lexigram-monitor

Observability, health checks, and metrics for the Lexigram Framework. Supports Prometheus, OpenTelemetry, structured log export, and /health endpoints that integrate with Kubernetes probes and load-balancer health checks.

Overview

lexigram-monitor provides metrics collection, distributed tracing, health checks, and alerting for Lexigram applications. It integrates with Prometheus and OpenTelemetry backends, supports composable health checks with liveness and readiness flavours, and includes decorators for instrumenting services with custom metrics and traces. All services are wired via MonitorProvider, which registers monitoring protocols with the DI container.

Full documentation: docs.lexigram.dev

Install

uv add lexigram-monitor
# Optional extras
uv add "lexigram-monitor[prometheus]"    # Prometheus + Grafana
uv add "lexigram-monitor[opentelemetry]" # OTLP / Jaeger / Zipkin

Quick Start

from lexigram import Application
from lexigram.monitor import MonitorModule


async def main() -> None:
    async with Application.boot(modules=[MonitorModule.configure()]) as app:
        # ... metrics, health checks and /health endpoints active ...
        ...


if __name__ == "__main__":
    import asyncio

    asyncio.run(main())

Configuration

Endpoint protection

HealthCheckProvider and PrometheusMiddleware expose their endpoints (/health and /metrics by default) without authentication — the intentional default, because Kubernetes probes and Prometheus scrapers usually run inside a trusted network and cannot always carry credentials.

If these endpoints are reachable from outside that boundary, pass an auth_token to require Authorization: Bearer <token> on every request:

from lexigram.monitor.middleware import HealthCheckProvider, PrometheusMiddleware

app = HealthCheckProvider(path="/health", auth_token=os.environ["HEALTH_TOKEN"])
app = PrometheusMiddleware(app, path="/metrics", auth_token=os.environ["METRICS_TOKEN"])

Requests without the matching token receive 401 with WWW-Authenticate: Bearer. Configure the same token on the scraper side (e.g. Prometheus scrape_configs → authorization.credentials).

Failed dependency checks never echo the raw driver message into the JSON health payload — the response carries only the exception type name ("ConnectionError: connection check failed"), while the full message is written to the application logs.

Module Factory Methods

Key Features

• Prometheus — Auto /metrics endpoint; request counters, histograms, gauges
• OpenTelemetry — Distributed tracing via OTLP exporter to Jaeger / Honeycomb
• Health checks — Composable checks with liveness + readiness flavours
• Cached checks — Per-check TTL to avoid thundering-herd on slow dependencies
• DB instrumentation — Automatic query timing and error tagging
• HTTP instrumentation — Outbound request tracking for lexigram-http
• Messaging instrumentation — Kafka / RabbitMQ consumer lag, publish rate
• Alerting — Configurable alert rules with tier-aware webhook delivery
• SLO Monitoring — Burn-rate evaluation with configurable suppression window
• Tiered alerts — P0 (PagerDuty) / P1 (business hours Slack) / P2 (weekly digest) routing
• Structured logging — json / text log output via logging.level / logging.format
• Grafana dashboards — Pre-built dashboard JSON in lexigram-monitor/dashboards/

Testing

async with Application.boot(modules=[MonitorModule.stub()]) as app:
    # your test code
    ...

Key Source Files

SLO Monitoring

Service Level Objectives are evaluated on a configurable interval. Each SLO tracks a metric percentile against a threshold and fires alerts on budget exhaustion.

Defining an SLO

from datetime import timedelta
from lexigram.contracts.monitor import ProjectionTier
from lexigram.monitor.slo import SLO, SLOMonitor

monitor = SLOMonitor()

slo = SLO(
    name="api.p99_latency",
    metric="http.request.duration",
    percentile=0.99,
    threshold_ms=200.0,
    window=timedelta(hours=1),
    tier=ProjectionTier.P1_BUSINESS_HOURS,
    owner="team-api",
    runbook_url="https://ops.runbook/api-slo",
)
monitor.register(slo)

Recording Samples

monitor.record_sample("http.request.duration", 150.0)
monitor.record_sample("http.request.duration", 350.0)

Evaluating and Dispatching

violations = await monitor.evaluate_and_dispatch()

Violations are routed through the configured AlertDispatcherProtocol. Alerts for the same SLO are suppressed within the suppression window (default 300s) to avoid storms.

Projection Tiers

Worker Configuration

Enable periodic evaluation via config:

# application.yaml
monitor:
  slo:
    enabled: true
    evaluation_interval: 60
    suppression_window_seconds: 300

Or via environment variables:

export LEX_MONITOR__SLO__ENABLED=true
export LEX_MONITOR__SLO__EVALUATION_INTERVAL=60
export LEX_MONITOR__SLO__SUPPRESSION_WINDOW_SECONDS=300