MapR 5.0 Documentation : Monitoring Performance

You can examine fine-grained analytics information about the nodes in your cluster by using the Node Metrics API with the maprcli command-line tool. You can use this information to examine specific aspects of your node's performance at a very granular level. The node metrics API returns data as a table sent to your terminal's standard output or as a JSON file. The JSON file includes in-depth reports on the activity on each CPU, disk partition, and network interface in your node.

The hoststats service writes raw metrics data to the /var/mapr/local/<hostname>/metrics volumes as hourly files. The files rotate as specified by the metric.file.rotate parameter in the db.conf file.

The Metrics database holds data in the following tables:

  • METRIC_TRANSACTION: Data is written every 10 seconds and partitioned by day. This table retains up to three days of data.
  • METRIC_TRANSACTION_EVENT: Holds data regarding system events such as service starts/stops and disk additions.
  • METRIC_TRANSACTION_SUMMARY_DAILY: Holds the aggregate, maximum, and minimum values for five-minute intervals of the data in the METRIC_TRANSACTION table. This table is partitioned by day and holds up to 15 days' worth of data.
  • METRIC_TRANSACTION_SUMMARY_YEARLY: Holds the aggregate, maximum, and minimum values for daily intervals of the data in the METRIC_TRANSACTION_SUMMARY_DAILY table. This table is partitioned by year and holds up to 100 years' worth of data.

Retention of Node Metrics Data

You can purge node metrics data from the database without affecting the cluster's stability. This data will not be restored after purging.

Node metrics cover the following general categories:

  • CPU time used
  • Memory used
  • RPC activity
  • Process activity
  • Storage used
  • TaskTracker resources used