Welcome to the ITSI module for Telegraf Apache Kafka smart monitoring documentation

The ITSI module for Telegraf Kafka monitoring provides smart insight monitoring for Apache Kafka monitoring, on top of Splunk and ITSI.

glass_table.png service_analyser.png main1.png

The ITSI provides builtin and native monitoring for all Apache Kafka components, as well as the Confluent stack components:

  • Zookeeper
  • Apache Kafka Brokers
  • Apache Kafka Connect
  • Confluent schema-registry
  • Confluent ksql-server
  • Confluent kafka-rest
  • Kafka SLA and end to end monitoring with the LinkedIn Kafka monitor
  • Kafka Consumers lag monitoring with Burrow (Kafka Connect connectors, Kafka Streams…)

Fully multi-tenant compatible, the ITSI module can manage different environments or data-centers using tags at metrics low level.

It is recommended to read the unified guide for Kafka and Confluent monitoring first:

https://splunk-guide-for-kafka-monitoring.readthedocs.io

Overview:

About

  • Author: Guilhem Marchand
  • First release published in October 2018
  • Purposes:

The ITSI module for Apache Kafka end to end monitoring leverages the best components to provide a key layer monitoring for your Kafka infrastructure :

The ITSI module provides a native and builtin integration with Splunk and ITSI:

  • Builtin entities discovery for Zookeeper servers, Kafka brokers, Kafka connect nodes, Kafka connect source and sink tasks, Kafka-monitor, Kafka topics, Kafka Consumers, Confluent schema-registry/ksql-servers/kafka-rest
  • Services templates and KPI base searches for Zookeeper, Kafka brokers, Kafka connect and source/sink tasks, Kafka LinkedIn monitor, Kafka topics, Kafka Consumers Lag monitoring, Confluent schema-registry
  • Rich entity health views to manage Operating System metrics ingested in the Splunk metric store
overview_diagram

Compatibility

Splunk compatibility

All the metrics are ingested into the high performance Splunk metric store, Splunk 7.0.x or higher is required.

ITSI compatibility

The ITSI module has been tested and qualified against reasonably fresh versions of ITSI, recommended version is 3.1.0 and higher, previous versions may work as well although it has not and will not be tested.

Telegraf compatibility

Telegraf supports various operating systems and process architectures including any version of Linux and Windows.

For more information:

Containers compatibility

If you are running Kafka in containers, you are at the right place, all of the components can natively run in docker.

Kafka and Confluent compatibility

Qualification and certification is made against Kafka V2.x and Confluent V5.x, earlier versions might however work with no issues but are not being tested.

Known Issues

There are no known issues at the moment.

Support

The ITSI module for Telegraf Apache Kafka smart monitoring is community supported.

To get support, use of one the following options:

Splunk Answers

Open a question in Splunk answers for the application:

Splunk community slack

Contact me on Splunk community slack, or even better, ask the community !

Open a issue in Git

To report an issue, request a feature change or improvement, please open an issue in Github:

Email support

However, previous options are far betters, and will give you all the chances to get a quick support from the community of fellow Splunkers.

Download

ITSI Module for Telegraf Apache Kafka smart monitoring

The ITSI Module for Telegraf Apache Kafka can be downloaded from:

Deployment and configuration:

Deployment & Upgrades

Deployment matrix

Splunk roles required
ITSI Search head yes
Indexer tiers no

If ITSI is running in Search Head Cluster (SHC), the ITSI module must be deployed by the SHC deployer.

The deployment and configuration of the ITSI module requires the creation of a dedicated metric index (by default called telegraf_kafka), see the implementation section.

Initial deployment

The deployment of the ITSI module for Telegraf Kafka is straight forward.

Deploy the ITSI module using one of the following options:

  • Using the application manager in Splunk Web (Settings / Manages apps)
  • Extracting the content of the tgz archive in the “apps” directory of Splunk
  • For SHC configurations (Search Head Cluster), extract the tgz content in the SHC deployer and publish the SHC bundle

Upgrades

Upgrading the ITSI module is pretty much the same operation than the initial deployment.

Upgrades of the components

Upgrading the different components (Telegraf, Jolokia, etc.) rely on each of the technologies, please consult the deployment main pages.

Implementation

Data collection diagram overview:

overview_diagram

Splunk configuration

Index definition

The ITSI module relies by default on the creation of a metrics index called “telegraf_kafka”:

indexes.conf example with no Splunk volume:

[telegraf_kafka]
coldPath = $SPLUNK_DB/telegraf_kafka/colddb
datatype = metric
homePath = $SPLUNK_DB/telegraf_kafka/db
thawedPath = $SPLUNK_DB/telegraf_kafka/thaweddb

indexes.conf example with Splunk volumes:

[telegraf_kafka]
coldPath = volume:cold/telegraf_kafka/colddb
datatype = metric
homePath = volume:primary/telegraf_kafka/db
thawedPath = $SPLUNK_DB/telegraf_kafka/thaweddb

In a Splunk distributed configuration (cluster of indexers), this configuration stands on the cluster master node.

All Splunk searches included in the added refer to the utilisation of a macro called “telegraf_kafka_index” included in:

  • DA-ITSI-TELEGRAF-KAFKA/default/macros.conf

If you wish to use a different index model, this macro shall be customized to override the default model.

HEC input ingestion and definition

The default recommended way of ingesting the Kafka metrics is using the HTTP Events Collector method which requires the creation of an HEC input.

inputs.conf example:

[http://telegraf_kafka_monitoring]
disabled = 0
index = telegraf_kafka
token = 205d43f1-2a31-4e60-a8b3-327eda49944a

If you create the HEC input via Splunk Web interface, it is not required to select an explicit value for source and sourcetype.

The HEC input will be ideally relying on a load balancer to provides resiliency and load balancing across your HEC input nodes.

Other ingesting methods

There are other methods possible to ingest the Kafka metrics in Splunk:

  • TCP input (graphite format with tags support)
  • KAFKA ingestion (Kafka destination from Telegraf in graphite format with tags support, and Splunk connect for Kafka)
  • File monitoring with standard Splunk input monitors (file output plugin from Telegraf)

Notes: In the very specific context of monitoring Kafka, it is not a good design to use Kafka as the ingestion method since you will most likely never be able to know when an issue happens on Kafka.

These methods require the deployment of an additional Technology addon: https://splunkbase.splunk.com/app/4193

These methods are heavily described here: https://da-itsi-telegraf-os.readthedocs.io/en/latest/telegraf.html

Telegraf installation and configuration

Telegraf installation, configuration and start

If you are running Telegraf as a regular process in machine, the standard installation of Telegraf is really straightforward, consult:

If you have a Splunk Universal Forwarder deployment, you can deploy, run and maintain Telegraf and its configuration through a Splunk application (TA), consult:

An example of a ready to use TA application can be found here:

For Splunk customers, this solution has various advantages as you can deploy and maintain using your existing Splunk infrastructure.

Telegraf is extremely container friendly, a container approach is very convenient as you can easily run multiple Telegraf containers to monitor each of the Kafka infrastructure components:

Telegraf output configuration

Whether you will be running Telegraf in various containers, or installed as a regular software within the different servers composing your Kafka infrastructure, a minimal configuration is required to teach Telegraf how to forward the metrics to your Splunk deployment.

Telegraf is able to send to data to Splunk in different ways:

  • Splunk HTTP Events Collector (HEC) - Since Telegraf v1.8
  • Splunk TCP inputs in Graphite format with tags support and the TA for Telegraf
  • Apache Kafka topic in Graphite format with tags support and the TA for Telegraf and Splunk connect for Kafka

Who watches for the watcher?

As you are running a Kafka deployment, it would seem very logical to produce metrics in a Kafka topic. However, it presents a specific concern for Kafka itself.

If you use this same system for monitoring Kafka itself, it is very likely that you will never know when Kafka is broken because the data flow for your monitoring system will be broken as well.

The recommendation is to rely either on Splunk HEC or TCP inputs to forward Telegraf metrics data for the Kafka monitoring.

A minimal configuration for telegraf.conf, running in container or as a regular process in machine and forwarding to HEC:

[global_tags]
  # the env tag is used by the application for multi-environments management
  env = "my_env"
  # the label tag is an optional tag used by the application that you can use as additional label for the services or infrastructure
  label = "my_env_label"

[agent]
  interval = "10s"
  flush_interval = "10s"
  hostname = "$HOSTNAME"

# outputs
[[outputs.http]]
   url = "https://splunk:8088/services/collector"
   insecure_skip_verify = true
   data_format = "splunkmetric"
    ## Provides time, index, source overrides for the HEC
   splunkmetric_hec_routing = true
    ## Additional HTTP headers
    [outputs.http.headers]
   # Should be set manually to "application/json" for json data_format
      Content-Type = "application/json"
      Authorization = "Splunk 205d43f1-2a31-4e60-a8b3-327eda49944a"
      X-Splunk-Request-Channel = "205d43f1-2a31-4e60-a8b3-327eda49944a"

If for some reasons, you have to use either of the 2 other solutions, please consult:

Jolokia JVM monitoring

Kafka components are being monitored through the very powerful Jolokia agent:

Basically, Jolokia JVM agent can be started in 2 modes, either as using the -javaagent argument during the start of the JVM, or on the fly by attaching Jolokia to the JVM running PID:

Starting Jolokia with the JVM

To start Jolokia agent using the -javaagent argument, use such option at the start of the JVM:

-javaagent:/opt/jolokia/jolokia-jvm-1.6.0-agent.jar=port=8778,host=0.0.0.0

Note: This method is the method used in the docker example within this documentation by using the environment variables of the container.

When running on dedicated servers or virtual machines, update the relevant systemd configuration file to start Jolokia automatically:

For Kafka brokers

Environment="KAFKA_OPTS=-javaagent:/opt/jolokia/jolokia-jvm-1.6.0-agent.jar=port=8778,host=0.0.0.0"

For Kafka Connect

Environment="KAFKA_OPTS=-javaagent:/opt/jolokia/jolokia-jvm-1.6.0-agent.jar=port=8778,host=0.0.0.0"

For Confluent schema-registry

Environment="KAFKA_OPTS=-javaagent:/opt/jolokia/jolokia-jvm-1.6.0-agent.jar=port=8778,host=0.0.0.0"

For Confluent ksql-server

Environment="KSQL_OPTS=-javaagent:/opt/jolokia/jolokia-jvm-1.6.0-agent.jar=port=8778,host=0.0.0.0"

For Confluent kafka-rest

Environment="KAFKAREST_OPTS=-javaagent:/opt/jolokia/jolokia-jvm-1.6.0-agent.jar=port=8778,host=0.0.0.0"

Starting Jolokia on the fly

To attach Jolokia agent to an existing JVM, identify its process ID (PID), simplistic example:

ps -ef | grep 'kafka.properties' | grep -v grep | awk '{print $1}'

Then:

java -jar /opt/jolokia/jolokia-jvm-1.6.0-agent.jar --host 0.0.0.0 --port 8778 start <PID>

Add this operation to any custom init scripts you use to start the Kafka components.

Zookeeper monitoring

Collecting with Telegraf

The Zookeeper monitoring is very simple and achieved by Telegraf and the Zookeeper input plugin.

The following configuration stands in telegraf.conf and configures the input plugin to monitor multiple Zookeeper servers from one source:

# zookeeper metrics
[[inputs.zookeeper]]
  servers = ["zookeeper-1:12181","zookeeper-2:22181","zookeeper-3:32181"]

If each server runs an instance of Zookeeper and you deploy Telegraf, you can simply collect from the localhost:

# zookeeper metrics
[[inputs.zookeeper]]
  servers = ["$HOSTNAME:2181"]

Full telegraf.conf example

The following telegraf.conf collects a cluster of 3 Zookeeper servers:

[global_tags]
  # the env tag is used by the application for multi-environments management
  env = "my_env"
  # the label tag is an optional tag used by the application that you can use as additional label for the services or infrastructure
  label = "my_env_label"

[agent]
  interval = "10s"
  flush_interval = "10s"
  hostname = "$HOSTNAME"

# outputs
[[outputs.http]]
   url = "https://splunk:8088/services/collector"
   insecure_skip_verify = true
   data_format = "splunkmetric"
    ## Provides time, index, source overrides for the HEC
   splunkmetric_hec_routing = true
    ## Additional HTTP headers
    [outputs.http.headers]
   # Should be set manually to "application/json" for json data_format
      Content-Type = "application/json"
      Authorization = "Splunk 205d43f1-2a31-4e60-a8b3-327eda49944a"
      X-Splunk-Request-Channel = "205d43f1-2a31-4e60-a8b3-327eda49944a"

# zookeeper metrics
[[inputs.zookeeper]]
  servers = ["zookeeper-1:12181","zookeeper-2:22181","zookeeper-3:32181"]

Visualization of metrics within the Splunk metrics workspace application:

zookeeper_metrics_workspace.png

Using mcatalog search command to verify data availability:

| mcatalog values(metric_name) values(_dims) where index=* metric_name=zookeeper.*

Kafka brokers monitoring with Jolokia

Jolokia

example: Jolokia start in docker environment:

environment:
  KAFKA_BROKER_ID: 1
  KAFKA_ZOOKEEPER_CONNECT: zookeeper-1:12181,zookeeper-2:12181,zookeeper-3:12181
  KAFKA_ADVERTISED_LISTENERS: PLAINTEXT://kafka-1:19092
  KAFKA_OPTS: "-javaagent:/opt/jolokia/jolokia-jvm-1.6.0-agent.jar=port=8778,host=0.0.0.0"

Collecting with Telegraf

Depending on how you run Kafka and your architecture preferences, you may prefer to collect all the brokers metrics from one Telegraf collector, or installed locally on the Kafka brocker machine.

Connecting to multiple remote Jolokia instances:

# Kafka JVM monitoring
[[inputs.jolokia2_agent]]
  name_prefix = "kafka_"
  urls = ["http://kafka-1:18778/jolokia","http://kafka-2:28778/jolokia","http://kafka-3:38778/jolokia"]

Connecting to the local Jolokia instance:

# Kafka JVM monitoring
[[inputs.jolokia2_agent]]
  name_prefix = "kafka_"
  urls = ["http://$HOSTNAME:8778/jolokia"]

Full telegraf.conf example

The following telegraf.conf collects a cluster of 3 Kafka brokers:

[global_tags]
  # the env tag is used by the application for multi-environments management
  env = "my_env"
  # the label tag is an optional tag used by the application that you can use as additional label for the services or infrastructure
  label = "my_env_label"

[agent]
  interval = "10s"
  flush_interval = "10s"
  hostname = "$HOSTNAME"

# outputs
[[outputs.http]]
   url = "https://splunk:8088/services/collector"
   insecure_skip_verify = true
   data_format = "splunkmetric"
    ## Provides time, index, source overrides for the HEC
   splunkmetric_hec_routing = true
    ## Additional HTTP headers
    [outputs.http.headers]
   # Should be set manually to "application/json" for json data_format
      Content-Type = "application/json"
      Authorization = "Splunk 205d43f1-2a31-4e60-a8b3-327eda49944a"
      X-Splunk-Request-Channel = "205d43f1-2a31-4e60-a8b3-327eda49944a"

# Kafka JVM monitoring

[[inputs.jolokia2_agent]]
  name_prefix = "kafka_"
  urls = ["http://kafka-1:18778/jolokia","http://kafka-2:28778/jolokia","http://kafka-3:38778/jolokia"]

[[inputs.jolokia2_agent.metric]]
  name         = "controller"
  mbean        = "kafka.controller:name=*,type=*"
  field_prefix = "$1."

[[inputs.jolokia2_agent.metric]]
  name         = "replica_manager"
  mbean        = "kafka.server:name=*,type=ReplicaManager"
  field_prefix = "$1."

[[inputs.jolokia2_agent.metric]]
  name         = "purgatory"
  mbean        = "kafka.server:delayedOperation=*,name=*,type=DelayedOperationPurgatory"
  field_prefix = "$1."
  field_name   = "$2"

[[inputs.jolokia2_agent.metric]]
  name     = "client"
  mbean    = "kafka.server:client-id=*,type=*"
  tag_keys = ["client-id", "type"]

[[inputs.jolokia2_agent.metric]]
  name         = "network"
  mbean        = "kafka.network:name=*,request=*,type=RequestMetrics"
  field_prefix = "$1."
  tag_keys     = ["request"]

[[inputs.jolokia2_agent.metric]]
  name         = "network"
  mbean        = "kafka.network:name=ResponseQueueSize,type=RequestChannel"
  field_prefix = "ResponseQueueSize"
  tag_keys     = ["name"]

[[inputs.jolokia2_agent.metric]]
  name         = "network"
  mbean        = "kafka.network:name=NetworkProcessorAvgIdlePercent,type=SocketServer"
  field_prefix = "NetworkProcessorAvgIdlePercent"
  tag_keys     = ["name"]

[[inputs.jolokia2_agent.metric]]
  name         = "topics"
  mbean        = "kafka.server:name=*,type=BrokerTopicMetrics"
  field_prefix = "$1."

[[inputs.jolokia2_agent.metric]]
  name         = "topic"
  mbean        = "kafka.server:name=*,topic=*,type=BrokerTopicMetrics"
  field_prefix = "$1."
  tag_keys     = ["topic"]

[[inputs.jolokia2_agent.metric]]
  name       = "partition"
  mbean      = "kafka.log:name=*,partition=*,topic=*,type=Log"
  field_name = "$1"
  tag_keys   = ["topic", "partition"]

[[inputs.jolokia2_agent.metric]]
  name       = "log"
  mbean      = "kafka.log:name=LogFlushRateAndTimeMs,type=LogFlushStats"
  field_name = "LogFlushRateAndTimeMs"
  tag_keys   = ["name"]

[[inputs.jolokia2_agent.metric]]
  name       = "partition"
  mbean      = "kafka.cluster:name=UnderReplicated,partition=*,topic=*,type=Partition"
  field_name = "UnderReplicatedPartitions"
  tag_keys   = ["topic", "partition"]

[[inputs.jolokia2_agent.metric]]
  name     = "request_handlers"
  mbean    = "kafka.server:name=RequestHandlerAvgIdlePercent,type=KafkaRequestHandlerPool"
  tag_keys = ["name"]

# JVM garbage collector monitoring
[[inputs.jolokia2_agent.metric]]
  name     = "jvm_garbage_collector"
  mbean    = "java.lang:name=*,type=GarbageCollector"
  paths    = ["CollectionTime", "CollectionCount", "LastGcInfo"]
  tag_keys = ["name"]

Visualization of metrics within the Splunk metrics workspace application:

kafka_kafka_metrics_workspace.png

Using mcatalog search command to verify data availability:

| mcatalog values(metric_name) values(_dims) where index=* metric_name=kafka_*.*

Kafka connect monitoring

Jolokia

example: Jolokia start in docker environment:

environment:
  KAFKA_OPTS: "-javaagent:/opt/jolokia/jolokia-jvm-1.6.0-agent.jar=port=18779,host=0.0.0.0"
command: "/usr/bin/connect-distributed /etc/kafka-connect/config/connect-distributed.properties-kafka-connect-1"

Collecting with Telegraf

Connecting to multiple remote Jolokia instances:

# Kafka-connect JVM monitoring
[[inputs.jolokia2_agent]]
  name_prefix = "kafka_connect."
  urls = ["http://kafka-connect-1:18779/jolokia","http://kafka-connect-2:28779/jolokia","http://kafka-connect-3:38779/jolokia"]

Connecting to local Jolokia instance:

# Kafka-connect JVM monitoring
 [[inputs.jolokia2_agent]]
   name_prefix = "kafka_connect."
   urls = ["http://$HOSTNAME:8778/jolokia"]

Full telegraf.conf example

bellow a full telegraf.conf example:

[global_tags]
  # the env tag is used by the application for multi-environments management
  env = "my_env"
  # the label tag is an optional tag used by the application that you can use as additional label for the services or infrastructure
  label = "my_env_label"

[agent]
  interval = "10s"
  flush_interval = "10s"
  hostname = "$HOSTNAME"

# outputs
[[outputs.http]]
   url = "https://splunk:8088/services/collector"
   insecure_skip_verify = true
   data_format = "splunkmetric"
    ## Provides time, index, source overrides for the HEC
   splunkmetric_hec_routing = true
    ## Additional HTTP headers
    [outputs.http.headers]
   # Should be set manually to "application/json" for json data_format
      Content-Type = "application/json"
      Authorization = "Splunk 205d43f1-2a31-4e60-a8b3-327eda49944a"
      X-Splunk-Request-Channel = "205d43f1-2a31-4e60-a8b3-327eda49944a"

# Kafka-connect JVM monitoring

[[inputs.jolokia2_agent]]
  name_prefix = "kafka_connect."
  urls = ["http://kafka-connect-1:18779/jolokia","http://kafka-connect-2:28779/jolokia","http://kafka-connect-3:38779/jolokia"]

[[inputs.jolokia2_agent.metric]]
  name         = "worker"
  mbean        = "kafka.connect:type=connect-worker-metrics"

[[inputs.jolokia2_agent.metric]]
  name         = "worker"
  mbean        = "kafka.connect:type=connect-worker-rebalance-metrics"

[[inputs.jolokia2_agent.metric]]
  name         = "connector-task"
  mbean        = "kafka.connect:type=connector-task-metrics,connector=*,task=*"
  tag_keys = ["connector", "task"]

[[inputs.jolokia2_agent.metric]]
  name         = "sink-task"
  mbean        = "kafka.connect:type=sink-task-metrics,connector=*,task=*"
  tag_keys = ["connector", "task"]

[[inputs.jolokia2_agent.metric]]
  name         = "source-task"
  mbean        = "kafka.connect:type=source-task-metrics,connector=*,task=*"
  tag_keys = ["connector", "task"]

[[inputs.jolokia2_agent.metric]]
  name         = "error-task"
  mbean        = "kafka.connect:type=task-error-metrics,connector=*,task=*"
  tag_keys = ["connector", "task"]

# Kafka connect return a status value which is non numerical
# Using the enum processor with the following configuration replaces the string value by our mapping
[[processors.enum]]
  [[processors.enum.mapping]]
    ## Name of the field to map
    field = "status"

    ## Table of mappings
    [processors.enum.mapping.value_mappings]
      paused = 0
      running = 1
      unassigned = 2
      failed = 3
      destroyed = 4

Visualization of metrics within the Splunk metrics workspace application:

kafka_kafka_connect_workspace.png

Using mcatalog search command to verify data availability:

| mcatalog values(metric_name) values(_dims) where index=* metric_name=kafka_connect.*

Kafka LinkedIn monitor - end to end monitoring

Installing and starting the Kafka monitor

LinkedIn provides an extremely powerful open source end to end monitoring solution for Kafka, please consult:

As a builtin configuration, the kafka-monitor implements a jolokia agent, so collecting the metrics with Telegraf cannot be more easy !

It is very straightforward to run the kafka-monitor in a docker container, first you need to create your own image:

Once your Kafka monitor is running, you need a Telegraf instance that will be collecting the JMX beans, example:

[global_tags]
  # the env tag is used by the application for multi-environments management
  env = "my_env"
  # the label tag is an optional tag used by the application that you can use as additional label for the services or infrastructure
  label = "my_env_label"

[agent]
  interval = "10s"
  flush_interval = "10s"
  hostname = "$HOSTNAME"

# outputs
[[outputs.http]]
   url = "https://splunk:8088/services/collector"
   insecure_skip_verify = true
   data_format = "splunkmetric"
    ## Provides time, index, source overrides for the HEC
   splunkmetric_hec_routing = true
    ## Additional HTTP headers
    [outputs.http.headers]
   # Should be set manually to "application/json" for json data_format
      Content-Type = "application/json"
      Authorization = "Splunk 205d43f1-2a31-4e60-a8b3-327eda49944a"
      X-Splunk-Request-Channel = "205d43f1-2a31-4e60-a8b3-327eda49944a"

# Kafka JVM monitoring

[[inputs.jolokia2_agent]]
  name_prefix = "kafka_"
  urls = ["http://kafka-monitor:8778/jolokia"]

[[inputs.jolokia2_agent.metric]]
  name         = "kafka-monitor"
  mbean        = "kmf.services:name=*,type=*"

Visualization of metrics within the Splunk metrics workspace application:

kafka_monitoring_metrics_workspace.png

Using mcatalog search command to verify data availability:

| mcatalog values(metric_name) values(_dims) where index=* metric_name=kafka_kafka-monitor.*

Confluent schema-registry

Jolokia

example: Jolokia start in docker environment:

environment:
  SCHEMA_REGISTRY_KAFKASTORE_CONNECTION_URL: zookeeper-1:12181,zookeeper-2:12181,zookeeper-3:12181
  SCHEMA_REGISTRY_HOST_NAME: schema-registry
  SCHEMA_REGISTRY_LISTENERS: "http://0.0.0.0:8081"
  SCHEMA_REGISTRY_OPTS: "-javaagent:/opt/jolokia/jolokia-jvm-1.6.0-agent.jar=port=18783,host=0.0.0.0"

Collecting with Telegraf

Connecting to multiple remote Jolokia instances:

[[inputs.jolokia2_agent]]
  name_prefix = "kafka_schema-registry."
  urls = ["http://schema-registry:18783/jolokia"]

Connecting to local Jolokia instance:

# Kafka-connect JVM monitoring
 [[inputs.jolokia2_agent]]
  name_prefix = "kafka_schema-registry."
   urls = ["http://$HOSTNAME:8778/jolokia"]

Full telegraf.conf example

bellow a full telegraf.conf example:

[global_tags]
  # the env tag is used by the application for multi-environments management
  env = "my_env"
  # the label tag is an optional tag used by the application that you can use as additional label for the services or infrastructure
  label = "my_env_label"

[agent]
  interval = "10s"
  flush_interval = "10s"
  hostname = "$HOSTNAME"

# outputs
[[outputs.http]]
   url = "https://splunk:8088/services/collector"
   insecure_skip_verify = true
   data_format = "splunkmetric"
    ## Provides time, index, source overrides for the HEC
   splunkmetric_hec_routing = true
    ## Additional HTTP headers
    [outputs.http.headers]
   # Should be set manually to "application/json" for json data_format
      Content-Type = "application/json"
      Authorization = "Splunk 205d43f1-2a31-4e60-a8b3-327eda49944a"
      X-Splunk-Request-Channel = "205d43f1-2a31-4e60-a8b3-327eda49944a"

# schema-registry JVM monitoring

[[inputs.jolokia2_agent]]
  name_prefix = "kafka_schema-registry."
  urls = ["http://schema-registry:18783/jolokia"]

[[inputs.jolokia2_agent.metric]]
  name         = "jetty-metrics"
  mbean        = "kafka.schema.registry:type=jetty-metrics"
  paths = ["connections-active", "connections-opened-rate", "connections-closed-rate"]

[[inputs.jolokia2_agent.metric]]
  name         = "master-slave-role"
  mbean        = "kafka.schema.registry:type=master-slave-role"

[[inputs.jolokia2_agent.metric]]
  name         = "jersey-metrics"
  mbean        = "kafka.schema.registry:type=jersey-metrics"

Visualization of metrics within the Splunk metrics workspace application:

confluent_schema-registry_metrics_workspace.png

Using mcatalog search command to verify data availability:

| mcatalog values(metric_name) values(_dims) where index=* metric_name=kafka_schema-registry.*

Confluent ksql-server

Jolokia

example: Jolokia start in docker environment:

environment:
  KSQL_BOOTSTRAP_SERVERS: PLAINTEXT://kafka-1:19092,PLAINTEXT://kafka-2:29092,PLAINTEXT://kafka-3:39092
  KSQL_KSQL_SERVICE_ID: confluent_standalone_1_
  SCHEMA_REGISTRY_LISTENERS: "http://0.0.0.0:8081"
  KSQL_OPTS: "-javaagent:/opt/jolokia/jolokia-jvm-1.6.0-agent.jar=port=18784,host=0.0.0.0"

Collecting with Telegraf

Connecting to multiple remote Jolokia instances:

[[inputs.jolokia2_agent]]
  name_prefix = "kafka_"
  urls = ["http://ksql-server-1:18784/jolokia"]

Connecting to local Jolokia instance:

[[inputs.jolokia2_agent]]
  name_prefix = "kafka_"
  urls = ["http://$HOSTNAME:18784/jolokia"]

Full telegraf.conf example

bellow a full telegraf.conf example:

[global_tags]
  # the env tag is used by the application for multi-environments management
  env = "my_env"
  # the label tag is an optional tag used by the application that you can use as additional label for the services or infrastructure
  label = "my_env_label"

[agent]
  interval = "10s"
  flush_interval = "10s"
  hostname = "$HOSTNAME"

# outputs
[[outputs.http]]
   url = "https://splunk:8088/services/collector"
   insecure_skip_verify = true
   data_format = "splunkmetric"
    ## Provides time, index, source overrides for the HEC
   splunkmetric_hec_routing = true
    ## Additional HTTP headers
    [outputs.http.headers]
   # Should be set manually to "application/json" for json data_format
      Content-Type = "application/json"
      Authorization = "Splunk 205d43f1-2a31-4e60-a8b3-327eda49944a"
      X-Splunk-Request-Channel = "205d43f1-2a31-4e60-a8b3-327eda49944a"

# ksql-server JVM monitoring

 [[inputs.jolokia2_agent]]
   name_prefix = "kafka_"
   urls = ["http://ksql-server:18784/jolokia"]

 [[inputs.jolokia2_agent.metric]]
   name         = "ksql-server"
   mbean        = "io.confluent.ksql.metrics:type=*"

Visualization of metrics within the Splunk metrics workspace application:

confluent_ksql_server_metrics_workspace.png

Using mcatalog search command to verify data availability:

| mcatalog values(metric_name) values(_dims) where index=* metric_name=kafka_ksql-server.*

Confluent kafka-rest

Jolokia

example: Jolokia start in docker environment:

environment:
  KAFKA_REST_ZOOKEEPER_CONNECT: "zookeeper-1:12181,zookeeper-2:22181,zookeeper-3:32181"
  KAFKA_REST_LISTENERS: "http://localhost:18089"
  KAFKA_REST_SCHEMA_REGISTRY_URL: "http://schema-registry-1:18083"
  KAFKAREST_OPTS: "-javaagent:/opt/jolokia/jolokia-jvm-1.6.0-agent.jar=port=18785,host=0.0.0.0"
  KAFKA_REST_HOST_NAME: "kafka-rest"

notes: KAFKAREST_OPTS is not a typo, this is (strangely) the right name to configuration java options.

Collecting with Telegraf

Connecting to multiple remote Jolokia instances:

[[inputs.jolokia2_agent]]
  name_prefix = "kafka_kafka-rest."
  urls = ["http://kafka-rest:8778/jolokia"]

Connecting to local Jolokia instance:

[[inputs.jolokia2_agent]]
  name_prefix = "kafka_kafka-rest."
  urls = ["http://$HOSTNAME:18785/jolokia"]

Full telegraf.conf example

bellow a full telegraf.conf example:

[global_tags]
  # the env tag is used by the application for multi-environments management
  env = "my_env"
  # the label tag is an optional tag used by the application that you can use as additional label for the services or infrastructure
  label = "my_env_label"

[agent]
  interval = "10s"
  flush_interval = "10s"
  hostname = "$HOSTNAME"

# outputs
[[outputs.http]]
   url = "https://splunk:8088/services/collector"
   insecure_skip_verify = true
   data_format = "splunkmetric"
    ## Provides time, index, source overrides for the HEC
   splunkmetric_hec_routing = true
    ## Additional HTTP headers
    [outputs.http.headers]
   # Should be set manually to "application/json" for json data_format
      Content-Type = "application/json"
      Authorization = "Splunk 205d43f1-2a31-4e60-a8b3-327eda49944a"
      X-Splunk-Request-Channel = "205d43f1-2a31-4e60-a8b3-327eda49944a"

 # kafka-rest JVM monitoring

 [[inputs.jolokia2_agent]]
   name_prefix = "kafka_kafka-rest."
   urls = ["http://kafka-rest:18785/jolokia"]

 [[inputs.jolokia2_agent.metric]]
   name         = "jetty-metrics"
   mbean        = "kafka.rest:type=jetty-metrics"
   paths = ["connections-active", "connections-opened-rate", "connections-closed-rate"]

 [[inputs.jolokia2_agent.metric]]
   name         = "jersey-metrics"
   mbean        = "kafka.rest:type=jersey-metrics"

Visualization of metrics within the Splunk metrics workspace application:

confluent_kafka_rest_metrics_workspace.png

Using mcatalog search command to verify data availability:

| mcatalog values(metric_name) values(_dims) where index=* metric_name=kafka_kafka_kafka-rest.*

Burrow Lag Consumers

As from their authors, Burrow is a monitoring companion for Apache Kafka that provides consumer lag checking as a service without the need for specifying thresholds.

See: https://github.com/linkedin/Burrow

Burrow workflow diagram:

burrow_diagram.png

Burrow is a very powerful application that monitors all consumers (Kafka Connect connectors, Kafka Streams…) to report an advanced state of the service automatically, and various useful lagging metrics.

Telegraf has a native input for Burrow which polls consumers, topics and partitions lag metrics and statuses over http, use the following telegraf minimal configuration:

See: https://github.com/influxdata/telegraf/tree/master/plugins/inputs/burrow

[global_tags]
  # the env tag is used by the application for multi-environments management
  env = "my_env"
  # the label tag is an optional tag used by the application that you can use as additional label for the services or infrastructure
  label = "my_env_label"

[agent]
  interval = "10s"
  flush_interval = "10s"
  hostname = "$HOSTNAME"

# outputs
[[outputs.http]]
   url = "https://splunk:8088/services/collector"
   insecure_skip_verify = true
   data_format = "splunkmetric"
    ## Provides time, index, source overrides for the HEC
   splunkmetric_hec_routing = true
    ## Additional HTTP headers
    [outputs.http.headers]
   # Should be set manually to "application/json" for json data_format
      Content-Type = "application/json"
      Authorization = "Splunk 205d43f1-2a31-4e60-a8b3-327eda49944a"
      X-Splunk-Request-Channel = "205d43f1-2a31-4e60-a8b3-327eda49944a"

# Burrow

[[inputs.burrow]]
  ## Burrow API endpoints in format "schema://host:port".
  ## Default is "http://localhost:8000".
  servers = ["http://dockerhost:9001"]

  ## Override Burrow API prefix.
  ## Useful when Burrow is behind reverse-proxy.
  # api_prefix = "/v3/kafka"

  ## Maximum time to receive response.
  # response_timeout = "5s"

  ## Limit per-server concurrent connections.
  ## Useful in case of large number of topics or consumer groups.
  # concurrent_connections = 20

  ## Filter clusters, default is no filtering.
  ## Values can be specified as glob patterns.
  # clusters_include = []
  # clusters_exclude = []

  ## Filter consumer groups, default is no filtering.
  ## Values can be specified as glob patterns.
  # groups_include = []
  # groups_exclude = []

  ## Filter topics, default is no filtering.
  ## Values can be specified as glob patterns.
  # topics_include = []
  # topics_exclude = []

  ## Credentials for basic HTTP authentication.
  # username = ""
  # password = ""

  ## Optional SSL config
  # ssl_ca = "/etc/telegraf/ca.pem"
  # ssl_cert = "/etc/telegraf/cert.pem"
  # ssl_key = "/etc/telegraf/key.pem"
  # insecure_skip_verify = false

Visualization of metrics within the Splunk metrics workspace application:

burrow_metrics_workspace.png

Using mcatalog search command to verify data availability:

| mcatalog values(metric_name) values(_dims) where index=* metric_name=burrow_*

Operating System level metrics

Monitoring the Operating System level metrics is fully part of the monitoring requirements of a Kafka infrastructure.

Bare metal servers and virtual machines

ITSI module for Telegraf Operating System

Telegraf has very powerful Operating System level metrics capabilities, checkout the ITSI module for Telegraf Operating System monitoring !

https://da-itsi-telegraf-os.readthedocs.io

itsi_module_telegraf.png
ITSI module for metricator Nmon

Another very powerful way of monitoring Operating System level metrics with a builtin ITSI module and the excellent nmon monitoring:

https://www.octamis.com/metricator-docs/itsi_module.html

itsi_module_metricator.pngg
ITSI module for OS

Last option is using the builtin ITSI module for OS which relies on the TA-nix or TA-Windows:

http://docs.splunk.com/Documentation/ITSI/latest/IModules/AbouttheOperatingSystemModule

Containers with Docker and container orchestrators

Telegraf docker monitoring

Telegraf has very powerful inputs for Docker and is natively compatible with a container orchestrator such as Kubernetes.

Specially with Kubernetes, it is very easy to run a Telegraf container as a daemonset in Kubernetes and retrieve all the performance metrics of the containers.

Docker testing templates

Docker compose templates are provided in the following repository:

https://github.com/guilhemmarchand/kafka-docker-splunk

Using the docker templates allows you to create a full pre-configured Kafka environment with docker, just in 30 seconds.

Integration with Kubernetes is documented here:

https://splunk-guide-for-kafka-monitoring.readthedocs.io

Example:

  • 3 x nodes Zookeeper cluster
  • 3 x nodes Apache Kafka brokers cluster
  • 3 x nodes Apache Kafka connect cluster
  • 1 x node Confluent schema-registry
  • 1 x Splunk standalone server running in docker
  • 1 x LinkedIn Kafka monitor node
  • 1 x Telegraf collector container to collect metrics from Zookeeper, Kafka brokers
  • 1 x Telegraf collector container to collect metrics from Kafka Connect (including source and sink tasks)
  • 1 x Telegraf collector container to collect metrics from LinkedIn Kafka monitor
docker_template

Start the template, have a very short coffee (approx. 30 sec), open Splunk, install the Metrics workspace app and observe the magic happening !

docker-templates.png

Entities discovery

The ITSI entities discovery is a fully automated process that will discover and properly configure your entities in ITSI depending on the data availability in Splunk.

All report rely on extremely fast and optimized queries with mcatalog, which has a negligible processing cost for the Splunk infrastructure.

Entities automatic import

In a nutshell, the following reports are automatically scheduled:

Purpose Report
Zookeeper servers detection DA-ITSI-TELEGRAF-KAFKA-Inventory_Search_zookeeper
Kafka brokers detection DA-ITSI-TELEGRAF-KAFKA-Inventory_Search_kafka_brokers
Kafka topics detection DA-ITSI-TELEGRAF-KAFKA-Inventory_Search_kafka_topics
Kafka connect detection DA-ITSI-TELEGRAF-KAFKA-Inventory_Search_kafka_connect
Kafka connect tasks detection DA-ITSI-TELEGRAF-KAFKA-Inventory_Search_kafka_connect_tasks
Kafka monitors detection DA-ITSI-TELEGRAF-KAFKA-Inventory_Search_linkedin_kafka_monitors
Kafka Consumers detection DA-ITSI-TELEGRAF-KAFKA-Inventory_Search_kafka_burrow_group_consumers
Confluent schema-registry DA-ITSI-TELEGRAF-KAFKA-Inventory_Search_kafka_schema-registry
Confluent ksql-server DA-ITSI-TELEGRAF-KAFKA-Inventory_Search_kafka-ksql-server
Confluent kafka-rest DA-ITSI-TELEGRAF-KAFKA-Inventory_Search_kafka-kafka-rest

When entities are discovered, entities will be added automatically using the itsi_role information field, in addition with several other info fields depending on the components.

Manual entities import

It is possible to manually import the entities in ITSI, and use the searches above:

Configure / Entities / New Entity / Import from Search

Then select the module name, and depending on your needs select the relevant search.

Zookeeper server detection

entities_zookeeper_detection.png

Kafka brokers detection

entities_kafka_brokers_detection.png

Kafka topics detection

entities_kafka_topics_detection.png

Kafka connect detection

entities_kafka_connect_detection.png

Kafka connect tasks detection

entities_kafka_connect_tasks_detection.png

Kafka consumers detection (Burrow)

entities_kafka_consumers_detection.png

Confluent schema-registry nodes detection

entities_schema_registry.png

Confluent ksql-server nodes detection

entities_ksql_server.png

Confluent kafka-rest nodes detection

entities_kafka_rest.png

LinkedIn Kafka monitor nodes detection

entities_kafka_kafka-monitor_detection.png

Services creation

The ITSI module for Telegraf Kafka smart monitoring provides builtin services templates, relying on several base KPIs retrieving data from the metric store.

  • Zookeeper monitoring
  • Kafka brokers monitoring
  • Kafka LinkedIn monitor
  • Kafka topic monitoring
  • Kafka connect monitoring
  • Kafka sink task monitoring
  • Kafka source task monitoring
  • Kafka Consumers lag monitoring
  • Confluent schema-registry monitoring
  • Confluent Confluent ksql-server monitoring
  • Confluent kafka-rest monitoring

As a general practice, if you first goal is designing the IT infrastructure in ITSI, a good generic recommendation is to create a main service container for your Kafka infrastructure.

As such, every service that will be designed will be linked to the main service. (the main service depends on them)

itsi_dep.png

Monitoring Zookeeper servers

To monitor your Zookeeper servers, create a new service using the “Zookeeper monitoring” template service and select the proper filters for your entities:

  • Configure / Service / Create new service / Zookeeper monitoring
service_zookeeper_pic1.png service_zookeeper_pic2.png

Monitoring Kafka Brokers

To monitor your Zookeeper servers, create a new service using the “Kafka brokers monitoring” template service and select the proper filters for your entities:

  • Configure / Service / Create new service / Kafka brokers monitoring
service_kafka_broker_pic1.png service_kafka_broker_pic2.png

Monitoring Kafka Topics

To monitor one or more Kafka topics, create a new service using the “Kafka topic monitoring” template service and select the proper filters for your entities corresponding to your topics:

  • Configure / Service / Create new service / Kafka topic monitoring
service_kafka_topic_pic1.png service_kafka_topic_pic2.png

Monitoring Kafka Connect

To monitor Kafka Connect, create a new service using the “Kafka connect monitoring” template service and select the proper filters for your entities:

  • Configure / Service / Create new service / Kafka connect monitoring
service_kafka_connect_pic1.png service_kafka_connect_pic2.png

Monitoring Kafka Connect Sink taks

To monitor one of more Kafka Connect Sink connectors, create a new service using the “Kafka sink task monitoring” template service and select the proper filters for your entities:

service_kafka_sink_task_pic1.png service_kafka_sink_task_pic2.png

Monitoring Kafka Connect Source taks

To monitor one of more Kafka Connect Source connectors, create a new service using the “Kafka source task monitoring” template service and select the proper filters for your entities:

service_kafka_source_task_pic1.png service_kafka_source_task_pic2.png

Monitoring Kafka Consumers

To monitor one or more Kafka Consumers, create a new service using the “Kafka Consumers lag monitoring” template service and select the proper filters for your entities corresponding to your topics:

  • Configure / Service / Create new service / Kafka lag monitoring
service_kafka_consumers_pic1.png service_kafka_consumers_pic2.png

Monitoring Confluent schema-registry

To monitor one of more Confluent schema-registry nodes, create a new service using the “Kafka schema-registry monitoring” template service and select the proper filters for your entities:

service_confluent_schema_registry_pic1.png service_confluent_schema_registry_pic2.png

Monitoring Confluent ksql-server

To monitor one of more Confluent ksql servers, create a new service using the “Confluent ksql-server monitoring” template service and select the proper filters for your entities:

service_confluent_ksql_server_pic1.png service_confluent_ksql_server_pic2.png

Monitoring Confluent kafka-rest

To monitor one of more Confluent kafka-rest nodes, create a new service using the “Confluent kafka-rest monitoring” template service and select the proper filters for your entities:

service_confluent_kafka_rest_pic1.png service_confluent_kafka_rest_pic2.png

End to end monitoring with LinkedIn Kafka monitor

To monitor your Kafka deployment using the LinkedIn Kafka monitor, create a new service using the “Kafka LinkedIn monitor” template service and select the proper filters for your entities:

  • Configure / Service / Create new service / Kafka LinkedIn monitor
service_kafka_monitor_pic1.png service_kafka_monitor_pic2.png

ITSI Entities dashboard (health views)

Through builtin ITSI deepdive links, you can automatically and easily access to an efficient dashboard that provides insight analytic for the component.

Accessing entities health views is a native ITSI feature, either by:

  • Entities lister (Configure / Entities)
entities_lister.png
  • deepdive link
deepdive_link.png

Zookeeper dashboard view

dashboard_zookeeper1.png

Kafka broker dashboard view

dashboard_kafka_broker_pic1.png dashboard_kafka_broker_pic2.png dashboard_kafka_broker_pic3.png dashboard_kafka_broker_pic4.png dashboard_kafka_broker_pic5.png dashboard_kafka_broker_pic6.png dashboard_kafka_broker_pic7.png dashboard_kafka_broker_pic8.png dashboard_kafka_broker_pic9.png

Kafka topic dashboard view

dashboard_kafka_topic_pic1.png dashboard_kafka_topic_pic2.png dashboard_kafka_topic_pic3.png

Kafka connect dashboard view

dashboard_kafka_connect_pic1.png dashboard_kafka_connect_pic2.png dashboard_kafka_connect_pic3.png dashboard_kafka_connect_pic4.png dashboard_kafka_connect_pic5.png

Kafka connect sink task dashboard view

dashboard_kafka_connect_sink_pic1.png dashboard_kafka_connect_sink_pic2.png dashboard_kafka_connect_sink_pic3.png

Kafka connect source task dashboard view

dashboard_kafka_connect_source_pic1.png dashboard_kafka_connect_source_pic2.png dashboard_kafka_connect_source_pic3.png

Kafka consumers lag monitoring dashboard view (Burrow)

dashboard_kafka_consumers_pic1.png dashboard_kafka_consumers_pic2.png dashboard_kafka_consumers_pic3.png dashboard_kafka_consumers_pic4.png

Confluent schema-registry dashboard view

dashboard_schema_registry_pic1.png dashboard_schema_registry_pic2.png

Confluent ksql-server dashboard view

dashboard_confluent_ksql_server_pic1.png

Confluent kafka-rest dashboard view

dashboard_confluent_kafka_rest_pic1.png dashboard_confluent_kafka_rest_pic2.png

LinkedIn Kafka monitor view

dashboard_kafka_monitor_pic1.png dashboard_kafka_monitor_pic2.png dashboard_kafka_monitor_pic3.png

Troubleshoot:

Troubleshoot & FAQ

Versioniong and build history:

Release notes

Version 1.1.5

  • fix: Static index reference in Kafka Brokers entities discovery report
  • feature: Drilldown to single forms for Offline and Under-replicated partitions in Overview and Kafka Brokers entities views

Version 1.1.4

  • fix: incompatibility for ksql-server with latest Confluent release (5.1.x) due to metric name changes in JMX model

Version 1.1.3

Burrow integration: Kafka Consumer Lag monitoring

  • feature: New KPI basesearch and Service Template for Kafka Consumers Lag Monitoring with Burrow
  • feature: New entity view for Kafka Consumers Lag monitoring

The Burrow integration provides advanced threshold less lag monitoring for Kafka Consumers, such as Kafka Connect connectors and Kafka Streams.

Version 1.1.2

  • unpublished

Version 1.1.1

CAUTION: Breaking changes and major release, telegraf modification is required to provide global tags for env and label dimensions!

https://da-itsi-telegraf-kafka.readthedocs.io/en/latest/kafka_monitoring.html#telegraf-installation-and-configuration

Upgrade path:

  • Upgrade telegraf configuration to provide the env and label tags
  • Upgrade the module, manage entities and rebuild your services

release notes:

  • fix: duplicated KPI id for topic/brokers under replicated replication leads in KPI rendering issues
  • fix: entity rendering issue with Kafka SLA monitor health view

Version 1.1.0

CAUTION: Breaking changes and major release, telegraf modification is required to provide global tags for env and label dimensions!

https://da-itsi-telegraf-kafka.readthedocs.io/en/latest/kafka_monitoring.html#telegraf-installation-and-configuration

Upgrade path:

  • Upgrade telegraf configuration to provide the env and label tags
  • Upgrade the module, manage entities and rebuild your services

release notes:

  • feature: Support for multi-environments / multi-dc deployments with metrics tagging
  • feature: Global rewrite of entities management and identification
  • fix: Moved from second interval to cron schedule for entities import to avoid dup entities at addon installation time
  • fix: Various fixes and improvements

Version 1.0.6

  • feature: Support for Confluent ksql-server
  • feature: Support for Confluent kafka-rest
  • feature: event logging integration with the TA-kafka-streaming-platform

Version 1.0.5

  • feature: Support for Confluent schema-registry
  • feature: Adding follower/leader info in Zookeeper entity view

Version 1.0.4

  • fix: typo on partitions in Kafka brokers view

Version 1.0.3

  • fix: missing entity filter in latency from Zookeeper view
  • fix: incorrect static filter in state from Sink task view

Version 1.0.2

  • fix: incorrect duration shown in Kafka Connect entity view
  • feature: minor improvements in UIs

Version 1.0.1

  • fix: error in state of Source/Sink connector in dashboards

Version 1.0.0

  • initial and first public release