Skip to content

RDS

Monitoring

General Purpose VS Provisioned IOPs

General Purpose VS Provisioned IOPs

Analyse

Amazon RDS with Amazon CloudWatch metrics

Metrics

  • BinLogDiskUsage 
    • Amount of disk space occupied by binary logs on the primary database. Applies to MySQL Read Replicas and is measured in bytes
  • DatabaseConnections
    • Number of database connections in use
  • DiskQueueDepth
    • The number of outstanding read and write requests waiting to access the disk
  • FreeStorageSpace 
    • Amount of available storage space
  • FreeableMemory 
    • Amount of available RAM
  • NetworkReceiveThroughput 
    • Inbound network traffic on the DB instance that includes both customer database traffic and Amazon RDS traffic used for monitoring and replication
  • NetworkTransmitThroughput 
    • Outbound network traffic on the DB instance that includes both customer database traffic and Amazon RDS traffic used for monitoring and replication
  • ReadIOPS/WriteIOPS
    • Average number of disk I/O operations per second
    • Use this to determine storage type changes
  • ReadLatency /WriteLatency
    • Average amount of time taken per disk I/O operation
    • More IOPS needed
  • ReadThroughput/WriteThroughput
    • Average number of bytes read from disk per second
  • ReplicaLag
    • Amount of time a Read Replica DB instance lags behind the source DB instance. Applies to MySQL, MariaDB, and PostgreSQL Read Replicas
  • SwapUsage 
    • Amount of swap space used on the DB instance
    • If increase - low or no available ram
  • High CPU or RAM consumption High values for CPU or RAM consumption might be appropriate, provided that they are within an expected range.
  • Disk space consumption Investigate disk space consumption if there is consistently less than 15 percent of available free space
  • Network traffic Investigate network traffic if throughput is consistently lower than expected
  • Database connections If an increase of usage causes performance issues, consider limiting the number of available database connections. The best number of user connections for a DB instance will depend on the instance class and the complexity of the operations performed.

  • IOPS metrics The expected values for IOPS metrics depend on disk specification and server configuration. Establish a baseline to know what utilization is typical. For best IOPS performance, make sure that the typical working set will fit into memory to minimize read and write operations.

  • Aurora 100% CPU utilization

    • Is it writes causing the issue? if so scale up (increase instance size)
    • Is it reads causing the issue? If so scale out (increase the number of read replicas)