You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@ignite.apache.org by "Semyon Danilov (Jira)" <ji...@apache.org> on 2022/04/06 09:35:00 UTC

[jira] [Created] (IGNITE-16803) RocksDB usage refinement

Semyon Danilov created IGNITE-16803:
---------------------------------------

             Summary: RocksDB usage refinement
                 Key: IGNITE-16803
                 URL: https://issues.apache.org/jira/browse/IGNITE-16803
             Project: Ignite
          Issue Type: Improvement
            Reporter: Semyon Danilov


Ignite 3 uses RocksDB in many different components: metastorage, partition storage, log storage and vault. At this moment, every component creates its own instance of rocksdb. Moreover, partition storage and log storage create multiple instances: partition storage creates an instance per table and log storage creates an instance per raft group. 
This creates high memory overhead and causes out of memory errors. We have to find a way to use as few instances as possible (ideally one per ignite node). 
For example, MySQL over RocksDB uses only one instance of RocksDB but "create table" query offers a way to put data into another column family. Maybe we can store all raft logs in one column family and data (partitions and meta) in another.




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