telegraf/docs/CONFIGURATION.md

Configuration

Telegraf's configuration file is written using TOML and is composed of three sections: global tags, agent settings, and plugins.

View the default telegraf.conf config file with all available plugins.

Generating a Configuration File

A default config file can be generated by telegraf:

telegraf config > telegraf.conf

To generate a file with specific inputs and outputs, you can use the --input-filter and --output-filter flags:

telegraf --input-filter cpu:mem:net:swap --output-filter influxdb:kafka config

Configuration Loading

The location of the configuration file can be set via the --config command line flag.

When the --config-directory command line flag is used files ending with .conf in the specified directory will also be included in the Telegraf configuration.

On most systems, the default locations are /etc/telegraf/telegraf.conf for the main configuration file and /etc/telegraf/telegraf.d for the directory of configuration files.

Environment Variables

Environment variables can be used anywhere in the config file, simply prepend them with $. Replacement occurs before file parsing. For strings the variable must be within quotes, e.g., "$STR_VAR", for numbers and booleans they should be unquoted, e.g., $INT_VAR, $BOOL_VAR.

When using the .deb or .rpm packages, you can define environment variables in the /etc/default/telegraf file.

Example:

/etc/default/telegraf:

USER="alice"
INFLUX_URL="http://localhost:8086"
INFLUX_SKIP_DATABASE_CREATION="true"
INFLUX_PASSWORD="monkey123"

/etc/telegraf.conf:

[global_tags]
  user = "$USER"

[[inputs.mem]]

[[outputs.influxdb]]
  urls = ["$INFLUX_URL"]
  skip_database_creation = $INFLUX_SKIP_DATABASE_CREATION
  password = "$INFLUX_PASSWORD"

The above files will produce the following effective configuration file to be parsed:

[global_tags]
  user = "alice"

[[outputs.influxdb]]
  urls = "http://localhost:8086"
  skip_database_creation = true
  password = "monkey123"

Intervals

Intervals are durations of time and can be specified for supporting settings by combining an integer value and time unit as a string value. Valid time units are ns, us (or µs), ms, s, m, h.

[agent]
  interval = "10s"

Global Tags

Global tags can be specified in the [global_tags] table in key="value" format. All metrics that are gathered will be tagged with the tags specified.

[global_tags]
  dc = "us-east-1"

Agent

The agent table configures Telegraf and the defaults used across all plugins.

• interval: Default data collection interval for all inputs.

• round_interval: Rounds collection interval to interval ie, if interval="10s" then always collect on :00, :10, :20, etc.

• metric_batch_size: Telegraf will send metrics to outputs in batches of at most metric_batch_size metrics. This controls the size of writes that Telegraf sends to output plugins.

• metric_buffer_limit: For failed writes, telegraf will cache metric_buffer_limit metrics for each output, and will flush this buffer on a successful write. Oldest metrics are dropped first when this buffer fills. This buffer only fills when writes fail to output plugin(s).

• collection_jitter: Collection jitter is used to jitter the collection by a random interval. Each plugin will sleep for a random time within jitter before collecting. This can be used to avoid many plugins querying things like sysfs at the same time, which can have a measurable effect on the system.

• flush_interval: Default flushing interval for all outputs. Maximum flush_interval will be flush_interval + flush_jitter

• flush_jitter: Jitter the flush interval by a random amount. This is primarily to avoid large write spikes for users running a large number of telegraf instances. ie, a jitter of 5s and interval 10s means flushes will happen every 10-15s

• precision: Collected metrics are rounded to the precision specified as an interval.

  Precision will NOT be used for service inputs. It is up to each individual service input to set the timestamp at the appropriate precision.

• debug: Run telegraf with debug log messages.

• quiet: Run telegraf in quiet mode (error log messages only).

• logfile: Specify the log file name. The empty string means to log to stderr.

• hostname: Override default hostname, if empty use os.Hostname()

• omit_hostname: If set to true, do no set the "host" tag in the telegraf agent.

Plugins

Telegraf plugins are divided into 4 types: inputs, outputs, processors, and aggregators.

Unlike the global_tags and agent tables, any plugin can be defined multiple times and each instance will run independantly. This allows you to have plugins defined with differing configurations as needed within a single Telegraf process.

Each plugin has a unique set of configuration options, reference the sample configuration for details. Additionally, several options are available on any plugin depending on its type.

Input Plugins

Input plugins gather and create metrics. They support both polling and event driven operation.

Parameters that can be used with any input plugin:

• interval: How often to gather this metric. Normal plugins use a single global interval, but if one particular input should be run less or more often, you can configure that here.
• name_override: Override the base name of the measurement. (Default is the name of the input).
• name_prefix: Specifies a prefix to attach to the measurement name.
• name_suffix: Specifies a suffix to attach to the measurement name.
• tags: A map of tags to apply to a specific input's measurements.

The metric filtering parameters can be used to limit what metrics are emitted from the input plugin.

Examples

Use the name_suffix parameter to emit measurements with the name cpu_total:

[[inputs.cpu]]
  name_suffix = "_total"
  percpu = false
  totalcpu = true

Use the name_override parameter to emit measurements with the name foobar:

[[inputs.cpu]]
  name_override = "foobar"
  percpu = false
  totalcpu = true

Emit measurements with two additional tags: tag1=foo and tag2=bar

NOTE: With TOML, order matters. Parameters belong to the last defined table header, place [inputs.cpu.tags] table at the end of the plugin definition.

[[inputs.cpu]]
  percpu = false
  totalcpu = true
  [inputs.cpu.tags]
    tag1 = "foo"
    tag2 = "bar"

Utilize name_override, name_prefix, or name_suffix config options to avoid measurement collisions when defining multiple plugins:

[[inputs.cpu]]
  percpu = false
  totalcpu = true

[[inputs.cpu]]
  percpu = true
  totalcpu = false
  name_override = "percpu_usage"
  fielddrop = ["cpu_time*"]

Output Plugins

Output plugins write metrics to a location. Outputs commonly write to databases, network services, and messaging systems.

Parameters that can be used with any output plugin:

• flush_interval: The maximum time between flushes. Use this setting to override the agent flush_interval on a per plugin basis.
• metric_batch_size: The maximum number of metrics to send at once. Use this setting to override the agent metric_batch_size on a per plugin basis.
• metric_buffer_limit: The maximum number of unsent metrics to buffer. Use this setting to override the agent metric_buffer_limit on a per plugin basis.

The metric filtering parameters can be used to limit what metrics are emitted from the output plugin.

Examples

Override flush parameters for a single output:

[agent]
  flush_interval = "10s"
  metric_batch_size = 1000

[[outputs.influxdb]]
  urls = [ "http://example.org:8086" ]
  database = "telegraf"

[[outputs.file]]
  files = [ "stdout" ]
  flush_interval = "1s"
  metric_batch_size = 10

Processor Plugins

Processor plugins perform processing tasks on metrics and are commonly used to rename or apply transformations to metrics. Processors are applied after the input plugins and before any aggregator plugins.

Parameters that can be used with any processor plugin:

• order: The order in which the processor(s) are executed. If this is not specified then processor execution order will be random.

The metric filtering parameters can be used to limit what metrics are handled by the processor. Excluded metrics are passed downstream to the next processor.

Examples

If the order processors are applied matters you must set order on all involved processors:

[[processors.rename]]
  order = 1
  [[processors.rename.replace]]
    tag = "path"
    dest = "resource"

[[processors.strings]]
  order = 2
  [[processors.strings.trim_prefix]]
    tag = "resource"
    prefix = "/api/"

Aggregator Plugins

Aggregator plugins produce new metrics after examining metrics over a time period, as the name suggests they are commonly used to produce new aggregates such as mean/max/min metrics. Aggregators operate on metrics after any processors have been applied.

Parameters that can be used with any aggregator plugin:

• period: The period on which to flush & clear each aggregator. All metrics that are sent with timestamps outside of this period will be ignored by the aggregator.
• delay: The delay before each aggregator is flushed. This is to control how long for aggregators to wait before receiving metrics from input plugins, in the case that aggregators are flushing and inputs are gathering on the same interval.
• drop_original: If true, the original metric will be dropped by the aggregator and will not get sent to the output plugins.
• name_override: Override the base name of the measurement. (Default is the name of the input).
• name_prefix: Specifies a prefix to attach to the measurement name.
• name_suffix: Specifies a suffix to attach to the measurement name.
• tags: A map of tags to apply to a specific input's measurements.

The metric filtering parameters can be used to limit what metrics are handled by the aggregator. Excluded metrics are passed downstream to the next aggregator.

Examples

Collect and emit the min/max of the system load1 metric every 30s, dropping the originals.

[[inputs.system]]
  fieldpass = ["load1"] # collects system load1 metric.

[[aggregators.minmax]]
  period = "30s"        # send & clear the aggregate every 30s.
  drop_original = true  # drop the original metrics.

[[outputs.file]]
  files = ["stdout"]

Collect and emit the min/max of the swap metrics every 30s, dropping the originals. The aggregator will not be applied to the system load metrics due to the namepass parameter.

[[inputs.swap]]

[[inputs.system]]
  fieldpass = ["load1"] # collects system load1 metric.

[[aggregators.minmax]]
  period = "30s"        # send & clear the aggregate every 30s.
  drop_original = true  # drop the original metrics.
  namepass = ["swap"]   # only "pass" swap metrics through the aggregator.

[[outputs.file]]
  files = ["stdout"]

Metric Filtering

Metric filtering can be configured per plugin on any input, output, processor, and aggregator plugin. Filters fall under two categories: Selectors and Modifiers.

Selectors

Selector filters include or exclude entire metrics. When a metric is excluded from a Input or an Output plugin, the metric is dropped. If a metric is excluded from a Processor or Aggregator plugin, it is skips the plugin and is sent onwards to the next stage of processing.

• namepass: An array of glob pattern strings. Only metrics whose measurement name matches a pattern in this list are emitted.

• namedrop: The inverse of namepass. If a match is found the metric is discarded. This is tested on metrics after they have passed the namepass test.

• tagpass: A table mapping tag keys to arrays of glob pattern strings. Only metrics that contain a tag key in the table and a tag value matching one of its patterns is emitted.

• tagdrop: The inverse of tagpass. If a match is found the metric is discarded. This is tested on metrics after they have passed the tagpass test.

Modifiers

Modifier filters remove tags and fields from a metric. If all fields are removed the metric is removed.

• fieldpass: An array of glob pattern strings. Only fields whose field key matches a pattern in this list are emitted.

• fielddrop: The inverse of fieldpass. Fields with a field key matching one of the patterns will be discarded from the metric. This is tested on metrics after they have passed the fieldpass test.

• taginclude: An array of glob pattern strings. Only tags with a tag key matching one of the patterns are emitted. In contrast to tagpass, which will pass an entire metric based on its tag, taginclude removes all non matching tags from the metric. Any tag can be filtered including global tags and the agent host tag.

• tagexclude: The inverse of taginclude. Tags with a tag key matching one of the patterns will be discarded from the metric. Any tag can be filtered including global tags and the agent host tag.

Filtering Examples

Using tagpass and tagdrop:

[[inputs.cpu]]
  percpu = true
  totalcpu = false
  fielddrop = ["cpu_time"]
  # Don't collect CPU data for cpu6 & cpu7
  [inputs.cpu.tagdrop]
    cpu = [ "cpu6", "cpu7" ]

[[inputs.disk]]
  [inputs.disk.tagpass]
    # tagpass conditions are OR, not AND.
    # If the (filesystem is ext4 or xfs) OR (the path is /opt or /home)
    # then the metric passes
    fstype = [ "ext4", "xfs" ]
    # Globs can also be used on the tag values
    path = [ "/opt", "/home*" ]

[[inputs.win_perf_counters]]
  [[inputs.win_perf_counters.object]]
    ObjectName = "Network Interface"
    Instances = ["*"]
    Counters = [
      "Bytes Received/sec",
      "Bytes Sent/sec"
    ]
    Measurement = "win_net"
  # Don't send metrics where the Windows interface name (instance) begins with isatap or Local
  [inputs.win_perf_counters.tagdrop]
    instance = ["isatap*", "Local*"]

Using fieldpass and fielddrop:

# Drop all metrics for guest & steal CPU usage
[[inputs.cpu]]
  percpu = false
  totalcpu = true
  fielddrop = ["usage_guest", "usage_steal"]

# Only store inode related metrics for disks
[[inputs.disk]]
  fieldpass = ["inodes*"]

Using namepass and namedrop:

# Drop all metrics about containers for kubelet
[[inputs.prometheus]]
  urls = ["http://kube-node-1:4194/metrics"]
  namedrop = ["container_*"]

# Only store rest client related metrics for kubelet
[[inputs.prometheus]]
  urls = ["http://kube-node-1:4194/metrics"]
  namepass = ["rest_client_*"]

Using taginclude and tagexclude:

# Only include the "cpu" tag in the measurements for the cpu plugin.
[[inputs.cpu]]
  percpu = true
  totalcpu = true
  taginclude = ["cpu"]

# Exclude the "fstype" tag from the measurements for the disk plugin.
[[inputs.disk]]
  tagexclude = ["fstype"]

Metrics can be routed to different outputs using the metric name and tags:

[[outputs.influxdb]]
  urls = [ "http://localhost:8086" ]
  database = "telegraf"
  # Drop all measurements that start with "aerospike"
  namedrop = ["aerospike*"]

[[outputs.influxdb]]
  urls = [ "http://localhost:8086" ]
  database = "telegraf-aerospike-data"
  # Only accept aerospike data:
  namepass = ["aerospike*"]

[[outputs.influxdb]]
  urls = [ "http://localhost:8086" ]
  database = "telegraf-cpu0-data"
  # Only store measurements where the tag "cpu" matches the value "cpu0"
  [outputs.influxdb.tagpass]
    cpu = ["cpu0"]

Routing metrics to different outputs based on the input. Metrics are tagged with influxdb_database in the input, which is then used to select the output. The tag is removed in the outputs before writing.

[[outputs.influxdb]]
  urls = ["http://influxdb.example.com"]
  database = "db_default"
  [outputs.influxdb.tagdrop]
    influxdb_database = ["*"]

[[outputs.influxdb]]
  urls = ["http://influxdb.example.com"]
  database = "db_other"
  tagexclude = ["influxdb_database"]
  [ouputs.influxdb.tagpass]
    influxdb_database = ["other"]

[[inputs.disk]]
  [inputs.disk.tags]
    influxdb_database = "other"