be06105f20 | ||
---|---|---|
mqtt2prom | ||
webserv | ||
.gitignore | ||
Makefile | ||
Readme.md | ||
docker-compose.yml |
Readme.md
mqtt2prom
This service is a mqtt-to-prometheus gateway.
it is configured with the following ENV-Variables, the ones suffiexed with a *
are mandatory, default values are after |
:
MQTT_HOST*
MQTT_PORT|1883
MQTT_USER
MQTT_PASS
MQTT_CLIENT_ID|mqtt2prometheus
MQTT_TOPIC|prometheus
MQTT_QOS|2
IGNORE_RETAINED|1
IGNORE_RETAINED
: when set, retained messages are dropped and will be reported as errors.
it will listen on the configured mqtt server at $MQTT_TOPIC
for metrics in json_format.
The payload needs to be an object containing only the following keys:
name
(string): the name of the metricvalue
(number): the value of the metriclabels
(object, optional): an object containing the labels for the metric (key-value pairs of strings)
According to prometheus conventions, the metric name and the label names need to be a lower or upper case letter or a underscore followed by all case leters, numbers or underscores.
These are all valid payloads:
{
"name": "temperature",
"value": 30.7,
"labels": {
"unit": "°C",
"location": "Werkstatt"
}
}
{
"name": "temperature",
"value": 30.7,
"labels": {}
}
{
"name": "temperature",
"value": 30.7
}
The service will expose the metrics on the /metrics
endpoint.
All metrics will be presented for about 5 minutes, and exported with their eventtime as a timestamp.
Metric types are not currently implemented, all metrics are exported without a type. This might change when prometheus supports it.
The service will report errors in received messages to STDERR and to $MQTT_TOPIC/error. Possible errors are:
- invalid json
- missing
name
orvalue
key - invalid
name
(prometheus conventions) - invalid
value
(not a number) - invalid
labels
(not an object) - invalid label names (not a sting or not a valid prometheus label name)
- invalid label values (not a string)
- additional information in the payload (not one of
name
,value
,labels
)
The error message describes the error type and repeats the payload that caused the error.
There is a comment block on top if the metric that shows some metadata for human consumption.
mem usage and mem usage peak is according to memory_get_usage() and memory_get_peak_usage() respectively. in both cases it gives 2 values, the first is the acutally used value, the second is tha allocated.
A new export is written either
- once oper second if new events are incoming.
- every 15 seconds if no new events are incoming, to update stats and remove stale entries.
There is a 60 second period on startup where metrics are received but not exported, in case the service crashes and there are previous exported metrics that are not polled by prometheus yet.