[jira] [Commented] (IGNITE-15885) Create SortedIndexStorage based on RocksDB

["Semyon Danilov (Jira)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/IGNITE-15885?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17459245#comment-17459245 ] 

Semyon Danilov commented on IGNITE-15885:
-----------------------------------------

LGTM! Thanks for the contribution, merged to the main branch.

> Create SortedIndexStorage based on RocksDB
> ------------------------------------------
>
>                 Key: IGNITE-15885
>                 URL: https://issues.apache.org/jira/browse/IGNITE-15885
>             Project: Ignite
>          Issue Type: Task
>            Reporter: Aleksandr Polovtcev
>            Assignee: Aleksandr Polovtcev
>            Priority: Major
>              Labels: ignite-3
>          Time Spent: 7.5h
>  Remaining Estimate: 0h
>
> h1. API
> * Add {{SortedIndexStorage}} interface:
> {code:java}
> public interface SortedIndexStorage extends AutoCloseable {
>     /** Index key factory */
>     IndexKeyFactory indexKeyFactory();
>     /** Index descriptor */
>     SortedIndexDescriptor indexDescriptor();
>     /** Adds a value */
>     void put(IndexKey key, SearchRow value);
>     /** Removes a value */
>     void remove(IndexKey key);
>     /** Range scan */
>     Cursor<SearchRow> range(IndexKeyPrefix lowerBound, IndexKeyPrefix upperBound);
>     /** Removes all values */
>     void destroy();
> }
> {code}
> * {{IndexKeyPrefix}} can contain a smaller number of columns, than the {{IndexKey}}, needed for the prefix search.
> {code:java}
> public interface IndexKey {
>     byte[] asBytes();
> }
> public interface IndexKeyPrefix {
>     Object[] prefixColumnValues();
> }
> {code}
> * {{SearchRow}} is a class from the {{storage-api}} that represents a key from the partition storage.
> h1. Implementation details
> For the initial implementation it is suggested to use {{BinaryRow}} serialization mechanism (it is proposed to implement the {{IndexKey}} on top of it, simply ignoring the value bytes) to store it in RocksDB. First implementation will also use a naive comparator, that will convert each {{BinaryRow}} into a {{Row}} and compare individual deserialized columns. 
> It is also proposed to store partition storage keys as values in the index storage.



--
This message was sent by Atlassian Jira
(v8.20.1#820001)