[
https://issues.apache.org/jira/browse/IGNITE-19276?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Roman Puchkovskiy updated IGNITE-19276:
---------------------------------------
Summary: Implement a mechanism to build indices distributively (was: Track
partitions while building indexes)
> Implement a mechanism to build indices distributively
> -----------------------------------------------------
>
> Key: IGNITE-19276
> URL: https://issues.apache.org/jira/browse/IGNITE-19276
> Project: Ignite
> Issue Type: Improvement
> Reporter: Ivan Bessonov
> Priority: Major
> Labels: ignite-3
>
> IGNITE-18539 states:
> {code:java}
> when index is created, we also create a table somewhere that looks similar to
> this:
> indexCompletion.<indexId>.0 = false
> ...
> indexCompletion.<indexId>.N = false {code}
> This part has not been implemented, and it's separated into another issue.
> Depending on the catalog service implementation, this data must either go to
> meta-storage directly, or into a corresponding section in private
> configuration.
> h3. Things to consider
> * RAFT replica listener node may die after sending last "build index"
> command, thus not updating, so new replica listener must compare states of
> indexes in schema with states of indexes in storage.
> * One of nodes must create new schema version when all partitions are built.
> Choosing such node is a hard problem, so we can simply ask all nodes to do
> this operation, making it idempotent and expecting that one of them would
> win. This part requires careful thinking about the specifics.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
