Metric Collection

Metrics are collected from each node in the cluster so that administrators can use the data to monitor the cluster. In general, the collectd service collects metrics every 10 seconds. The exception is volume metrics which are collected every 10 minutes.

When collectd writes metrics to streams, tags are assigned to each metric so that administrators can filter metric data to create dashboards that are specific to their needs.

By default, each metric contains the following tags:

fqdn: Displays values for a specified node.
clusterid: Displays values for a specific cluster.
clustername: As of MEP 3.0, displays values for a specific cluster.

However, many metrics have additional tags that you can use to filter metric data.

Streams store metrics in OpenTSDB with the following schema:

<metrictype.name> <fqdn:fqdnvalue> <clusterid:clusteridvalue> <clustername:clusternamevalue>[<AdditionalTagA:AdditionalTagAvalue> <AdditionalTagB:AdditionalTagBvalue>...] <metricvalue> <timestamp>

Note: A negative value shown in the metrics indicates that the maximum value configured for that metric is exceeded. The maximum value for GUT metrics is int32 (2^31-1).

For more information on using tags and dashboards, see Metric Visualization.

Configure Metric Retention
By default, OpenTSDB stores two weeks of metrics. Based on your requirements, you can change metric retention period.
Configure Queue Filters for mapr.rm.<value> Metrics
The YARN application metrics that are collected by JMX have the metric name syntax mapr.rm.<metric_name> and the metric values are aggregated among all the queues in the default queue. However, you can configure collectd to create a filter for each queue. As an alternative, you can use the REST API queue metrics (mapr.rm_queue.<metric_name>) which are by default set up for filtering by queue.
Configure the Collectd Service Heap Size
The collectd service uses an embedded JVM when it gathers metrics from the CLDB, Node Manager, Resource Manager, and Drill. You can edit the Plugin Java section of collectd.conf to configure limits to the collectd virtual memory footprint.
CPU Metrics
Every 10 seconds, the collectd service uses the cpu plugin to gather the following CPU metrics on each node in the cluster.
Disk Free Metrics
Every 10 seconds, the collectd service uses the df plugin to gather the following disk free metrics on each node in the cluster.
Disk Metrics
Every 10 seconds, the collectd service uses the disk plugin to gather the following disk metrics on each node in the cluster.
Drill Metrics
Every 10 seconds, the collectd service uses the plugin to gather the following Drill metrics on each node in the cluster.
Hive JMX Metrics
Every 10 seconds, the collectd service uses the Hive plug-in to gather the following Hive JMX metrics on each node in the cluster. Descriptions for the Hive metrics are not currently available.
Load Metrics
Every 10 seconds, the collectd service uses the load plugin to gather the following load metrics on each node in the cluster.
MapR Alarm Metrics
Every 10 seconds, the collectd service uses a plugin to gather the cluster alarms.
MapR Cache Metrics
Every 10 seconds, the collectd service uses a plugin to gather the following MapR filesystem cache metrics on each node in the cluster.
MapR CLDB Metrics
Every 10 seconds, the collectd service uses a HPE Data Fabric plugin to gather the following CLDB metrics on the master CLDB node in the cluster.
HPE Ezmeral Data Fabric Database Metrics
Every 10 seconds, the collectd service uses a plugin to gather HPE Ezmeral Data Fabric Database metrics on each node in the cluster. HPE Ezmeral Data Fabric Database provides both node and table metrics.
HPE Ezmeral Data Fabric Event Store Metrics
Every 10 seconds, the collectd service uses a plugin to gather the following Streams metrics on each node in the cluster.
filesystem Metrics
Every 10 seconds, the collectd service uses a HPE Data Fabric plugin to gather the following filesystem metrics on each node in the cluster.
MapR Process Metrics
Every 10 seconds, the collectd service uses a plugin to gather the following MapR process metrics on each node in the cluster.
MapR I/O Metrics
Every 10 seconds, the collectd service uses a plugin to gather the following MapR I/O metrics on each node in the cluster.
MapR RPC Metric
Every 10 seconds, the collectd service uses a plugin to gather the following MapR RPC metrics on each node in the cluster.
Spark JMX Metrics
Every 10 seconds, the collectd service gathers the following Spark JMX metrics on each node in the cluster.
MapR Topology Metrics
Every 60 seconds, the collectd service uses a plugin to gather the following topology metrics on each node in the cluster. Use these metrics to understand disk utilization across a topology or rack. By default, these metrics include all racks and topologies associated with the cluster. However, you can use tags to specify which rack(s) or topologies(s) to include. Note: Racks and topologies can span multiple nodes and one rack can be associated with multiple topologies.
MapR Volume Metrics
Every 10 seconds, the collectd service uses a plugin to gather the following MapR volume metrics on each CLDB node in the cluster.
Virtual Memory Metrics
Every 10 seconds, the collectd service uses the vmem plugin to gather the following memory metrics on each node in the cluster.
Memory Metrics
Every 10 seconds, the collectd service uses the memory and swap plugins to gather the following memory metrics on each node in the cluster.
Network Metrics
Every 10 seconds, the collectd service uses the interface plugin to gather network metrics on each node in the cluster.
Node Manager Metrics
Every 10 seconds, the collectd service uses a plugin to gather the following Node Manager metrics on each node in the cluster.
Resource Manager Metrics
Every 10 seconds, the collectd service uses a HPE Ezmeral Data Fabric plugin to gather Resource Manager metrics on the active Resource Manager. Collectd gathers metrics on the Resource Manager JVM process, YARN applications, and nodes that are managed by the Resource Manager. The method used to gather the metrics differs based on the metric type.

Parent topic: Using HPE Ezmeral Data Fabric Monitoring (Spyglass Initiative)