Metrics Reporting

Solr includes a developer API and instrumentation for the collection of detailed performance-oriented metrics throughout the life-cycle of Solr service and its various components.

Internally this feature uses the Dropwizard Metrics API, which uses the following classes of meters to measure events:

  • counters - simply count events. They provide a single long value, e.g., the number of requests.

  • meters - additionally compute rates of events. Provide a count (as above) and 1-, 5-, and 15-minute exponentially decaying rates, similar to the Unix system load average.

  • histograms - calculate approximate distribution of events according to their values. Provide the following approximate statistics, with a similar exponential decay as above: mean (arithmetic average), median, maximum, minimum, standard deviation, and 75th, 95th, 98th, 99th and 999th percentiles.

  • timers - measure the number and duration of events. They provide a count and histogram of timings.

  • gauges - offer instantaneous reading of a current value, e.g., current queue depth, current number of active connections, free heap size.

Each group of related metrics with unique names is managed in a metric registry. Solr maintains several such registries, each corresponding to a high-level group such as: jvm, jetty, node, and core (see Metric Registries below).

For each group (and/or for each registry) there can be several reporters, which are components responsible for communication of metrics from selected registries to external systems. Currently implemented reporters support emitting metrics via JMX, Ganglia, Graphite and SLF4J.

There is also a dedicated /admin/metrics handler that can be queried to report all or a subset of the current metrics from multiple registries.

Metric Registries

Solr includes multiple metric registries, which group related metrics.

Metrics are maintained and accumulated through all lifecycles of components from the start of the process until its shutdown - e.g., metrics for a particular SolrCore are tracked through possibly several load, unload and/or rename operations, and are deleted only when a core is explicitly deleted. However, metrics are not persisted across process restarts; restarting Solr will discard all collected metrics.

These are the major groups of metrics that are collected:

JVM Registry

This registry is returned at solr.jvm and includes the following information. When making requests with the Metrics API, you can specify &group=jvm to limit to only these metrics.

  • direct and mapped buffer pools

  • class loading / unloading

  • OS memory, CPU time, file descriptors, swap, system load

  • GC count and time

  • heap, non-heap memory and GC pools

  • number of threads, their states and deadlocks

  • System properties such as Java information, various installation directory paths, ports, and similar information. You can control what appears here by modifying solr.xml.

Node / CoreContainer Registry

This registry is returned at solr.node and includes the following information. When making requests with the Metrics API, you can specify &group=node to limit to only these metrics.

  • handler requests (count, timing): collections, info, admin, configSets, etc.

  • number of cores (loaded, lazy, unloaded)

Core (SolrCore) Registry

The Core (SolrCore) Registry includes solr.core.<collection>, one for each core. When making requests with the Metrics API, you can specify &group=core to limit to only these metrics.

  • all common RequestHandlers report: request timers / counters, timeouts, errors. Handlers that support process distributed shard requests also report shardRequests sub-counters for each type of distributed request.

  • index-level events: meters for minor / major merges, number of merged docs, number of deleted docs, gauges for currently running merges and their size.

  • shard replication and transaction log replay on replicas (TBD, SOLR-9856)

  • open / available / pending connections for shard handler and update handler

Jetty Registry

This registry is returned at solr.jetty and includes the following information. When making requests with the Metrics API, you can specify &group=jetty to limit to only these metrics.

  • threads and pools,

  • connection and request timers,

  • meters for responses by HTTP class (1xx, 2xx, etc.)

In the future, metrics will be added for shard leaders and cluster nodes, including aggregations from per-core metrics.

Metrics Configuration

The metrics available in your system can be customized by modifying the <metrics> element in solr.xml.

See also the section Format of Solr.xml for more information about the solr.xml file, where to find it, and how to edit it.

The <metrics><hiddenSysProps> Element

This section of solr.xml allows you to define the system properties which are considered system-sensitive and should not be exposed via the Metrics API.

If this section is not defined, the following default configuration is used which hides password and authentication information:

<metrics>
  <hiddenSysProps>
    <str>javax.net.ssl.keyStorePassword</str>
    <str>javax.net.ssl.trustStorePassword</str>
    <str>basicauth</str>
    <str>zkDigestPassword</str>
    <str>zkDigestReadonlyPassword</str>
  </hiddenSysProps>
</metrics>

The <metrics><reporters> Element

Reporters consume the metrics data generated by Solr. See the section Reporters below for more details on how to configure custom reporters.

The <metrics><suppliers> Element

Suppliers help Solr generate metrics data. The <metrics><suppliers> section of solr.xml allows you to define your own implementations of metrics and configure parameters for them.

Implementation of a custom metrics supplier is beyond the scope of this guide, but there are other customizations possible with the default implementation, via the elements described below.

<counter>

This element defines the implementation and configuration of a Counter supplier. The default implementation does not support any configuration.

<meter>

This element defines the implementation of a Meter supplier. The default implementation supports an additional parameter:

<str name="clock">

The type of clock to use for calculating EWMA rates. The supported values are:

  • user, the default, which uses System.nanoTime()

  • cpu, which uses the current thread’s CPU time

<histogram>

This element defines the implementation of a Histogram supplier. This element also supports the clock parameter shown above with the meter element, and also:

<str name="reservoir">

The fully-qualified class name of the Reservoir implementation to use. The default is com.codahale.metrics.ExponentiallyDecayingReservoir but there are other options available with the Codahale Metrics library that Solr uses. The following parameters are supported, within the mentioned limitations:

  • size, the reservoir size. The default is 1028.

  • alpha, the decay parameter. The default is 0.015. This is only valid for the ExponentiallyDecayingReservoir.

  • window, the window size, in seconds, and only valid for the SlidingTimeWindowReservoir. The default is 300 (5 minutes).

<timer>

This element defines an implementation of a Timer supplier. The default implementation supports the clock and reservoir parameters described above.

As an example of a section of solr.xml that defines some of these custom parameters, the following defines the default Meter supplier with a non-default clock and the default Timer is used with a non-default reservoir:

<metrics>
  <suppliers>
    <meter>
      <str name="clock">cpu</str>
    </meter>
    <timer>
      <str name="reservoir">com.codahale.metrics.SlidingTimeWindowReservoir</str>
      <long name="window">600</long>
    </timer>
  </suppliers>
</metrics>

Reporters

Reporter configurations are specified in solr.xml file in <metrics><reporter> sections, for example:

<solr>
 <metrics>
  <reporter name="graphite" group="node, jvm" class="org.apache.solr.metrics.reporters.SolrGraphiteReporter">
    <str name="host">graphite-server</str>
    <int name="port">9999</int>
    <int name="period">60</int>
  </reporter>
  <reporter name="collection1Updates" registry="solr.core.collection1" class="org.apache.solr.metrics.reporters.SolrSlf4jReporter">
    <int name="period">300</int>
    <str name="prefix">example</str>
    <str name="logger">updatesLogger</str>
    <str name="filter">QUERYHANDLER./update</str>
  </reporter>
 </metrics>
...
</solr>

This example configures two reporters: Graphite and SLF4J. See below for more details on how to configure reporters.

Reporter Arguments

Reporter plugins use the following arguments:

name

The unique name of the reporter plugin (required).

class

The fully-qualified implementation class of the plugin, which must extend SolrMetricReporter (required).

group

One or more of the predefined groups (see above).

registry

One or more of valid fully-qualified registry names.

If both group and registry attributes are specified only the group attribute is considered. If neither attribute is specified then the plugin will be used for all groups and registries. Multiple group or registry names can be specified, separated by comma and/or space.

Additionally, several implementation-specific initialization arguments can be specified in nested elements. There are some arguments that are common to SLF4J, Ganglia and Graphite reporters:

period

The period in seconds between reports. Default value is 60.

prefix

A prefix to be added to metric names, which may be helpful in logical grouping of related Solr instances, e.g., machine name or cluster name. Default is empty string, i.e., just the registry name and metric name will be used to form a fully-qualified metric name.

filter

If not empty then only metric names that start with this value will be reported. Default is no filtering, i.e., all metrics from the selected registry will be reported.

Reporters are instantiated for every group and registry that they were configured for, at the time when the respective components are initialized (e.g., on JVM startup or SolrCore load).

When reporters are created their configuration is validated (and e.g., necessary connections are established). Uncaught errors at this initialization stage cause the reporter to be discarded from the running configuration.

Reporters are closed when the corresponding component is being closed (e.g., on SolrCore close, or JVM shutdown) but metrics that they reported are still maintained in respective registries, as explained in the previous section.

The following sections provide information on implementation-specific arguments. All implementation classes provided with Solr can be found under org.apache.solr.metrics.reporters.

JMX Reporter

The JMX Reporter uses the org.apache.solr.metrics.reporters.SolrJmxReporter class.

It takes the following arguments:

domain

The JMX domain name. If not specified then the registry name will be used.

serviceUrl

The service URL for a JMX server. If not specified, Solr will attempt to discover if the JVM has an MBean server and will use that address. See below for additional information on this.

agentId

The agent ID for a JMX server. Note either serviceUrl or agentId can be specified but not both - if both are specified then the default MBean server will be used.

Object names created by this reporter are hierarchical, dot-separated but also properly structured to form corresponding hierarchies in e.g., JConsole. This hierarchy consists of the following elements in the top-down order:

  • registry name (e.g., solr.core.collection1.shard1.replica1). Dot-separated registry names are also split into ObjectName hierarchy levels, so that metrics for this registry will be shown under /solr/core/collection1/shard1/replica1 in JConsole, with each domain part being assigned to dom1, dom2, …​ domN property.

  • reporter name (the value of reporter’s name attribute)

  • category, scope and name for request handlers

  • or additional name1, name2, …​ nameN elements for metrics from other components.

The JMX Reporter replaces the JMX functionality available in Solr versions before 7.0. If you have upgraded from an earlier version and have an MBean Server running when Solr starts, Solr will automatically discover the location of the local MBean server and use a default configuration for the SolrJmxReporter.

You can start a local MBean server with a system property at startup by adding -Dcom.sun.management.jmxremote to your start command. This will not add the reporter configuration to solr.xml, so if you enable it with a system property, you must always start Solr with the system property or JMX will not be enabled in subsequent starts.

SLF4J Reporter

The SLF4J Reporter uses the org.apache.solr.metrics.reporters.SolrSlf4jReporter class.

It takes the following arguments, in addition to the common arguments above.

logger

The name of the logger to use. Default is empty, in which case the group or registry name will be used if specified in the plugin configuration.

Users can specify logger name (and the corresponding logger configuration in e.g., Log4j configuration) to output metrics-related logging to separate file(s), which can then be processed by external applications.

Each log line produced by this reporter consists of configuration-specific fields, and a message that follows this format:

type=COUNTER, name={}, count={}

type=GAUGE, name={}, value={}

type=TIMER, name={}, count={}, min={}, max={}, mean={}, stddev={}, median={}, p75={}, p95={}, p98={}, p99={}, p999={}, mean_rate={}, m1={}, m5={}, m15={}, rate_unit={}, duration_unit={}

type=METER, name={}, count={}, mean_rate={}, m1={}, m5={}, m15={}, rate_unit={}

type=HISTOGRAM, name={}, count={}, min={}, max={}, mean={}, stddev={}, median={}, p75={}, p95={}, p98={}, p99={}, p999={}

(curly braces added only as placeholders for actual values).

Graphite Reporter

The Graphite Reporter uses the org.apache.solr.metrics.reporters.SolrGraphiteReporter) class.

It takes the following attributes, in addition to the common attributes above.

host

The host name where Graphite server is running (required).

port

The port number for the server (required).

pickled

If true, use "pickled" Graphite protocol which may be more efficient. Default is false (use plain-text protocol).

When plain-text protocol is used (pickled==false) it’s possible to use this reporter to integrate with systems other than Graphite, if they can accept space-separated and line-oriented input over network in the following format:

dot.separated.metric.name[.and.attribute] value epochTimestamp

For example:

example.solr.node.cores.lazy 0 1482932097
example.solr.node.cores.loaded 1 1482932097
example.solr.jetty.org.eclipse.jetty.server.handler.DefaultHandler.2xx-responses.count 21 1482932097
example.solr.jetty.org.eclipse.jetty.server.handler.DefaultHandler.2xx-responses.m1_rate 2.5474287707930614 1482932097
example.solr.jetty.org.eclipse.jetty.server.handler.DefaultHandler.2xx-responses.m5_rate 3.8003171557510305 1482932097
example.solr.jetty.org.eclipse.jetty.server.handler.DefaultHandler.2xx-responses.m15_rate 4.0623076220244245 1482932097
example.solr.jetty.org.eclipse.jetty.server.handler.DefaultHandler.2xx-responses.mean_rate 0.5698031798408144 1482932097

Ganglia Reporter

The Ganglia reporter uses the org.apache.solr.metrics.reporters.SolrGangliaReporter class.

It take the following arguments, in addition to the common arguments above.

host

The host name where Ganglia server is running (required).

port

The port number for the server.

multicast

When true use multicast UDP communication, otherwise use UDP unicast. Default is false.

Shard and Cluster Reporters

These two reporters can be used for aggregation of metrics reported from replicas to shard leader (the "shard" reporter), and from any local registry to the Overseer node.

Metric reports from these reporters are periodically sent as batches of regular SolrInputDocuments, so they can be processed by any Solr handler. By default they are sent to /admin/metrics/collector handler (an instance of MetricsCollectorHandler) on a target node, which aggregates these reports and keeps them in additional local metric registries so that they can be accessed using /admin/metrics handler, and re-reported elsewhere as necessary.

In case of shard reporter the target node is the shard leader, in case of cluster reporter the target node is the Overseer leader.

Shard Reporter

This reporter uses predefined shard group, and the implementing class must be (a subclass of) solr.SolrShardReporter. It publishes selected metrics from replicas to the node where shard leader is located. Reports use a target registry name that is the replica’s registry name with a .leader suffix, e.g., for a SolrCore name collection1_shard1_replica_n3 the target registry name is solr.core.collection1.shard1.replica_n3.leader.

The following configuration properties are supported:

handler

The handler path where reports are sent. Default is /admin/metrics/collector.

period

How often reports are sent, in seconds. Default is 60. Setting this to 0 disables the reporter.

filter

An optional regular expression(s) matching selected metrics to be reported.

The following filter expressions are used by default:

TLOG.*
CORE\.fs.*
REPLICATION.*
INDEX\.flush.*
INDEX\.merge\.major.*
UPDATE\./update/.*requests
QUERY\./select.*requests

Example configuration:

      <reporter name="test" group="shard" class="solr.SolrShardReporter">
        <int name="period">11</int>
        <str name="filter">UPDATE\./update/.*requests</str>
        <str name="filter">QUERY\./select.*requests</str>
      </reporter>

Cluster Reporter

This reporter uses predefined cluster group and the implementing class must be (a subclass of) solr.SolrClusterReporter. It publishes selected metrics from any local registry to the Overseer leader node.

The following configuration properties are supported:

handler

The handler path where reports are sent. Default is /admin/metrics/collector.

period

How often reports are sent, in seconds. Default is 60. Setting this to 0 disables the reporter.

report

report configuration(s), see below.

Each report configuration consists of the following properties:

registry

A regular expression pattern matching local source registries (see SolrMetricManager.registryNames(String…​)), may contain regex capture groups (required).

group

The target registry name where metrics will be grouped. This can be a regular expression pattern that contains back-references to capture groups collected by registry pattern (required).

label

An optional prefix to prepend to metric names, may contain back-references to capture groups collected by registry pattern.

filter

An optional regular expression(s) matching selected metrics to be reported.

The following report specifications are used by default (their result is a single additional metric registry in Overseer, called solr.cluster):

   <lst name="report">
    <str name="group">cluster</str>
    <str name="registry">solr\.jetty</str>
    <str name="label">jetty</str>
   </lst>
   <lst name="report">
    <str name="group">cluster</str>
    <str name="registry">solr\.node</str>
    <str name="label">node</str>
    <str name="filter">CONTAINER\.cores\..*</str>
    <str name="filter">CONTAINER\.fs\..*</str>
   </lst>
   <lst name="report">
     <str name="group">cluster</str>
     <str name="label">jvm</str>
     <str name="registry">solr\.jvm</str>
     <str name="filter">memory\.total\..*</str>
     <str name="filter">memory\.heap\..*</str>
     <str name="filter">os\.SystemLoadAverage</str>
     <str name="filter">os\.FreePhysicalMemorySize</str>
     <str name="filter">os\.FreeSwapSpaceSize</str>
     <str name="filter">os\.OpenFileDescriptorCount</str>
     <str name="filter">threads\.count</str>
   </lst>
   <lst name="report">
    <str name="group">cluster</str>
    <str name="registry">solr\.core\.(.*)\.leader</str>
    <str name="label">leader.$1</str>
    <str name="filter">QUERY\./select/.*</str>
    <str name="filter">UPDATE\./update/.*</str>
    <str name="filter">INDEX\..*</str>
    <str name="filter">TLOG\..*</str>
   </lst>

Example configuration:

         <reporter name="test" group="cluster" class="solr.SolrClusterReporter">
           <str name="handler">/admin/metrics/collector</str>
           <int name="period">11</int>
           <lst name="report">
             <str name="group">aggregated_jvms</str>
             <str name="label">jvm</str>
             <str name="registry">solr\.jvm</str>
             <str name="filter">memory\.total\..*</str>
             <str name="filter">memory\.heap\..*</str>
             <str name="filter">os\.SystemLoadAverage</str>
             <str name="filter">threads\.count</str>
           </lst>
           <lst name="report">
             <str name="group">aggregated_shard_leaders</str>
             <str name="registry">solr\.core\.(.*)\.leader</str>
             <str name="label">leader.$1</str>
             <str name="filter">UPDATE\./update/.*</str>
           </lst>
         </reporter>

Core Level Metrics

These metrics are available only on a per-core basis. Metrics can be aggregated across cores using Shard and Cluster reporters.

Index Merge Metrics

These metrics are collected in respective registries for each core (e.g., solr.core.collection1…​.), under the INDEX category.

Basic metrics are always collected - collection of additional metrics can be turned on using boolean parameters in the /config/indexConfig/metrics section of solrconfig.xml:

<config>
  ...
  <indexConfig>
    <metrics>
      <majorMergeDocs>524288</majorMergeDocs>
      <bool name="mergeDetails">true</bool>
    </metrics>
    ...
  </indexConfig>
...
</config>

The following metrics are collected:

  • INDEX.merge.major - timer for merge operations that include at least "majorMergeDocs" (default value for this parameter is 512k documents).

  • INDEX.merge.minor - timer for merge operations that include less than "majorMergeDocs".

  • INDEX.merge.errors - counter for merge errors.

  • INDEX.flush - meter for index flush operations.

Additionally, the following gauges are reported, which help to monitor the momentary state of index merge operations:

  • INDEX.merge.major.running - number of running major merge operations (depending on the implementation of MergeScheduler that is used there can be several concurrently running merge operations).

  • INDEX.merge.minor.running - as above, for minor merge operations.

  • INDEX.merge.major.running.docs - total number of documents in the segments being currently merged in major merge operations.

  • INDEX.merge.minor.running.docs - as above, for minor merge operations.

  • INDEX.merge.major.running.segments - number of segments being currently merged in major merge operations.

  • INDEX.merge.minor.running.segments - as above, for minor merge operations.

If the boolean flag mergeDetails is true then the following additional metrics are collected:

  • INDEX.merge.major.docs - meter for the number of documents merged in major merge operations

  • INDEX.merge.major.deletedDocs - meter for the number of deleted documents expunged in major merge operations

Metrics API

The admin/metrics endpoint provides access to all the metrics for all metric groups.

A few query parameters are available to limit your request to only certain metrics:

group

The metric group to retrieve. The default is all to retrieve all metrics for all groups. Other possible values are: jvm, jetty, node, and core. More than one group can be specified in a request; multiple group names should be separated by a comma.

type

The type of metric to retrieve. The default is all to retrieve all metric types. Other possible values are counter, gauge, histogram, meter, and timer. More than one type can be specified in a request; multiple types should be separated by a comma.

prefix

The first characters of metric name that will filter the metrics returned to those starting with the provided string. It can be combined with group and/or type parameters. More than one prefix can be specified in a request; multiple prefixes should be separated by a comma. Prefix matching is also case-sensitive.

regex

A regular expression matching metric names. Note: dot separators in metric names must be escaped, e.g., QUERY\./select\..* is a valid regex that matches all metrics with the QUERY./select. prefix.

property

Allows requesting only this metric from any compound metric. Multiple property parameters can be combined to act as an OR request. For example, to only get the 99th and 999th percentile values from all metric types and groups, you can add &property=p99_ms&property=p999_ms to your request. This can be combined with group, type, and prefix as necessary.

key

fully-qualified metric name, which specifies one concrete metric instance (parameter can be specified multiple times to retrieve multiple concrete metrics). NOTE: when this parameter is used, other selection methods listed above are ignored. Fully-qualified name consists of registry name, colon and metric name, with optional colon and metric property. Colons in names can be escaped using back-slash \ character. Examples:

  • key=solr.node:CONTAINER.fs.totalSpace

  • key=solr.core.collection1:QUERY./select.requestTimes:max_ms

  • key=solr.jvm:system.properties:user.name

compact

When false, a more verbose format of the response will be returned. Instead of a response like this:

{"metrics": [
    "solr.core.gettingstarted",
    {
      "CORE.aliases": {
        "value": ["gettingstarted"]
      },
      "CORE.coreName": {
        "value": "gettingstarted"
      },
      "CORE.indexDir": {
        "value": "/solr/example/schemaless/solr/gettingstarted/data/index/"
      },
      "CORE.instanceDir": {
        "value": "/solr/example/schemaless/solr/gettingstarted"
      },
      "CORE.refCount": {
        "value": 1
      },
      "CORE.startTime": {
        "value": "2017-03-14T11:43:23.822Z"
      }
    }
  ]}

The response will look like this:

{"metrics": [
    "solr.core.gettingstarted",
    {
      "CORE.aliases": [
        "gettingstarted"
      ],
      "CORE.coreName": "gettingstarted",
      "CORE.indexDir": "/solr/example/schemaless/solr/gettingstarted/data/index/",
      "CORE.instanceDir": "/solr/example/schemaless/solr/gettingstarted",
      "CORE.refCount": 1,
      "CORE.startTime": "2017-03-14T11:43:23.822Z"
    }
  ]}

Like other request handlers, the Metrics API can also take the wt parameter to define the output format.

Examples

Request only "counter" type metrics in the "core" group, returned in JSON:

http://localhost:8983/solr/admin/metrics?type=counter&group=core

Request only "core" group metrics that start with "INDEX", returned in XML:

http://localhost:8983/solr/admin/metrics?wt=xml&prefix=INDEX&group=core

Request only "core" group metrics that end with ".requests":

http://localhost:8983/solr/admin/metrics?regex=.*\.requests&group=core

Request only "user.name" property of "system.properties" metric from registry "solr.jvm":

http://localhost:8983/solr/admin/metrics?wt=xml?key=solr.jvm:system.properties:user.name