APPARATUS FOR ELASTIC DATABASE PROCESSING WITH HETEROGENEOUS DATA

• US 20120036146A1
• Filed: 10/04/2011
• Published: 02/09/2012
• Est. Priority Date: 10/26/2010
• Status: Abandoned Application

• Current Assignee: Tesora, Inc. (Stratoscale Ltd.)
• Original Assignee: Tesora, Inc. (Stratoscale Ltd.)
• Inventors: Annapragada, Mrithyunjaya
• Application Number:
  US13/252,742
• Publication Number:
  US 20120036146A1
• Time in Patent Office: Days
• US Class Current: 707/764
• CPC Class Codes: G06F 16/2471 Distributed queries
• CPC Class Codes: G06F 16/27 Replication, distribution or synchronisation of data between databases or within a distributed database system; Distributed database system architectures therefor
• CPC Class Codes: G06F 16/278 Data partitioning, e.g. horizontal or vertical partitioning

Abstract

A database management system implemented in a cloud computing environment. Operational nodes are assigned as groups of controller-nodes, compute-nodes or storage-nodes. Queries specify one or more tables for an associated database operation, with each table being assigned to respective storage nodegroup(s). The number of nodes executing a given query may change, by (a) changing the compute-nodes associated with a connection, or (b) adding or removing nodes associated with a connection; and/or distributing data to a storage nodegroup based on a Distribution Method which may be either data dependent or data independent. A controller node further executes a Dynamic Query Planner (DQP) process that develops a query plan.

Claims

1. A database management system comprising:

a network interface, for receiving database queries from two or more client application processes as a network database service, the client application processes originating from two different users, the system providing a least one connection into the system for each such client application process;

a group of two or more operational nodes for executing the queries as database operations, each operational node implemented as a logical collection of software components that execute on one or more physical machines;

where the number of physical machines is not necessarily the same as the number of operational nodes;

with the operational nodes assigned as controller-nodes, compute-nodes or storage-nodes, and groups of controller-nodes forming controller nodegroups, and groups of compute-nodes forming compute nodegroups, and groups of storage nodes forming storage nodegroups;

the number of operational nodes, and their available assignment as compute-nodes or storage-nodes varying during execution of the queries;

each client connection being assigned to an associated compute nodegroup;

the queries also specifying one or more tables for an associated database operation, with each such table being assigned to a respective storage nodegroup;

the operational nodes further;

operating in parallel;

with the number of operational nodes executing a given query or queries changing during a given time interval by at least one of;

(a) changing the compute-nodegroup associated with a connection, or

(b) adding or removing nodes from the compute nodegroup associated with a connection; and

distributing data from the tables among the nodes in the storage nodegroup to which the table is assigned based on a Distribution Method which may be either data dependent or data independent; and

at least one of the controller nodes further;

executing a Dynamic Query Planner (DQP) process that transforms queries received from the client into a query plan that includes an ordered series of steps that are executed in parallel on multiple operational nodes where possible, the query plan further stipulating, for each step, which compute node it must be performed on, which storage nodes it must access, and other steps that this step depends on.

2. The apparatus of claim 1 wherein the operational nodes include two or more Query Execution Engines, a Data Distribution Manager, and a Persistent Data Store.

3. The apparatus of claim 1 wherein the software components comprising the operational nodes execute on some combination of physical machines and virtual machines.

4. The apparatus of claim 1 wherein the storage nodes further provide a Persistent Data Store as at least one of relational database, flat files, or non-relational databases.

5. The apparatus of claim 1 wherein

the storage and compute nodes contain multiple potentially heterogeneous data stores that can be presented and queried as if they were one database;

and

wherein the steps in the query plan generated by the DQP process are expressed in a form that is understood by an underlying data store on the node(s) executing the step(s); and

wherein the form in which the client communicates the query to the system may or may not be in the same form used to communicate with the data stores on the compute and storage nodes.

6. The apparatus of claim 1 wherein each nodegoup may store more than one table.

7. The apparatus of claim 1 wherein transient tables created during query execution are stored on a compute nodegroup associated with the connection, and unless otherwise specified a nodegroup designated as the default compute nodegroup is used for storing the transient tables.

8. The apparatus of claim 1 wherein a query may specify a nodegroup on which to store a table, but if a query does not, a storage nodegroup designated as the default storage nodegroup is used for storing the table.

9. The apparatus of claim 1 wherein compute and storage nodegroups are further assigned reference counters and table counters, with

(a) a reference counter being incremented when a nodegroup is assigned as the default storage or default compute nodegroup;

(b) a reference counter being incremented when a nodegroup is assigned as the compute nodegroup for a connection;

(c) the reference counter being decremented when a nodegroup is no longer the default storage or default compute nodegroup;

(d) the reference counter being decremented when a nodegroup is no longer the compute nodegroup for a connection; and

(e) the table counter being incremented each time a table is associated with the nodegroup; and

(f) the table counter being decremented each time a table is disassociated from a nodegroup.

10. The apparatus of claim 9 wherein the reference counter and table counter are initialized to an initial-value, and wherein a nodegoup can be deleted when both its associated reference counter and table counter reach the initial value.

11. The apparatus of claim 1 wherein the data independent distribution methods are broadcast, round-robin and random.

12. The apparatus of claim 1 wherein the data dependent distribution methods are either inelastic (IDD) methods or elastic (EDD) methods.

13. The apparatus of claim 1 wherein the DQP process further identifies groups of tables in the query, and assigns database operations that can be executed in parallel and at the same time to different selected storage and/or compute nodes, such that a final result will be as expected by the client application process(es).

14. The apparatus of claim 1 wherein steps in the plan may operate on sets of tables, and such that steps in the plan may (a) produce intermediate result tables for later use (b) return results to the client;

(c) delete some or all of the intermediate result tables;

or (d) perform some other operation specified by the client application process(es).

15. The apparatus of claim 1 wherein the system further maintains a reference counter associated with each table referenced in a query, and the table can be deleted when it is no longer being referenced by any query.

16. The apparatus of claim 1 wherein a Query Execution Engine receives a plan from the DQP and then identifies the number of other Query Execution Engines required to execute the plan, identifies other tables addressed in the query, and identifies physical or virtual machines where Query Execution Engines may be used for the plan, and contacts Machine Managers to provision Query Execution Engines from the operational nodes, and where appropriate additional physical or virtual machines are provisioned for the same.

17. The apparatus of claim 1 wherein the operational nodes are provisioned as one or more Query Execution Engines, and are provided with client connection state information prior to executing a step, and the connection state information is saved for later use at the completion of execution of a step.

18. The apparatus of claim 17 wherein after query execution, some or all of the Query Execution Engines may be released, and if any physical or virtual machines were provisioned, they may be freed as well.

19. The system of claim 1 wherein

intermediate processing of queries is decoupled from activities that are closely tied to where the data is stored by:

storing intermediate tables generated as part of a query plan on the compute-nodegroup associated with the connection; and

storing persistent user data on a storage-nodegroup; and

wherein client connection state is decoupled from the connection and provided to the query execution engine that executes each step in the query plan.

20. A database management system comprising:

a network interface, for receiving database queries from two or more client application processes as a network database service, the client application processes originating from two different users, the system providing a least one connection into the system for each such client application processes;

a group of two or more operational nodes for executing the queries as database operations, each operational node implemented as a logical collection of software components that execute on one or more physical machines;

where the number of physical machines is not necessarily the same as the number of operational nodes;

with the operational nodes assigned as controller-nodes, compute-nodes or storage-nodes, and groups of controller-nodes forming controller nodegroups, and groups of compute-nodes forming compute nodegroups, and groups of storage nodes forming storage nodegroups;

wherein the storage nodes further provide a Persistent Data Store as at least one of relational database, flat files, or non-relational databases;

with one of the compute nodegroups assigned as a default compute nodegroup and one of the storage nodegroups defined as a default storage nodegroup;

the number of operational nodes, and their available assignment as compute-nodes or storage-nodes varying during execution of the queries;

each client connection being assigned to an associated compute nodegroup;

the queries also specifying one or more tables for an associated database operation, with each such table being assigned to a respective storage nodegroup;

the operational nodes further;

operating in parallel;

with the number of operational nodes executing a given query or queries changing during a given time interval by at least one of;

(a) changing the compute-nodegroup associated with a connection, or

(b) adding or removing nodes from the compute nodegroup associated with a connection; and

distributing data from the tables among the nodes in the storage nodegroup to which the table is assigned based on a Distribution Method which may be either data dependent or data independent; and

at least one of the controller nodes further;

executing a Dynamic Query Planner (DQP) process that transforms queries received from the client into a query plan that includes an ordered series of steps that are executed in parallel on multiple operational nodes where possible, the query plan further stipulating, for each step, which compute node it must be performed on, which storage nodes it must access, and other steps that this step depends on; and

wherein nodegroups are further assigned reference counters and table counters, with

(a) a reference counter being incremented when a nodegroup is assigned as the default storage or default compute nodegroup;

(b) a reference counter being incremented when a nodegroup is assigned as the compute nodegroup for a connection;

(c) the reference counter being decremented when a nodegroup is no longer the default storage or default compute nodegroup; and

(d) the reference counter being decremented when a nodegroup is no longer the compute nodegroup for a connection; and

(e) the table counter being incremented each time a table is associated with the nodegroup; and

(f) the table counter being decremented each time a table is disassociated from a nodegroup,

wherein the reference counter and table counter are initialized to an initial-value; and

wherein a nodegroups can be deleted when both its associated reference counter and table counter reach the initial value; and

intermediate processing of queries is decoupled from activities that are closely tied to where the data is stored by;

storing intermediate tables generated as part of a query plan on the compute-nodegroup associated with the connection; and

storing persistent user data on a storage-nodegroup; and

wherein client connection state is decoupled from the connection and provided to the query execution engine that executes each step in the query plan.