spacearth-metrics 1.1.0

Gather metrics from Python services using either AWS CloudWatch or Prometheus.

Metrics library

This repo contains a metrics library implemented in Go and Python. The goal of the library is to be transparent inside the deployment: if metrics are disabled, the application code does not change.

Installation

Python

pip install spacearth-metrics

Go

go get github.com/Spacearth-NAV/metrics-lib

Usage

Configuration

Each backend requires specific configuration before the server is initialized.

AWS

Set the following environment variables before starting your application:

• AWS_ACCESS_KEY_ID
• AWS_SECRET_ACCESS_KEY
• AWS_DEFAULT_REGION

Refer to the AWS SDK configuration documentation for more info.

Prometheus

The Prometheus backend starts an HTTP server that exposes metrics at /metrics.

• Python: port is set via the port keyword argument (default 8080).
• Go: port defaults to 8080 if not set via WithPort.

Note: Prometheus requires all label names for a metric to be declared upfront. The label schema is locked on the first call for each metric name. Any subsequent call with a different set of label keys will cause an error: a ValueError in Python, a panic in Go. Make sure to use the same label keys consistently across all calls to the same metric.

Note: Label keys passed at call-site must not overlap with fixed label keys. Passing a key that matches a fixed label key will cause a ValueError in Python and a panic in Go.

Security

The /metrics endpoint is served over plain HTTP with no authentication or TLS. This follows the standard Prometheus pull model, where the Prometheus server scrapes from within a trusted network. Do not expose the metrics port to untrusted networks. Restrict access at the network level (security groups, firewall rules, or a service mesh policy) so that only the Prometheus scraper can reach the port.

No-op

No configuration required. All calls are silently ignored.

---

Initialization

The library is designed to be transparent: initialization is the only place where the backend is chosen. All metric recording calls are identical regardless of the backend in use.

Python

import os
from spacearth.metrics import MetricServer

provider    = os.getenv("METRIC_PROVIDER", "aws")
namespace   = os.getenv("METRIC_NAMESPACE", "default")
environment = os.getenv("ENVIRONMENT", "development")

labels     = {"environment": environment}
extra_args = {}

# Prometheus requires a port
if provider == "prometheus":
    extra_args["port"] = int(os.getenv("PROMETHEUS_PORT", "8080"))

metric_server = MetricServer.create_server(provider, namespace, labels, **extra_args)
# AWS:        set AWS_ACCESS_KEY_ID, AWS_SECRET_ACCESS_KEY, AWS_DEFAULT_REGION
# Prometheus: set PROMETHEUS_PORT (default: 8080)
# No-op:      no configuration required

Go

import (
    "os"

    metrics "github.com/Spacearth-NAV/metrics-lib/go"
)

func envOr(key, fallback string) string {
    if v := os.Getenv(key); v != "" {
        return v
    }
    return fallback
}

provider    := envOr("METRIC_PROVIDER", "aws")
namespace   := envOr("METRIC_NAMESPACE", "default")
environment := envOr("ENVIRONMENT", "development")

opts := []metrics.Option{
    metrics.WithFixedLabels(metrics.Label{Key: "environment", Value: environment}),
}

if provider == "prometheus" {
    port, _ := strconv.Atoi(envOr("PROMETHEUS_PORT", "8080"))
    opts = append(opts, metrics.WithPort(port))
}

metricsServer, err := metrics.NewServer(metrics.ServerType(provider), namespace, opts...)
if err != nil {
    // handle error
}
// AWS:        set AWS_ACCESS_KEY_ID, AWS_SECRET_ACCESS_KEY, AWS_DEFAULT_REGION
// Prometheus: starts on :8080 by default; override with metrics.WithPort(port)
// No-op:      no configuration required

---

Recording metrics

All backends share the same interface. Fixed labels passed at initialization are automatically added to every metric.

Counters — add_observation / AddObservation

Records a single event count.

Python

metric_server.add_observation("requests_received", 1, labels={"endpoint": "/login"})

Go

metricsServer.AddObservation("requests_received", 1, metrics.Label{"endpoint", "/login"})

Histograms — measure_time / MeasureTime

Records a duration. Python accepts seconds as a float; Go accepts a time.Duration.

Python

import time

t_start = time.time()
# ... do work ...
metric_server.measure_time("processing_time", time.time() - t_start, labels={"step": "auth"})

Go

start := time.Now()
// ... do work ...
metricsServer.MeasureTime("processing_time", time.Since(start), metrics.Label{"step", "auth"})

Gauges — increment_value / decrement_value / set_value

Tracks a value that goes up and down.

Python

def on_connection(conn):
    metric_server.increment_value("active_connections", labels={"endpoint": "/ws"})
    try:
        while conn.connected:
            pass
    finally:
        metric_server.decrement_value("active_connections", labels={"endpoint": "/ws"})

# or set an absolute value
metric_server.set_value("queue_depth", 42)

Go

metricsServer.IncrementValue("active_connections", 1, metrics.Label{"endpoint", "/ws"})
metricsServer.DecrementValue("active_connections", 1, metrics.Label{"endpoint", "/ws"})

metricsServer.SetValue("queue_depth", 42)

---

Known limitations

No graceful shutdown (Prometheus backend)

The Prometheus backend starts an HTTP server in a background goroutine. Neither the Server interface nor any concrete implementation exposes a Close or Shutdown method. The HTTP listener is held until the process exits.

If your application needs to stop the metrics server cleanly — for example, in integration tests that create multiple servers — this must be handled at the process level (e.g. via os.Signal → os.Exit), not through this library.