DWS Resource Management Overview

Video Tutorial

Functions

The resource management function consists of the following parts:

• Resource management configuration: you can configure whether to enable resource management and the maximum number of global concurrent requests. The system allows a limited number of simultaneous tasks per CN. Disabling the resource manager turns off all its features.
• Resource pools: Compute resources are isolated and restricted to prevent cluster-level exceptions caused by abnormal SQL queries.
• Resource management plan: Resources are managed automatically based on a preconfigured plan, which can flexibly cope with complex scenarios.
• Schema space management: Storage is managed at user and schema level to prevent disk exhaustion, which makes the database read only. Currently, the management plane only supports schema space management.
• Exception rules: To avoid query blocking or performance deterioration, you can configure exception rules to let the service automatically identify and handle abnormal queries, preventing slow SQL statements from occupying too many resources for a long time.
• Query filter: The query filter creates rules to block problematic SQL statements early, preventing issues like poor performance or complete cluster failure.

Notes and Constraints

• Resource management is supported only by clusters of version 8.0.0 or later.
• Only clusters of version 8.1.3 and later versions support the dedicated CPU limit.
• Resource management plans are supported only by clusters of version 8.1.0.100 or later.

Concurrency Management

GaussDB(DWS) supports fine-grained resource management. Before resource management is implemented, queries are classified into complex queries (with long execution time and high resource consumption) and simple queries (with short execution time and low resource consumption). Simple and complex queries also differ in their estimated memory usage.

• The estimated memory usage of a simple query is less than 32 MB.
• The estimated memory usage of a complex query is 32 MB or higher.

In a hybrid load database, complex queries often occupy a large number of resources for a long time. A simple query queued after a complex query is time consuming, because it has to wait for the complex query to complete and resources to be freed up. To improve execution efficiency and system throughput, GaussDB(DWS) provides the short query acceleration function, managing simple queries separately. In the Short Query Configuration area, you can determine whether to enable the short query acceleration function. To change the number of concurrent simple statements (the default value is -1, and 0 or -1 indicates no control), you can enable short query acceleration.

The concurrency management rules are as follows:

• If short query acceleration is enabled, complex queries are under resource pool concurrency control, and simple queries are under short query concurrency control. Simple queries and complex queries are managed separately. Simple queries do not need to compete with complex queries for resources.
• If short query acceleration is disabled, complex and simple queries are both under resource pool concurrency control. Short query concurrency control is invalid. Both simple and complex queries handle resources similarly. Memory-based query division relies on accurate memory estimates. However, query runtime and CPU usage do not always align with memory usage. For less critical or specialized workloads, disabling short query acceleration allows better resource control and error handling for simple tasks.

While one simple task uses minimal resources, many such tasks together consume significant amounts. When short query acceleration is active, managing concurrency—limiting the number of simultaneous queries—helps reduce resource conflicts and ensures efficiency. Without this management, simple queries bypass resource controls, and exception rules will not apply, potentially impacting overall performance and system capacity.

Memory Management

Each resource pool occupies a certain percentage of memory.

Memory management aims to prevent out of memory (OOM) in a database, isolate the memory of different resource pools, and to control memory usage. Memory is managed from the following aspects:

• Global memory management

  To prevent OOM, set the global memory upper limit (max_process_memory) to a proper value. Global memory management before a query controls memory usage to prevent OOM management. Global memory management during a query prevents errors during query execution.

  • Management before a query

    The service checks the estimated memory usage of a query in the slow queue, and compares it with the actual usage. The estimation will be adjusted if it is smaller than the actual usage. Before a query is executed, the service checks whether the available memory is sufficient for the query. If yes, the query can be executed directly. If no, the query needs to be queued and executed after other queries release resources. Concurrency and memory are managed in this phase.

  • Management during a query

    During a query, the service checks whether the requested memory exceeds a certain limit. If yes, an error will be reported, and memory occupied by the query will be released. Memory, CPUs, storage space, and exception rules are managed in this phase.

• Resource pool memory management

  Resource pool memory management puts a limit on dedicated quotas. A workload queue can only use the memory allocated to it, and cannot use idle memory in other resource pools.

  The resource pool memory is allocated in percentage. The value range is 0 to 100. The value 0 indicates that the resource pool does not perform memory management. The value 100 indicates that the resource pool performs memory management and can use all the global memory.

  The sum of memory percentages allocated to all resource pools cannot exceed 100. Resource pool memory management is performed only before a query in the slow queue starts. It works in a way similar to the global memory management before a query. Before a query in the slow queue in a resource pool is executed, its memory usage is estimated. If the estimation is greater than the resource pool memory, the query needs to be queued and can be executed only after earlier queries in the pool are complete and resources released.

CPU Management

CPU share and CPU limit can be managed.

• CPU share: If the system is heavily loaded, CPU resources are allocated to resource pools based on the specific CPU shares. If the system not busy, this configuration does not take effect.
• CPU limit: It specifies the maximum number of CPU cores used by a resource pool. The resource usage of jobs in the resource pool cannot exceed this limit no matter whether the system is busy or not.

In the resource configuration area, you can modify the CPU time limit and CPU usage limit.

Choose either of the preceding management methods as needed. In CPU share management, CPUs can be shared and fully utilized, but resource pools are not isolated and may affect the query performance of each other. In CPU limit management, the CPUs of different resource pools are isolated, but this may result in the waste of idle resources.