Integrations

SketchLog is built to fit neatly into modern observability stacks without creating architectural friction.

OpenTelemetry Exporter

You can seamlessly bridge SketchLog into any OpenTelemetry-compatible observability backend (like Jaeger, Datadog, Honeycomb, or New Relic) using the built-in OTLP asynchronous integration.

This exporter registers asynchronous OTel gauges and maps StreamLog percentiles, cardinality estimates, resettable totals, and selected event frequencies into OTLP metrics. Windowed logs create a bounded temporary merged sketch during an observation callback; ordinary and thread-safe logs are observed under their normal snapshot/locking semantics.

from sketchlog import StreamLog
from sketchlog.integrations.opentelemetry import SketchLogOTelPublisher

log = StreamLog()
# You can use the high-level publisher to immediately spin up a Periodic MetricReader
publisher = SketchLogOTelPublisher(
    streamlog=log,
    export_interval_millis=60000,
    export_events=["cache_miss", "db_query"]
)

log.add_latency(42.0)
log.add_unique("user_123")
log.add_event("cache_miss")

# At shutdown
publisher.shutdown()

If you prefer to wire it into an existing MeterProvider, you can instantiate OpenTelemetryAdapter directly.

Prometheus Exporter

Easily expose sketch metrics to Prometheus using the built-in HTTP exporter. It exports p50/p95/p99 quantiles, unique counts, total events, and memory usage.

from sketchlog import ThreadSafeStreamLog
from sketchlog.integrations.prometheus import PrometheusExporter

log = ThreadSafeStreamLog()
exporter = PrometheusExporter(log)

# Loopback is the secure default. Bind explicitly when Prometheus is remote.
exporter.start(host="0.0.0.0", port=9090)

# Add metrics in your main thread
log.add_latency(42.0)

The lightweight exporter has no authentication or TLS termination. When binding beyond loopback, restrict access with a firewall, sidecar, ingress, or service mesh.

FastAPI / Starlette Middleware

Track request latency, endpoint hits, and status codes across your entire API without configuring separate prometheus clients or statsd servers.

from fastapi import FastAPI
from sketchlog import StreamLog
from sketchlog.integrations.fastapi import SketchLogMiddleware

app = FastAPI()
log = StreamLog()

# Pass the log instance explicitly
app.add_middleware(SketchLogMiddleware, log=log)

@app.get("/api/users")
def get_users():
    return {"status": "ok"}