apache/gravitino

[Improvement] Improve the current SQL of JDBCBackend

Open

#4 024 ouverte le 2 juil. 2024

Voir sur GitHub
 (6 commentaires) (0 réactions) (0 assignés)Java (887 forks)auto 404
help wantedimprovement

Métriques du dépôt

Stars
 (3 058 stars)
Métriques de merge PR
 (Métriques PR en attente)

Description

What would you like to be improved?

In the current implementation of xxxMetaMapper and xxxMetaService, we seldom use JOIN and transactions, for example:

The list catalog operation will issue two SQL queries, the first one is to get a metalake id, then using this id to get the catalog list.

Instead of issuing two queries, we can use JOIN to join catalog and metalake table to get catalog list by one query, this will save the IO time and avoid inconsistent problem.

Also for updateCatalog

We have several steps:

  1. get metalake id.
  2. get catalog PO.
  3. update the catalog object.
  4. update the catalog in RDBMS.

These 4 steps are not in the transaction, which will potentially meet the inconsistent problem, and we highly leverage the lock to avoid the concurrent problem. A better solution is to put these 4 steps into a transaction.

How should we improve?

So basically, we can improve the current SQL to:

  1. reduce the queries and increase the performance.
  2. Put multiple step actions into one transaction to leverage DB's transaction to keep consistent.

Guide contributeur