Parallel execution of query sub-plans
Sub-plans are executed in parallel using a plurality of execution nodes, which can be part of a data platform. In particular, sub-plans (e.g., fragments or portions of one or more child operators) of a root operator are identified in a query plan such that the identified sub-plans that are candidates for execution on a single execution node, determine a cost estimate for causing the candidate sub-plans to be executed in parallel using multiple execution nodes, and cause the candidate sub-plans to be executed in parallel based on the cost estimate.
1. A system comprising:
at least one hardware processor; and
at least one memory storing instructions that cause the at least one hardware processor to perform operations comprising:
accessing a query plan that comprises at least one root operator, the at least one root operator being associated with a plurality of sub-plans;
identifying, from the plurality of sub-plans, a sub-plurality of sub-plans that each involves processing a data set that satisfies a set of criteria;
estimating a first cost for executing individual sub-plans of the sub-plurality of sub-plans successively using a plurality of execution nodes;
estimating a second cost for executing the sub-plurality of sub-plans in parallel using the plurality of execution nodes;
comparing the first cost to the second cost;
determining, based on the comparing, whether the sub-plurality of sub-plans is to be executed in parallel using the plurality of execution nodes;
in response to determining that the sub-plurality of sub-plans is to be executed in parallel using the plurality of execution nodes, causing the sub-plurality of sub-plans to be executed in parallel using the plurality of execution nodes; and
in response to determining that the sub-plurality of sub-plans is not to be executed in parallel using the plurality of execution nodes, causing the individual sub-plans of the sub-plurality of sub-plans to be executed successively such that all execution nodes of the plurality of execution nodes are reserved for execution of the individual sub-plan until the individual sub-plan completes execution.
2. The system of claim 1 , wherein the determining, based on the comparing, whether the sub-plurality of sub-plans is to be executed in parallel using the plurality of execution nodes comprises:
based on the comparing, determining whether the first cost is greater than the second cost; and
in response to determining that the first cost is greater than the second cost, determining to execute the sub-plurality of sub-plans in parallel using the plurality of execution nodes.
3. The system of claim 1 , wherein the determining, based on the comparing, whether the sub-plurality of sub-plans is to be executed in parallel using the plurality of execution nodes comprises:
based on the comparing, determining whether the first cost is at least greater than the second cost by a threshold value; and
in response to determining that the first cost is at least greater than the second cost by the threshold value, determining to execute the sub-plurality of sub-plans in parallel using the plurality of execution nodes.
4. The system of claim 3 , wherein the threshold value comprises a ratio of the first cost and the second cost.
5. The system of claim 1 , wherein the causing of the individual sub-plans of the sub-plurality of sub-plans to be executed successively comprises:
successively scheduling the individual sub-plans of the sub-plurality of sub-plans to be executed using all execution nodes of the plurality of execution nodes, the individual sub-plans being successively scheduled such that a next sub-plan of the sub-plurality of sub-plans is not scheduled until a last sub-plan of the sub-plurality of sub-plans has completed execution.
6. The system of claim 5 , wherein the successively scheduling comprises performing a separate scheduling operation for each execution node of the plurality of execution nodes to be used to execute a portion of a current sub-plan.
7. The system of claim 1 , wherein the causing of the sub-plurality of sub-plans to be executed in parallel using the plurality of execution nodes comprises:
for each single sub-plan in the sub-plurality of sub-plans, performing a single scheduling operation to assign execution of the single sub-plan to a single execution node of the plurality of execution nodes.
8. The system of claim 7 , wherein the single execution node is configured to generate a local schedule queue to process the single sub-plan assigned to the single execution node.
9. The system of claim 7 , wherein the single scheduling operation to assign execution of the single sub-plan to the single execution node comprises:
adding the single sub-plan to a current scheduling queue of the single execution node.
10. The system of claim 1 , wherein the data set comprises one or more files.
11. The system of claim 1 , wherein the at least one root operator comprising a join operator.
12. The system of claim 1 , wherein the at least one root operator comprising a union all operator.
13. The system of claim 1 , wherein sub-plans in the sub-plurality of sub-plans have a same set of dependencies.
14. The system of claim 12 , wherein the operations comprise:
maintaining a dependency map that describes one or more dependencies between individual operators to be executed, the dependency map being initialized based on the query plan, and the identifying the sub-plurality of sub-plans being based on the dependency map.
15. The system of claim 1 , wherein the operations comprise:
selecting the at least one root operator from a global scheduling queue.
16. The system of claim 1 , wherein the set of criteria includes one or more criteria for identifying the data set that is capable of being executed by a single execution node.
17. A method comprising:
accessing, by a hardware processor, a query plan that comprises at least one root operator, the at least one root operator being associated with a plurality of sub-plans;
identifying, by the hardware processor and from the plurality of sub-plans, a sub-plurality of sub-plans that each involves processing a data set that satisfies a set of criteria;
estimating, by the hardware processor, a first cost for executing individual sub-plans of the sub-plurality of sub-plans successively using a plurality of execution nodes;
estimating, by the hardware processor, a second cost for executing the sub-plurality of sub-plans in parallel using the plurality of execution nodes;
comparing, by the hardware processor, the first cost to the second cost;
determining, by the hardware processor and based on the comparing, that the sub-plurality of sub-plans is to be executed in parallel using the plurality of execution nodes;
in response to determining that the sub-plurality of sub-plans is to be executed in parallel using the plurality of execution nodes, causing, by the hardware processor, the sub-plurality of sub-plans to be executed in parallel using the plurality of execution nodes; and
executed in parallel using the plurality of execution nodes, causing the individual sub-plans of the sub-plurality of sub-plans to be executed successively such that all execution nodes of the plurality of execution nodes are reserved for execution of the individual sub-plan until the individual sub-plan completes execution.
18. The method of claim 17 , wherein the determining, based on the comparing, whether the sub-plurality of sub-plans is to be executed in parallel using the plurality of execution nodes comprises:
based on the comparing, determining whether the first cost is greater than the second cost; and
in response to determining that the first cost is greater than the second cost, determining to execute the sub-plurality of sub-plans in parallel.
19. The method of claim 17 , wherein the determining, based on the comparing, whether the sub-plurality of sub-plans is to be executed in parallel using the plurality of execution nodes comprises:
based on the comparing, determining whether the first cost is at least greater than the second cost by a threshold value; and
in response to determining that the first cost is at least greater than the second cost by the threshold value, determining to execute the sub-plurality of sub-plans in parallel using the plurality of execution nodes.
20. The method of claim 19 , wherein the threshold value comprises a ratio of the first cost and the second cost.
21. The method of claim 17 , wherein the causing of the individual sub-plans of the sub-plurality of sub-plans to be executed successively comprises:
successively scheduling the individual sub-plans of the sub-plurality of sub-plans to be executed using all execution nodes of the plurality of execution nodes, the individual sub-plans being successively scheduled such that a next sub-plan of the sub-plurality of sub-plans is not scheduled until a last sub-plan of the sub-plurality of sub-plans has completed execution.
22. The method of claim 21 , wherein the successively scheduling comprises performing a separate scheduling operation for each execution node of the plurality of execution nodes to be used to execute a portion of a current sub-plan.
23. The method of claim 17 , wherein the causing of the sub-plurality of sub-plans to be executed in parallel using the plurality of execution nodes comprises:
for each single sub-plan in the sub-plurality of sub-plans, performing a single scheduling operation to assign execution of the single sub-plan to a single execution node of the plurality of execution nodes.
24. The method of claim 23 , wherein the single execution node is configured to generate a local schedule queue to process the single sub-plan assigned to the single execution node.
25. The method of claim 23 , wherein the single scheduling operation to assign execution of the single sub-plan to the single execution node comprises:
adding the single sub-plan to a current scheduling queue of the single execution node.
26. The method of claim 17 , wherein the data set comprises one or more files.
27. The method of claim 17 , wherein the at least one root operator comprising at least one of a join operator or a union all operator.
28. A non-transitory computer-readable medium comprising instructions that, when executed by one or more processors of a machine, configure the machine to perform operations comprising:
accessing a query plan that comprises at least one root operator, the at least one root operator being associated with a plurality of sub-plans;
identifying, from the plurality of sub-plans, a sub-plurality of sub-plans that each involves processing a data set that satisfies a set of criteria;
estimating a first cost for executing individual sub-plans of the sub-plurality of sub-plans successively using a plurality of execution nodes;
estimating a second cost for executing the sub-plurality of sub-plans in parallel using the plurality of execution nodes using the plurality of execution nodes;
comparing the first cost to the second cost;
determining, based on the comparing, whether the sub-plurality of sub-plans is to be executed in parallel using the plurality of execution nodes;
in response to determining that the sub-plurality of sub-plans is to be executed in parallel using the plurality of execution nodes, causing the sub-plurality of sub-plans to be executed in parallel using the plurality of execution nodes; and
in response to determining that the sub-plurality of sub-plans is not to be plans of the sub-plurality of sub-plans to be executed successively such that all individual sub-plan until the individual sub-plan completes execution.