Running LaunchDarkly?

LaunchDarkly flags dropping or targeting broken? We ship it properly.

LaunchDarkly works beautifully until the streaming connection flaps on a release day, a targeting rule you did not know existed starts serving the wrong variant, or someone counts the flags in your codebase and realises nobody has deleted one in two years. We take the stale flags, the timed-out evals and the inherited rule trees, and put the system back under control.

  • Since 2015Fixing technology
  • FreeEstimates
  • US teamReal engineers

launchdarkly-sdk.log

still failing

What it is

LaunchDarkly at a glance

LaunchDarkly is the market-leading feature management platform, used for feature flags, targeted rollouts, A/B tests and kill switches. It was founded in 2014 and is the tool most engineering teams reach for when they want to decouple deploy from release.

The platform offers SDKs for every major server and client language, a streaming architecture that keeps flag state current without polling, a relay proxy for environments where you do not want every client reaching the SaaS, and an experimentation layer built on top of the same flag machinery. It is capable and heavily used, which means every mature install also carries years of stale flags, overlapping segments and targeting rules whose owner has since moved on.

Type
Feature management and experimentation platform
First released
2014
License
Proprietary SaaS, per-seat and MAU pricing
Maintained by
LaunchDarkly (Catamorphic Co.)
Typically used for
Feature flags, progressive rollouts, kill switches, experimentation, targeting by user, device or environment

The honest version

LaunchDarkly pros and cons

No technology is all upside. This is where LaunchDarkly earns its place and where it tends to cause trouble.

Where LaunchDarkly is strong

  • The SDK coverage is broadOfficial libraries for most server and client languages, including the awkward ones. Polyglot stacks rarely need a third-party wrapper.
  • Streaming keeps flags currentThe SDK keeps an open connection to the LaunchDarkly stream, so flag changes propagate in seconds rather than at the next poll interval.
  • Targeting is genuinely flexibleRules can combine user attributes, segments, prerequisite flags and percentage rollouts, which covers almost any release strategy you can describe.
  • Relay proxy for air-gapped and high-volumeThe relay proxy lets you keep SDK traffic inside your network and share a single connection to LaunchDarkly, which matters at scale.
  • Audit log is auditor-friendlyEvery flag change is recorded with author, time and before-and-after values, and the log exports cleanly for a compliance review.
  • Experimentation layer on the same primitivesIf you already flag a feature, promoting it to an experiment is a few clicks rather than a parallel integration.

Where LaunchDarkly bites

  • Pricing climbs quicklyMAU-based pricing with add-ons for experimentation and data export means the second-year invoice often doubles the first.
  • Flag debt accumulatesEvery team creates flags; very few teams delete them. A mature install carries hundreds of flags that nobody owns and nobody dares remove.
  • Streaming failures are subtleAn SDK that cannot reach the stream serves the fallback value, and the application behaves as if the flag is off. If the fallback is wrong, outages happen without a visible error.
  • Targeting rules get unreadableThree overlapping segments, a prerequisite flag and a rollout bucket together produce behavior that no single person in the room can predict.
  • Relay proxy is one more thing to runIf you need the relay you need to operate it, monitor it, and keep its TLS and cert verification working across versions.
  • Experiment sample sizes biteSmall-traffic features never reach significance, and dashboards that look conclusive are often reading noise.

Sound familiar?

Common LaunchDarkly problems we fix

The LaunchDarkly problems that bring people to us, and what you have once each one is fixed.

evaluation took longer than 5000 ms

Flag evaluation times out in production

The SDK could not reach the streaming endpoint at startup, the init timeout fired, and every evaluation falls back to the hardcoded default, which is almost never what you wanted.

SDK initialised, stream healthy

Streaming connection closed

Streaming connection keeps flapping

A proxy, firewall or load balancer is closing the long-lived connection to stream.launchdarkly.com, and the SDK is reconnecting every minute instead of staying open.

Streaming stable, flags live

Flag toggled 847 days ago

The flag catalog is full of dead flags

Flags created for a launch two years ago are still in the codebase and in the dashboard, with branching that nobody can trace. Removing the wrong one breaks something nobody remembers.

Flag catalog reviewed and pruned

Experiment: insufficient sample size

Experiments never reach significance

Traffic is too thin for the metric lift you are trying to detect, or the experiment was stopped too early on a noisy swing. Decisions get made on data that cannot support them.

Experiments powered or parked

Relay proxy upstream TLS

Relay proxy fails TLS verification

A certificate bundle update changed the chain, the relay proxy cannot verify upstream, and every SDK that routes through the relay starts serving defaults.

Relay proxy connected cleanly

Prerequisite flag off

Dependent flags behave unexpectedly

A flag's prerequisite was switched off months ago, and no one realised the downstream flags became dormant. A release went out and the feature never lit up.

Prerequisite chain understood

Rate limit exceeded on /api/v2/flags

The management API is being hammered

An automation syncing flag state polls too aggressively, exhausts the rate limit, and the UI and CI integration both grind.

API traffic shaped, limits respected

What we do

LaunchDarkly help, from rescue to upkeep

Reach out for one stalled release, or let us own the whole feature-flag program.

Rescue
Streaming down, flags serving defaults in production, or a rollout that the targeting rules refuse to honor. We stabilize the user-facing behavior first and read the logs afterward.
Flag cleanup
Inventorying every flag, sorting live from dormant, working with each owning team to retire the dead ones, and leaving a smaller catalog that the next release can trust.
SDK and streaming reliability
Getting the init, timeout, retry and fallback behavior right for your language and runtime, so an SDK that cannot reach the stream fails loudly rather than quietly.
Targeting review
Walking the rule trees, segments and prerequisites until the behavior of each flag is one that anyone on the team can describe without opening the UI.
Relay proxy operation
Deploying the relay proxy where it is actually needed, monitoring it, and keeping the TLS chain, versions and routing working across upgrades.
Managed Tech
A standing arrangement to prune the flag catalog, validate experiments, and respond when the SDK starts throwing timeouts at the worst possible hour.

How it works

From first message to fixed

  1. Look Free estimate

    Tell us what's going wrong. We look at it, then tell you plainly what's wrong and what it will cost.

  2. Fix

    We fix the cause, not the symptom. If something is down, getting it back up comes first.

  3. Harden

    Backups, monitoring, updates and access, so the same problem can't quietly come back.

  4. Hand back or stay on

    You get a plain-language write-up. Or we stay on for hosting, upkeep and new work.

Often found together

Running LaunchDarkly alongside something else?

We work on the whole stack, so the problem can be wherever it actually is.

Questions

LaunchDarkly questions

Something else? Call+1 737 727 7327

Why is my LaunchDarkly SDK serving default values?

The SDK could not reach the streaming endpoint in time, so initialization fell back to defaults. It usually means a network path is blocking the long-lived connection to stream.launchdarkly.com, or the SDK key is wrong for the environment. We trace the connection from the host, confirm the key, and raise the init timeout to a value that matches the real reachability.

How do we stop flag debt from piling up?

Flag cleanup only sticks when it is someone's job. We set a monthly rhythm, require every new flag to carry an expected end date, and automate reminders when a flag has been on or off for longer than planned. The dashboard also has a stale-flag view that becomes useful once a team commits to acting on it.

Do we need the LaunchDarkly relay proxy?

You need it if you cannot reach the SaaS from every pod or if the SDK fan-out is going to generate more connections than you want to pay for. For small services and a well-connected network, the relay is overkill. For a Kubernetes cluster with hundreds of pods it usually pays for itself in connection stability.

Why do our experiments never show a winner?

The common causes are insufficient traffic for the effect size, a metric that is too noisy, and tests stopped too early because stakeholders wanted to call it. We rework the metric, estimate the required sample at the start, and set a minimum duration before any read is treated as conclusive.

Can we move from LaunchDarkly to a cheaper option?

Technically yes, but it is a project. Flag semantics vary across vendors, targeting syntax rarely translates one for one, and the SDK init behavior of the replacement is almost always different. We model the cost gap honestly and only recommend the migration when it is material and the catalog is small enough to carry across.

How much does LaunchDarkly help cost?

Our estimate is free. Describe the flag that is misbehaving or the rollout that stalled, and we come back with a fixed price before touching the account.

Free estimate

LaunchDarkly giving you trouble?

Tell us what you're running and what's going wrong. A person reads it and comes back with an estimate.

  • No charge to find out what it will take
  • If it's urgent, say so

Replies come from a person. By sending, you authorize us to email you back. No marketing.