Skip to content

Tencent Cloud MariaDB

Use the 「Guance Synchronization」 series of script packages in the script market to synchronize data from cloud monitoring cloud assets to the observation cloud.

Config

Install Func

Recommend opening 「Integrations - Extension - DataFlux Func (Automata)」: All preconditions are installed automatically, Please continue with the script installation

If you deploy Func yourself,Refer to Self-Deployment of Func

Installation script

Tip:Please prepare Tencent AK that meets the requirements in advance(For simplicity's sake,,You can directly grant the global read-only permissionReadOnlyAccess

To synchronize the monitoring data of Tencent MariaDB Private resources, we install the corresponding collection script:「Guance Integration(Tencent Cloud-MariaDBCollect)」(ID:guance_tencentcloud_mariadb)

Click 【Install】 and enter the corresponding parameters: Tencent AK, Tencent account name.。

tap【Deploy startup Script】,The system automatically creates Startup script sets,And automatically configure the corresponding startup script。

In addition, the corresponding automatic trigger configuration is displayed in「Management / Crontab Config」. tap【Run】,It can be executed immediately once, without waiting for a periodic time。After a while, you can view task execution records and corresponding logs.

We collected some configurations by default, as described in the Metrics column Configure custom cloud object metrics

Verify

  1. In「Management / Crontab Config」check whether the automatic triggering configuration exists for the corresponding task,In addition, you can view task records and logs to check whether exceptions exist
  2. On the observation cloud platform, click 「Infrastructure / Custom」 to check whether asset information exists
  3. On the observation cloud platform, press 「Metrics」 to check whether monitoring data exists

Metric

Configure Tencent Cloud - cloud monitoring. The default metric set is as follows. You can collect more metrics by configuring them Tencent Cloud Monitor Metrics Details

Index English name Index Chinese name Metric specification unit Statistical granularity
ActiveThreadCount Active Thread Count Instance-level monitoring metric, calculated as the sum of active thread counts across all primary and secondary nodes in the shards. indivual InstanceId
BinlogDiskAvailable Remaining Binlog Disk Space Instance-level monitoring metric, calculated as the sum of the monitored values of the BinlogDiskAvailableShard metric across all shards. GB InstanceId
BinlogUsedDisk Used Binlog Disk Space Instance-level monitoring metric, calculated as the sum of the used Binlog disk space of the primary nodes across all shards. GB InstanceId
ConnUsageRate DB Connection Usage Rate Instance-level monitoring metric, taking the maximum value of the DB connection usage rate across all primary and standby nodes of the instance. % InstanceId
CpuUsageRate CPU Utilization Rate Instance-level monitoring metric, taking the maximum value of the CPU usage rate across all primary nodes of the instance. % InstanceId
DataDiskAvailable Available Data Disk Space Instance-level monitoring metric, calculated by summing up the available data disk space of all primary nodes across various shards. GB InstanceId
DataDiskUsedRate Data Disk Space Utilization Instance-level monitoring metric, the value is the maximum data disk space utilization rate among the primary nodes of each shard in the instance. % InstanceId
DeleteTotal DELETE Request Count Instance-level monitoring metric, calculated by summing up the DELETE request count from the primary nodes of each shard in the instance. Times/second InstanceId
InnodbBufferPoolReads innodb Disk Read Page Count Instance-level monitoring metric, calculated by summing up the InnoDB disk read page count from the primary and standby nodes of each shard in the instance. times InstanceId
InnodbBufferPoolReadAhead innodb Buffer Pool Read Page Count Instance-level monitoring metric, calculated by summing up the InnoDB buffer pool pre-read page count from the primary and standby nodes of each shard in the instance. times InstanceId
InnodbBufferPoolReadRequests innodb Buffer Pool Read Page Count Instance-level monitoring metric, calculated by summing up the InnoDB buffer pool read page count from the primary and standby nodes of each shard in the instance. times InstanceId
InnodbRowsDeleted InnoDB Delete Rows Count Instance-level monitoring metric, calculated by summing up the InnoDB DELETE rows executed count from the primary nodes of each shard in the instance. rows InstanceId
InnodbRowsInserted InnoDB Insert Rows Count Instance-level monitoring metric, calculated by summing up the InnoDB INSERT rows executed count from the primary nodes of each shard in the instance. rows InstanceId
InnodbRowsRead InnoDB Read Rows Count Instance-level monitoring metric, calculated by summing up the InnoDB READ rows executed count from the primary and standby nodes of each shard in the instance. rows InstanceId
InnodbRowsUpdated InnoDB Update Rows Count Instance-level monitoring metric, calculated by summing up the InnoDB UPDATE rows executed count from the primary nodes of each shard in the instance. rows InstanceId
InsertTotal Number of INSERT Requests Instance-level monitoring metric, calculated by summing up the INSERT request count from the primary nodes of each shard in the instance. Times/second InstanceId
LongQueryCount Number of Slow Queries Instance-level monitoring metric, calculated by summing up the slow query count from the primary nodes of each shard in the instance. times InstanceId
MemAvailable Available Cache Space Instance-level monitoring metric, calculated by summing up the available cache space from the primary nodes of each shard in the instance. GB InstanceId
MemHitRate Cache Hit Rate Instance-level monitoring metric, the value of which is the minimum cache hit rate among the primary nodes of each shard in the instance. % InstanceId
ReplaceSelectTotal REPLACE_SELECT Request Count Instance-level monitoring metric, calculated by summing up the REPLACE-SELECT request counts from the primary nodes of each shard in the instance. Times/second InstanceId
ReplaceTotal REPLACE Request Count Instance-level monitoring metric, calculated by summing up the REPLACE request counts from the primary nodes of each shard in the instance. Times/second InstanceId
RequestTotal Total Request Count Instance-level monitoring metric, calculated by summing up the total request counts from all primary nodes and the SELECT request counts from all secondary nodes in the instance. Times/second InstanceId
SelectTotal SELECT Request Count Instance-level monitoring metric, calculated by summing up the SELECT request counts from all primary and secondary nodes across all shards in the instance. Times/second InstanceId
SlaveDelay Secondary Lag Instance-level monitoring metric, calculated by first calculating the replication delay for each shard's replicas and then taking the maximum value among them as the replication delay for the instance. The replication delay for a shard is determined by finding the minimum delay among all replicas in that shard. second InstanceId
UpdateTotal Number of UPDATE Requests Instance-level monitoring metric, calculated by summing up the number of UPDATE requests across all primary nodes of the shards in the instance. Times/second InstanceId
ThreadsConnected Current Open Connections Instance-level monitoring metric, calculated by summing up the current number of open connections across all primary and replica nodes of the shards in the instance. times InstanceId
ConnMax Maximum Connections Instance-level monitoring metric, calculated by summing up the maximum connection limit across all primary and replica nodes of the shards in the instance. indivual InstanceId
ClientConnTotal Total Client Connections Instance-level monitoring metric, calculated by summing up all connections on the instance's proxy. This metric accurately shows how many clients are connected to the database instance. indivual InstanceId
SQLTotal Total SQL Statements Instance-level monitoring metric that indicates how many SQL statements are sent to the database instance. items InstanceId
ErrorSQLTotal SQL Error Count Instance-level monitoring metric that indicates how many SQL statements are executed with errors. items InstanceId
SuccessSQLTotal SQL Success Count Instance-level monitoring metric that represents the number of successfully executed SQL statements. indivual InstanceId
TimeRange0 Number of Requests with Latency (<5ms) Instance-level monitoring metric that represents the number of requests with execution time less than 5 ms. Times/second InstanceId
TimeRange1 Number of Requests with Latency (5-20ms) Instance-level monitoring metric that represents the number of requests with execution time between 5 and 20 ms. Times/second InstanceId
TimeRange2 Number of Requests with Latency (20~30ms) Instance-level monitoring metric that represents the number of requests with execution time between 20 and 30 ms. Times/second InstanceId
TimeRange3 Number of Requests with Latency (>30ms) Instance-level monitoring metric that represents the number of requests with execution time greater than 30 ms. Times/second InstanceId
MasterSwitchedTotal Number of Master-Slave Switches Instance-level monitoring metric that represents the number of times master-slave switch occurs in the instance. Times InstanceId
IOUsageRate IO Utilization Instance-level monitoring metric, which is the maximum value of I/O utilization across all primary nodes of the shards in the instance. % InstanceId
MaxSlaveCpuUsageRate Maximum Secondary Node CPU Utilization Instance-level monitoring metric, which is the maximum value of CPU utilization among all backup nodes in the instance. % InstanceId
ThreadsRunningCount Sum of Running Threads Instance-level monitoring metric, calculated as the cumulative value of the Threads_running number from all nodes in the instance. Threads_running is obtained by executing the command "show status like 'Threads_running'". indivual InstanceId

Object

The collected Tencent Cloud CLB Private object data structure can be seen from the「Infrastructure-Custom」object data

{
  "measurement": "tencentcloud_mariadb",
  "tags": {
    "AppId": "1311xxx185",
    "AutoRenewFlag" : "0",
    "DbEngine"      : "MariaDB",
    "DbVersion"     : "10.1",
    "DbVersionId"   : "10.1",
    "InstanceId"    : "tdsql-ewqokixxxxxhu",
    "InstanceName"  : "tdsql-ewqoxxxxxxihu",
    "InstanceType"  : "2",
    "Paymode"       : "postpaid",
    "ProjectId"     : "0",
    "RegionId"      : "ap-shanghai",
    "Status"        : "0",
    "StatusDesc"    : "Creating",
    "TdsqlVersion"  : "Designed based on MariaDB 10.1 (compatible with MySQL 5.6)",
    "Uin"           : "100xxxx113474",
    "Vip"           : "",
    "Vport"         : "3306",
    "WanDomain"     : "",
    "WanPort"       : "0",
    "WanVip"        : "",
    "Zone"          : "ap-shanghai-5",
    "name"          : "tdsql-ewqokihu",
    "WanVIP"        : ""
  },
  "fields": {
    "Cpu"           : 1,
    "CreateTime"    : "2023-08-17 17:55:03",
    "Memory"        : 2,
    "NodeCount"     : 2,
    "PeriodEndTime" : "0001-01-01 00:00:00",
    "Qps"           : 2100,
    "Storage"       : 10,
    "UpdateTime"    : "2023-08-17 17:55:03",
    "message"       : "{Instance JSON data}"
  }
}

Note: Fields in tags and fields may change in subsequent updates.

Tip 1: The value of tags.name serves as a unique identifier.

Tip 2:fields.messagefields.InstanceNode are JSON serialized strings.

Feedback

Is this page helpful? ×