Manage Storage and Database Capacity for MariaDB Xpand

Overview

Xpand monitors the amount of space available within your cluster and proactively warns of potential capacity issues. The thresholds for determining a cluster's capacity are configurable and described below.

Note

For additional information, see "Allocate Disk Space for Fault Tolerance and Availability with MariaDB Xpand" or "In-Memory Tables with MariaDB Xpand" for guidance in allocating memory and configuring alerts.

Types of Storage

To understand how to manage the Device and Database Utilization, one must first understand how Xpand allocates disk space. Xpand creates and allocates space in two different files:

device1 (main storage)

The main device1 storage is used for all database data, undo logs, temporary tables, binlogs, Xpand system tables, as well as temporary storage used for query execution. The initial size of the device1 file is auto-detected by the Xpand installer, but can also be configured manually. Post-installation, the device1 file's size can be extended using ALTER CLUSTER RESIZE DEVICES. To decrease the size of the device1 file, see "Decrease device1 Size for MariaDB Xpand".

Xpand expects the device1 file to be the same size on every node. By default, on database startup, Xpand will automatically attempt to resize the device1 file on each node to match the largest device1 file in the cluster. To disable this feature, set device_auto_resize_to_largest=false.

Temporary storage is used for sorting and grouping of large query results and is stored in device1. There are two global variables to control temp space usage:

  • device_temporary_space_limit_bytes limits the amount of space usable for temporary storage.

  • device_temporary_space_preallocate_bytes specifies the amount of space that will be pre-allocated for temp space (guaranteed for use by temp).

Setting device_temporary_space_limit_bytes allows additional temp space to be used, but does not guarantee additional space will be available for temp. Increasing these values takes effect immediately, while decreasing it takes effect after database restart.

device1-redo (write-ahead log)

The write-ahead log (WAL) is stored in the device1-redo file. This size of this file is 4GB and is not configurable.

Checking Storage Utilization

See how much space is in use by using clx command-line administration tool

shell> /opt/clustrix/bin/clx space
nid |   Hostname   | Status  |       Undo      |       Perm      |       WAL        |    Temp    |       Used      | DB Total | FS Free
----+--------------+---------+-----------------+-----------------+------------------+------------+-----------------+----------+--------
 16 |  eukanuba003 |    OK   |  321.8M (0.04%) |  674.7G (79.4%) |  1024.0M (0.12%) |  0 (0.00%) |  760.1G (89.4%) |   850.0G |  113.9G
 17 |  karma183    |    OK   |  313.5M (0.04%) |  664.6G (78.2%) |  1024.0M (0.12%) |  0 (0.00%) |  750.1G (88.2%) |   850.0G |  113.9G
 18 |  eukanuba002 |    OK   |  324.3M (0.04%) |  669.5G (78.8%) |  1024.0M (0.12%) |  0 (0.00%) |  755.0G (88.8%) |   850.0G |  113.9G
 19 |  eukanuba001 |    OK   |  339.7M (0.04%) |  671.0G (78.9%) |  1024.0M (0.12%) |  0 (0.00%) |  756.4G (89.0%) |   850.0G |  113.9G
 20 |  eukanuba005 |    OK   |  277.3M (0.03%) |  668.7G (78.7%) |  1024.0M (0.12%) |  0 (0.00%) |  754.1G (88.7%) |   850.0G |  113.9G
 21 |  eukanuba004 |    OK   |  420.3M (0.05%) |  678.6G (79.8%) |  1024.0M (0.12%) |  0 (0.00%) |  764.1G (89.9%) |   850.0G |  113.9G
 22 |  eukanuba006 |    OK   |  397.0M (0.05%) |  670.4G (78.9%) |  1024.0M (0.12%) |  0 (0.00%) |  755.9G (88.9%) |   850.0G |  113.9G
 23 |  karma184    |    OK   |  479.9M (0.06%) |  674.8G (79.4%) |  1024.0M (0.12%) |  0 (0.00%) |  760.3G (89.5%) |   850.0G |  113.9G
----+--------------+---------+-----------------+-----------------+------------------+------------+-----------------+----------+--------
                                  2.8G (0.04%) |    5.2T (79.0%) |     8.0G (0.12%) |  0 (0.00%) |    5.9T (89.1%) |     6.6T |  910.9G

Global Variables

The default values for these global variables are optimal for most workloads.

Variable

Description

Default

device_auto_resize_to_largest

Automatically resize all (online) devices in the cluster to match the largest device

true

device_temporary_space_limit_bytes

Maximum number of bytes allowed to be used for temporary containers.

5368709120

device_temporary_space_preallocate_bytes

The amount of space that will be pre-allocated for temporary storage

5368709120

Database Storage Thresholds

Global variables establish the database storage thresholds for a cluster. When the first level of thresholds are exceeded, alerts are sent. If storage utilization continues to increase, user queries will begin to fail once the next set of thresholds are exceeded. Finally, if storage utilization continues to grow, system queries (including for critical internal processes) will be killed. Once the database is completely full, the database may become inoperable. For suggestions on freeing space, see "Resolving Low Space Issues" below.

Variables

The following variables are use to set thresholds for device1 utilization.

Variable

Description

Default Value

Allowed Values

databasefull_message_interval_s

Database almost full message interval in seconds.

120

Minimum: 10

Maximum: 600

databasefull_user_warn_percentage

Warn about user queries when space usage surpasses this percentage.

80

Minimum: 50

Maximum: databasefull_user_error_percentage - 1

databasefull_user_error_percentage

Fail user queries when space usage surpasses this percentage.

90

Minimum: databasefull_user_warn_percentage + 1

Maximum: databasefull_system_warn_percentage - 1

databasefull_system_warn_percentage

Warn about system queries when space usage surpasses this percentage.

95

Minimum: databasefull_user_error_percentage + 1

Maximum: databasefull_system_error_percentage - 1

databasefull_system_error_percentage

Fail user queries when space usage surpasses this percentage.

97

Minimum: databasefull_system_warn_percentage + 1

Maximum: >99

User queries are transactions which originate with an end user whereas system queries are internal Xpand processes such the Rebalancer, binlog deletes, statd, etc.

Alert Messages

The following alerts are triggered when the corresponding global variable is exceeded. This is evaluated each time Xpand allocates space and any alerts necessary are sent every databasefull_message_interval_s seconds. If multiple alerts are detected, only the most critical will appear. For additional information about Xpand's Alerter, see "Email Alerts for MariaDB Xpand".

Global Variable Evaluated

Alert Triggered

Level

Description

Message Shown

databasefull_user_warn_percentage

DATABASE_SPACE_LOW

warning

Database space low

Database space is nn% used. Soon user queries will fail.

databasefull_user_error_percentage

DATABASE_SPACE_EXTREME

warning

Database space extreme

Database space is nn% used. User queries will now fail.

databasefull_system_warn_percentage

DATABASE_SPACE_CRITICAL

critical

Database space critical

Database space is nn% used. User queries will fail, and soon system queries will fail.

databasefull_system_error_percentage

DATABASE_SPACE_EXHAUSTED

critical

Database space exhausted

Database space is nn% used. User queries and system queries will now fail.

Resolving Low Space Issues

When you receive any of the alerts above, some action will be necessary to prevent the capacity of device1 from reaching the next threshold.

Some resolutions to consider:

  • Add nodes to the cluster by Scaling Out.

  • Increase available space on the cluster by:

  • Enlarge the size of the device1 file on all nodes by using ALTER CLUSTER RESIZE DEVICES.

  • Terminate and reschedule long running transactions such as ALTERs, Backups, and long-running transactions. These halt garbage collection and cause the undo log to temporarily grow in size.

If you need assistance, please contact MariaDB Support.