StatsD Receiver
StatsD receiver for ingesting StatsD messages(https://github.com/statsd/statsd/blob/master/docs/metric_types.md) into the OpenTelemetry Collector.
Use case: it does not support horizontal pool of collectors. Desired work case is that customers use the receiver as an agent with a single input at the same time.
Configuration
The Following settings are optional:
-
endpoint
: Address and port to listen on.
- For
udp
and tcp
based transport
, this config will default to localhost:8125
- For
unixgram
transport
, this config will default to /var/run/statsd-receiver.sock
-
transport
(default = udp
): Protocol used by the StatsD server. Currently supported transports can be found in this file.
-
aggregation_interval: 70s
(default value is 60s): The aggregation time that the receiver aggregates the metrics (similar to the flush interval in StatsD server)
-
enable_metric_type: true
(default value is false): Enable the statsd receiver to be able to emit the metric type(gauge, counter, timer(in the future), histogram(in the future)) as a label.
-
enable_ip_only_aggregation
(default value is false): Enables metric aggregation on Client+IP
only. Normally, aggregation is performed on Client+IP+Port
. This setting is useful when the client sends metrics from a random ports or the receiver should aggregate metrics from the same client but different ports.
-
enable_simple_tags: true
(default value is false): Enable parsing tags that do not have a value, e.g. #mykey
instead of #mykey:myvalue
. DogStatsD supports such tagging.
-
is_monotonic_counter
(default value is false): Set all counter-type metrics the statsd receiver received as monotonic.
-
timer_histogram_mapping:
(default value is below): Specify what OTLP type to convert received timing/histogram data to.
"statsd_type"
specifies received Statsd data type. Possible values for this setting are "timing"
, "timer"
, "histogram"
and "distribution"
.
"observer_type"
specifies OTLP data type to convert to. We support "gauge"
, "summary"
, and "histogram"
. For "gauge"
, it does not perform any aggregation.
For "summary
, the statsD receiver will aggregate to one OTLP summary metric for one metric description (the same metric name with the same tags). By default, it will send percentile 0, 10, 50, 90, 95, 100 to the downstream. The "histogram"
setting selects an auto-scaling exponential histogram configured with only a maximum size, as shown in the example below.
TODO: Add a new option to use a smoothed summary like Prometheus: https://github.com/open-telemetry/opentelemetry-collector-contrib/pull/3261
Example:
receivers:
statsd:
statsd/2:
endpoint: "localhost:8127"
aggregation_interval: 70s
enable_metric_type: true
is_monotonic_counter: false
timer_histogram_mapping:
- statsd_type: "histogram"
observer_type: "gauge"
- statsd_type: "timing"
observer_type: "histogram"
histogram:
max_size: 100
- statsd_type: "distribution"
observer_type: "summary"
summary:
percentiles: [0, 10, 50, 90, 95, 100]
The full list of settings exposed for this receiver are documented here
with detailed sample configurations here.
Aggregation
Aggregation is done in statsD receiver. The default aggregation interval is 60s. The receiver only aggregates the metrics with the same metric name, metric type, label keys and label values. After each aggregation interval, the receiver will send all metrics (after aggregation) in this aggregation interval to the following workflow.
It supports:
Counter(transferred to int):
- statsdTestMetric1:3000|c|#mykey:myvalue
statsdTestMetric1:4000|c|#mykey:myvalue
(get the value after incrementation: 7000)
- statsdTestMetric1:3000|c|#mykey:myvalue
statsdTestMetric1:20|c|@0.25|#mykey:myvalue
(get the value after incrementation with sample rate: 3000+20/0.25=3080)
When the receiver receives valid sample rate (greater than 0 and less than 1), we covert the count value to float, divide by the sample rate and then covert back to integer.
The official doc does not support negative counter, we follow this pattern at this time. There are some requests for negative counters, we need to ake a look if we want to support later. For example:
https://github.com/influxdata/telegraf/issues/1898
https://thenewstack.io/collecting-metrics-using-statsd-a-standard-for-real-time-monitoring/
https://docs.datadoghq.com/developers/metrics/dogstatsd_metrics_submission/#count
Gauge(transferred to double):
- statsdTestMetric1:500|g|#mykey:myvalue
statsdTestMetric1:400|g|#mykey:myvalue
(get the latest value: 400)
- statsdTestMetric1:500|g|#mykey:myvalue
statsdTestMetric1:+2|g|#mykey:myvalue
statsdTestMetric1:-1|g|#mykey:myvalue
(get the value after calculation: 501)
Metrics
General format is:
<name>:<value>|<type>|@<sample-rate>|#<tag1-key>:<tag1-value>,<tag2-k/v>
Counter
<name>:<value>|c|@<sample-rate>|#<tag1-key>:<tag1-value>
It supports sample rate.
TODO: Need to change the implementation part for sample rate after OTLP supports sample rate as a parameter later.
Gauge
<name>:<value>|g|@<sample-rate>|#<tag1-key>:<tag1-value>
Timer
<name>:<value>|ms|@<sample-rate>|#<tag1-key>:<tag1-value>
<name>:<value>|h|@<sample-rate>|#<tag1-key>:<tag1-value>
It supports sample rate.
Testing
Full sample collector config
receivers:
statsd:
endpoint: "localhost:8125" # default
aggregation_interval: 60s # default
enable_metric_type: false # default
is_monotonic_counter: false # default
timer_histogram_mapping:
- statsd_type: "histogram"
observer_type: "histogram"
histogram:
max_size: 50
- statsd_type: "distribution"
observer_type: "histogram"
histogram:
max_size: 50
- statsd_type: "timing"
observer_type: "summary"
exporters:
file:
path: ./test.json
service:
pipelines:
metrics:
receivers: [statsd]
exporters: [file]
Send StatsD message into the receiver
A simple way to send a metric to localhost:8125
:
echo "test.metric:42|c|#myKey:myVal" | nc -w 1 -u -4 localhost 8125;
echo "test.metric:42|c|#myKey:myVal" | nc -w 1 -u -6 localhost 8125;
Which sends a UDP packet using both IPV4 and IPV6, which is needed because the receiver's UDP server only accepts one or the other.