criteo / blackbox-prober

Blackbox prober - A prometheus blackbox probe for distributed systems

Introduction

The blackbox prober is both a framework for probes and also the implementation of probes for the following databases:

Database	Maturity
Aerospike	RC
Memcached	Alpha

Coming soon:

Database
Elasticsearch
Cassandra/ScyllaDB

Checks

The goal of the probe is to perform checks against multiple distributed systems clusters of the same kind. Those distributed system are typically databases.

Checks can be used to compute SLIs:

• Availability
• Latency
• Durability (dataloss)
• Replication durability/delay

Check can be executed at two levels:

• Node: executed every interval per node on each clusters. Useful to calculate SLIs about a single node (such as latency)

• Cluster: executed every interval per cluster. Useful to calculate SLIs about a whole cluster (such as durability)

Usage

Building

make build

Running

$ build/aerospike_probe -h
usage: aerospike_probe [<flags>]

Aerospike blackbox probe

Flags:
  -h, --help               Show context-sensitive help (also try --help-long and --help-man).
      --web.listen-address="0.0.0.0:8080"  
                           Address to listen on for UI, API, and telemetry.
      --config.path="config.yaml"  
                           Path to the probe configuration file
      --log.level=info     Only log messages with the given severity or above. One of: [debug, info, warn, error]
      --log.format=logfmt  Output format of log messages. One of: [logfmt, json]

Testing

make test

Architecture

Topology

The probe works with the concept of topology:

A ClusterMap contains the toplogy of the clusters discovered by the service discovery. The prober will schedule checks (either cluster or node level) for each endpoints.

Workflow

Adding your own probe

Conventions

Metrics should use seconds as unit or prefix with the unit (_ms)

Implementations

Probes are defined in the probes/<name of db> directory. The probes have to define 3 things:

Endpoints

A implementation of the toplogy.ProbeableEndpoint. It contains the client (and everything necessary) to connect to the database. State, locks or connections should not be shared between endpoints as it may affect recorded latencies.

See probes/aerospike/endpoint.go for an example

Endpoint builder

The discovery is responsible for creating the topology. When creating a discoverer, a builder function should be provided: topologyBuilderFn func(log.Logger, []ServiceEntry) (topology.ClusterMap, error) This function takes all the service entries (services found by the service discovery) and return a topology. A generic version of this function is available:

GetGenericTopologyBuilder(
	ClusterFn func(log.Logger, []ServiceEntry) (topology.ProbeableEndpoint, error),
	NodeFn func(log.Logger, ServiceEntry) (topology.ProbeableEndpoint, error)

ClusterFn should generate a cluster level endpoint NodeFn should generate a node level endpoint

See probes/aerospike/discovery.go for an example

Checks

Example:

p.RegisterNewClusterCheck(scheduler.Check{
    Name:       "latency_check",
    PrepareFn:  scheduler.Noop,
    CheckFn:    scheduler.Noop,
    TeardownFn: scheduler.Noop,
    Interval:   config.AerospikeChecksConfigs.LatencyCheckConfig.Interval,
})

scheduler.Noop is a placeholder function

See probes/aerospike/main.go for an example