Memory and Resource Management with Resource Groups

    Memory, CPU, and concurrent transaction management have a significant impact on performance in a Greenplum Database cluster. Resource groups are a newer resource management scheme that enforce memory, CPU, and concurrent transaction limits in Greenplum Database.

    While it is not always possible to increase system memory, you can avoid many out-of-memory conditions by configuring resource groups to manage expected workloads.

    The following operating system and Greenplum Database memory settings are significant when you manage Greenplum Database resources with resource groups:

    • vm.overcommit_memory

      This Linux kernel parameter, set in , identifies the method that the operating system uses to determine how much memory can be allocated to processes. vm.overcommit_memory must always be set to 2 for Greenplum Database systems.

    • vm.overcommit_ratio

      This Linux kernel parameter, set in /etc/sysctl.conf, identifies the percentage of RAM that is used for application processes; the remainder is reserved for the operating system. Tune the setting as necessary. If your memory utilization is too low, increase the value; if your memory or swap usage is too high, decrease the setting.

    • The percentage of system memory to allocate to Greenplum Database. The default value is .7 (70%).

    • gp_workfile_limit_files_per_query

      Set gp_workfile_limit_files_per_query to limit the maximum number of temporary spill files (workfiles) allowed per query. Spill files are created when a query requires more memory than it is allocated. When the limit is exceeded the query is terminated. The default is zero, which allows an unlimited number of spill files and may fill up the file system.

    • gp_workfile_compress_algorithm

      If there are numerous spill files then set gp_workfile_compress_algorithm to compress the spill files. Compressing spill files may help to avoid overloading the disk subsystem with IO operations.

    • memory_spill_ratio

      Set memory_spill_ratio to increase or decrease the amount of query operator memory Greenplum Database allots to a query. When memory_spill_ratio is larger than 0, it represents the percentage of resource group memory to allot to query operators. If concurrency is high, this memory amount may be small even when memory_spill_ratio is set to the max value of 100. When you set memory_spill_ratio to 0, Greenplum Database uses the statement_mem setting to determine the initial amount of query operator memory to allot.

    • When memory_spill_ratio is 0, Greenplum Database uses the statement_mem setting to determine the amount of memory to allocate to a query.

    Other considerations:

    • Do not configure the operating system to use huge pages.
    • When you configure resource group memory, consider memory requirements for mirror segments that become primary segments during a failure to ensure that database operations can continue when primary segments or segment hosts fail.

    Available memory for resource groups may be limited on systems that use low or no swap space, and that use the default and gp_resource_group_memory_limit settings. To ensure that Greenplum Database has a reasonable per-segment-host memory limit, you may be required to increase one or more of the following configuration settings:

    1. The system’s vm.overcommit_ratio setting.
    2. The resource group gp_resource_group_memory_limit setting.

    Greenplum Database resource groups provide a powerful mechanism for managing the workload of the cluster. Consider these general guidelines when you configure resource groups for your system:

    • A transaction submitted by any Greenplum Database role with SUPERUSER privileges runs under the default resource group named admin_group. Keep this in mind when scheduling and running Greenplum administration utilities.
    • Ensure that you assign each non-admin role a resource group. If you do not assign a resource group to a role, queries submitted by the role are handled by the default resource group named default_group.
    • Use the CONCURRENCY resource group parameter to limit the number of active queries that members of a particular resource group can run concurrently.
    • Use the MEMORY_LIMIT and MEMORY_SPILL_RATIO parameters to control the maximum amount of memory that queries running in the resource group can consume.
    • Greenplum Database assigns unreserved memory (100 - (sum of all resource group MEMORY_LIMITs) to a global shared memory pool. This memory is available to all queries on a first-come, first-served basis.
    • Alter resource groups dynamically to match the real requirements of the group for the workload and the time of day.
    • Use the gptoolkit views to examine resource group resource usage and to monitor how the groups are working.

    A low setting (for example, in the 10MB range) has been shown to increase the performance of queries with low memory requirements. Use the memory_spill_ratio and statement_mem server configuration parameter to override the setting on a per-query basis. For example:

    The default resource group for database transactions initiated by Greenplum Database SUPERUSERs is the group named admin_group. The default CONCURRENCY value for the admin_group resource group is 10.

    Certain Greenplum Database administrative utilities may use more than one CONCURRENCY slot at runtime, such as gpbackup that you invoke with the --jobs option. If the utility(s) you run require more concurrent transactions than that configured for admin_group, consider temporarily increasing the group’s MEMORY_LIMIT and values to meet the utility’s requirement, making sure to return these parameters back to their original settings when the utility completes.

    Parent topic: