← LaunchDarkly renewal off-ramp

What Is a LaunchDarkly Service Connection?

It's the line item that quietly drives your LaunchDarkly bill, and the one finance teams most often can't explain. Here's exactly what a service connection is, how to count yours, and why it costs almost nothing to actually serve.

The short version

A service connection is one running process that has initialized a server-side LaunchDarkly SDK and is holding a live streaming connection to receive flag updates. One backend pod = roughly one service connection. Ten replicas of a service = ten connections. Background workers, cron jobs, and sidecars that evaluate flags each count too.

On LaunchDarkly's current pricing, each one is billed at $12 per connection per month. That's $144/year per process, so 30 pods is $4,320/year before you've served a single user.

Why it costs LaunchDarkly almost nothing

Here's the part worth understanding: a service connection is just a long-running, open GET request. The SDK opens an HTTP streaming connection (server-sent events) and holds it so the server can push flag changes down. Between changes, it sits idle. No compute, no per-request work, just an open socket.

Maintaining idle open connections is one of the cheapest things a modern server can do. Charging $12/month for each one is a pricing decision, not an infrastructure cost. It's a meter that scales with how your architecture happens to be deployed, more microservices, more pods, higher bill, even if your flag usage hasn't changed at all.

How to count your service connections

You want the number of running processes that initialize a server-side flag SDK. Three ways to get it, depending on who you are:

1. Count your pods (engineering)

If you're on Kubernetes, sum the replicas of every deployment that uses flags:

kubectl get deployments -A \
  -o custom-columns='NS:.metadata.namespace,NAME:.metadata.name,REPLICAS:.status.replicas'

# Or a quick total across the cluster:
kubectl get pods -A --field-selector=status.phase=Running --no-headers | wc -l

Then narrow that to the services that actually initialize the LaunchDarkly SDK. Don't forget autoscaled deployments, your connection count is the peak replica count, not the idle one.

2. Check your metrics (platform / SRE)

If you export SDK or runtime metrics, your Grafana dashboards already have this. Look for a gauge of active SDK connections, or process/replica counts for flag-using services. The LaunchDarkly dashboard itself also graphs active service connections over time, check the connections panel and read off the peak.

3. Ask your engineering team (finance / product)

If you're trying to forecast or sanity-check a renewal quote and you don't run the infrastructure, you don't need to. Send your platform or engineering team one question:

"Across all environments, how many running processes initialize the LaunchDarkly SDK at peak? Roughly the sum of replicas for every service that uses flags."

That number, times $12, times 12 months, is your annual service-connection spend, before MAU charges and add-ons.

How ShipSilently prices this differently

We charge $0 per service connection. Open as many as your architecture needs. The only thing that costs anything to serve is an actual edge evaluation: a real flag decision, and even that is bundled into one flat monthly price ($49/month on Pro). No connection meter, no MAU meter, no add-on surcharges.

That's the whole pitch: we don't bill you for how your services happen to be deployed. See the full breakdown on the pricing page, the SDK details in the docs, or run the numbers on the renewal comparison calculator.

Questions teams ask