| CPC G06F 16/214 (2019.01) [G06F 16/24552 (2019.01); G06F 16/283 (2019.01)] | 18 Claims |
|
1. A method for performing an auto-parallel-load operation that loads a set of target objects, currently residing in an on-disk database system, into an in-memory database system, comprising:
initiating the auto-parallel-load operation by a single load call made to an auto-parallel load module, wherein the single load call includes information that indicates a particular set of objects;
in response to the single load call, initiating a state object that includes metadata for one or more target objects in the set of target objects; and
invoking each pipeline module of a plurality of pipeline modules in sequence:
wherein the plurality of pipeline modules includes one or more pipeline modules that:
generate estimates based on the metadata, contained in the state object, about the one or more target objects; and
augment the state object by storing the estimates in the state object;
wherein the auto-parallel-load operation involves a plurality of load-related tasks;
wherein each module of the plurality of pipeline modules performs a distinct load-related task of the plurality of load-related tasks;
wherein invoking each pipeline module includes:
invoking a particular pipeline module that generates, for each target object in the set of target objects, an estimate based on:
features of the target object; and
output generated by a trained machine learning (ML) model that has been trained to generate estimates for target objects based on the features of the target objects; and
invoking a generate-and-execute module that is configured to generate database statements which, when executed, cause the in-memory database system to load the target objects based on the estimates that were generated for the target objects;
wherein the method is performed by one or more computing devices.
|