replica or the other
> one, because we are processing operations through the primary (e.g., RW
> transaction) or through the other (e.g., RO transaction).
> Currently, we may get an exception when trying to access a closing storage or
> getting some inappropriate state in the partition replica listener, etc.
> h3. Implementation notes
> # Probably, implementation based on busy locks for each replica will satisfy
> this ticket.
> # We also need to busy lock the replica in actions like building the
> indexes, because they try to await/get primary replica and interact with its
> storage without sending a request, see IGNITE-22191 .
> # We also will apply this approach to recovery procedure, which was worked
> around here IGNITE-23125
> h3. Definition of done
> The replica stop process has to be thread safe regarding other operations
> with replica.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
