If you find that the execution speed of some query does not reach the expectation, the optimizer might choose the wrong index to run the query.
You can first view the health state of tables in the statistics, and then solve this issue according to the different health states.
The low health state means TiDB has not performed the
ANALYZE statement for a long time. You can update the statistics by running the
ANALYZE command. After the update, if the optimizer still uses the wrong index, refer to the next section.
The near 100% health state suggests that the
ANALYZE statement is just completed or was completed a short time ago. In this case, the wrong index issue might be related to TiDB's estimation logic for the number of rows.
For equivalence queries, the cause might be Count-Min Sketch. You can check whether Count-Min Sketch is the cause and take corresponding solutions.
If the cause above does not apply to your problem, you can force-select indexes by using the
use index optimzer hint (see USE_INDEX for details). Also, you can change the query behavior by using SQL Plan Management in a non-intrusive way.
Apart from the aforementioned situations, the wrong index issue might also be caused by data updates which renders all the indexes no longer applicable. In such cases, you need to perform analysis on the conditions and data distribution to see whether new indexes can speed up the query. If so, you can add new indexes by running the
ADD INDEX command.