Prometheus SQL

Service that generates basic metrics for SQL result sets and exposing them as Prometheus metrics.

This service relies on the SQL Agent service to execute and return the SQL result sets.

Status

This is a prototype and may be merged into the SQL Agent repo directly.

Behavior

A static configuration file is used to define the queries to monitor.
Each query has a designated worker for execution.
An interval is used to define how often to execute the query.
Failed queries are automatically retried using a backoff mechanism.
The result set of each query is held in memory for a subsequent comparison.
Metrics are emitted for each query using a label (see caveats below).
The state of a query is exposed at /state/<query>

Install

go get -u github.com/chop-dbhi/prometheus-sql

Usage

Create a queries.yml file in the current directory that defines the set of named queries to monitor (see the example file in this repository).

prometheus-sql

or for an alternate path, use the -queries option:

prometheus-sql -queries /path/to/queries.yml

Docker

Run the SQL agent service.

docker run -d --name sqlagent dbhi/sql-agent

Run this service. Mount the queries.yml file and link the SQL Agent service.

docker run -d \
    -p 8080:8080 \
    -v /path/to/queries.yml:/queries.yml \
    --link sqlagent:sqlagent \
    dbhi/prometheus-sql

To view a plain text version of the metrics, open up the browser to the http://localhost:8080 (or http://192.168.59.103:8080/metrics for boot2docker users).

Compose

Alternately, use the docker-compose.yml file included in this repository. The volumes section be added for mounting the queries.yml file.

Caveats

Currently, the Prometheus client library does not support creating a separate registry (of metrics) for each query, so metrics are differentiated using a query:<name> label. The side effect to this is that there is a single endpoint for all queries rather than one endpoint per query. See prometheus/client_golang#46 for details.

ncabatoff / prometheus-sql