prometheus-kafka-druid-ingestion

At noris network we're sending Prometheus data via `prometheus-kafka-adapter to Kafka and then use Apache Druid's Kafka Ingestion for storing those metrics.

This repo creates an Apache Druid ingestion spec from prometheus-kafka-adapter data.

Background

Prometheus data can be sent via remote_write to prometheus-kafka-adapter, which in turn sends it to Kafka. The prometheus-kafka-adapter message will have the following structure:

{
  "timestamp": "1970-01-01T00:00:00Z",
  "value": "9876543210",
  "name": "up",

  "labels": {
    "__name__": "up",
    "label1": "value1",
    "label2": "value2"
  }
}

This data can be send to Apache Druid for long term storage, using Druid's Kafka Ingestion feature. For the data to be consumed and saved, we need to create an ingestion spec, including the final schema in the database and the Kafka brokers to consume the metrics from.

Install

From Source

Clone the repo and cd into it, then run make:

make build

The binary will be available in bin/generate-ingestion.

Usage

$ ./bin/generate-ingestion -h

Usage:
  generate-ingestion [flags]

Flags:
  -a, --address string             The address of the Prometheus server to send the query to (default "http://prometheus:9090")
  -d, --druid-data-source string   The druid data source (default "prometheus")
  -f, --file string                The file to save the ingestion spec to
  -h, --help                       help for generate-ingestion
      --ingest-via-ssl             Enables data ingestion from Kafka to Druid via SSL (default true)
  -b, --kafka-brokers string       The Kafka brokers for druid to ingest data from (default "kafka01:9092,kafka02:9092,kafka03:9092")
  -t, --kafka-topic string         The Kafka topic for druid to ingest data from (default "prometheus")
  -q, --query string               The query to send to the Prometheus server (default "{__name__=~\"job:.+\"}")
      --tls-skip-verify            Skip TLS certificate verification
  -o, --toStdout                   Prints the JSON ingestion spec to STDOUT (default true)

Executing the file sends the query specified with the -q / --query flag to a Prometheus server running at -a / --address. The script will then extract the unique labels of all time series returned and build an opinionated ingestion spec from those labels.

If all recording rules with the prefix job: are sent to prometheus-kafka-adapter via remote_write, the PromQL query {__name__=~"job:.+"} would retrieve those series.

By default Druid will consume messages from Kafka via SSL, meaning the following block will be populated in the spec:

            # ...
            "security.protocol": "SSL",
            "ssl.truststore.type": "PKCS12",
            "ssl.enabled.protocols": "TLSv1.2",
            "ssl.truststore.location": "/var/private/ssl/truststore.p12",
            "ssl.truststore.password": {
                "type": "environment",
                "variable": "DRUID_TRUSTSTORE_PASSWORD"
            },
            "ssl.keystore.location": "/var/private/ssl/keystore.p12",
            "ssl.keystore.password": {
                "type": "environment",
                "variable": "DRUID_KEYSTORE_PASSWORD"
            }
        },

This behaviour can be disabled with the --ingest-via-ssl=false flag.

By default the ingestion spec is displayed to stdout, but can be saved with the -f flag:

$ generate-ingestion -f ingestion.json

{
    "type": "kafka",
    "dataSchema": {
        "dataSource": "prometheus",
        "parser": {
            "type": "string",
            "parseSpec": {
                "format": "json",
                "timestampSpec": {
                    "column": "timestamp",
                    "format": "iso"
                },
                "flattenSpec": {
                    "fields": [
                        {
                            "type": "path",
                            "name": "clustername",
                            "expr": "$.labels.clustername"
                        },
                        {
                            "type": "path",
                            "name": "source",
                            "expr": "$.labels.source"
                        },
                        {
                            "type": "root",
                            "name": "name",
                            "expr": "name"
                        },
                        {
                            "type": "root",
                            "name": "value",
                            "expr": "value"
                        }
                    ]
                },
                "dimensionsSpec": {
                    "dimensions": [
                        "name",
                        "clustername",
                        "source"
                    ]
                }
            }
        },
        "metricsSpec": [
            {
                "name": "count",
                "type": "count"
            },
            {
                "name": "value",
                "type": "doubleMax",
                "fieldName": "value"
            }
        ],
        "granularitySpec": {
            "type": "uniform",
            "segmentGranularity": "HOUR",
            "queryGranularity": "MINUTE"
        }
    },
    "ioConfig": {
        "topic": "prometheus",
        "consumerProperties": {
            "bootstrap.servers": "kafka01:9092,kafka02:9092,kafka03:9092",
            "security.protocol": "SSL",
            "ssl.truststore.type": "PKCS12",
            "ssl.enabled.protocols": "TLSv1.2",
            "ssl.truststore.location": "/var/private/ssl/truststore.p12",
            "ssl.truststore.password": {
                "type": "environment",
                "variable": "DRUID_TRUSTSTORE_PASSWORD"
            },
            "ssl.keystore.location": "/var/private/ssl/keystore.p12",
            "ssl.keystore.password": {
                "type": "environment",
                "variable": "DRUID_KEYSTORE_PASSWORD"
            }
        },
        "taskDuration": "PT10M",
        "useEarliestOffset": true
    }
}