Based on query patterns, the algorithm strives to place rows that are most frequently joined closer together and within the same, or adjoining partitions on the same node. Other Autopilot functions include optimizations for loading and deciding where to physically place data. With Autopilot, this scenario can be avoided as it predicts the execution time of each query, then prioritizes short queries over long ones to reduce or eliminate waits for interactive queries.
Without such smart query scheduling, short queries, which are expected to be interactive, could get delayed by long workloads unless a database administrator takes action. In this case, the output of the model is making recommendations, including an explanation for the recommendation to the customer, who ultimately decides on cluster sizing.Ĭonversely, in the case of query execution, the model automatically schedules queries using logic to reduce contention between short- and long-running queries characteristic of mixed workloads.
This is based on predicting the amount of memory that will be required for holding the table and the intermediate results in memory. The latest release of MySQL HeatWave also increases cluster sizing to 64, more than doubling capacity limits to 32-TByte instances, and raises scalability by 20%.įor instance, when provisioning the database, Autopilot conducts a smart scan of the table using adaptive sampling techniques to predict the most optimal cluster sizing. And with machine learning, Autopilot can iteratively improve query plans based on results from other similar, different queries. It assists users by making recommendations on how to provision and load the database, and implements closed-loop automation for failure handling/error recovery and query execution.
Specifically, MySQL Autopilot uses internalized machine learning to assist users for some tasks, and applies closed-loop automation to handle others.
0 kommentar(er)
