Running GrowthBook?

GrowthBook self-hosted or sticky bucketing off? We fix it.

GrowthBook gives you LaunchDarkly-shaped functionality for a fraction of the cost, until the SDK proxy drifts out of sync with the control plane, or the sticky bucketing is subtly different from what your analytics is counting. We pick up the self-hosted pieces, the bucket drift and the data-warehouse connections and get the platform telling the same story on every side.

  • Since 2015Fixing technology
  • FreeEstimates
  • US teamReal engineers

growthbook-sdk.log

still failing

What it is

GrowthBook at a glance

GrowthBook is an open-source feature flag and experimentation platform. The hosted and self-hosted options share the same source code, which is a differentiator against the proprietary competition. It launched in 2021 and has grown steadily with teams that want to own their experimentation data in their own warehouse rather than send it to a vendor.

GrowthBook sits on top of your data warehouse (Snowflake, BigQuery, Redshift, Postgres, Mixpanel, ClickHouse) rather than keeping exposure data in its own silo. That is a strength for teams that already invested in a warehouse, and an operational cost for teams that have not.

Type
Open-source feature flags and experimentation
First released
2021
License
MIT core, with Enterprise features under commercial license
Maintained by
GrowthBook, Inc.
Typically used for
Feature flags, A/B tests, bandits and experimentation on top of your own data warehouse

The honest version

GrowthBook pros and cons

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

Where GrowthBook is strong

  • Open source and self-hostableThe MIT-licensed core runs on your own infrastructure, which is a strong story for teams that do not want exposure data leaving the environment.
  • Warehouse-native experimentationExposure and metric data live in your warehouse. You do not pay a vendor to store what you already store, and queries happen in a system your analysts trust.
  • SDK coverage is solidOfficial SDKs for the major server and client languages, with clean APIs and small footprints.
  • Sticky bucketing is first-classA user placed in a variant stays in that variant across sessions, which is what you want for most product experiments.
  • Visual editor for marketing experimentsA browser-based editor lets marketing and product tweak copy and layout for experiments without a deploy.
  • Reasonable pricingThe free self-hosted tier is actually useful, and the paid tiers are not priced on per-MAU.

Where GrowthBook bites

  • Self-hosting is yours to operateMongoDB, the API, the proxy and the webhook delivery are all pieces you have to run, watch, back up and upgrade.
  • Smaller ecosystem than LaunchDarklyFewer integrations out of the box, fewer worked examples online, and more pattern work falls to you.
  • Warehouse query cost is not zeroEvery metric query runs against your warehouse. Experiments running continuously can add a noticeable line to your Snowflake or BigQuery bill.
  • Sticky bucket store can biteIf the configured store is missing or slow, the SDK generates a fresh bucket assignment each time, and exposure data gets noisy in ways that take a while to notice.
  • Visual editor is fragile against real pagesSelectors that worked at design time break when the page changes, and an experiment can look live while serving the control.
  • Upgrade path requires careSelf-hosted installs that skip releases encounter migrations, schema changes and SDK compatibility jumps that the hosted version absorbs for you.

Sound familiar?

Common GrowthBook problems we fix

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

Proxy: connection to control plane refused

SDK proxy stops serving current features

The proxy cannot reach the GrowthBook control plane, so it serves the last cached feature payload. Flag changes sit undelivered until someone restarts the proxy.

Proxy reconnected, features current

Sticky bucket store returned null

Sticky bucketing is inconsistent

The configured bucket store was unreachable or empty, so users were re-bucketed on each visit, scrambling experiment reads for days before anyone noticed.

Users bucketed once and stably

trackingCallback not set

Exposures are not being logged

The SDK was integrated without a trackingCallback, so the warehouse never receives exposure events and the experiment analysis cannot compute metrics.

Exposure pipeline complete

Snowflake query timed out 300s

Warehouse queries time out for metrics

A metric definition joins across large tables, the query exceeds the timeout, and the experiment dashboard never populates.

Queries tuned, metrics loading

MongoDB pool exhausted

Self-hosted MongoDB is overloaded

A small MongoDB instance sized for the first week of use now serves a year of feature history, and connections pile up until the API starts returning errors.

Database resized and tuned

Visual editor: target not found

Visual editor experiment is serving the control

A page's markup changed since the experiment was authored, the selector no longer matches, and visitors are silently falling into the control.

Selectors robust, experiment live

Webhook delivery failed 500

Webhooks are not delivering feature changes

A downstream service that is supposed to react to flag changes is 500ing, and GrowthBook keeps retrying until the queue backs up.

Webhook consumers reliable

What we do

GrowthBook help, from self-host to experiment analysis

Hire us for one stuck experiment, or let us run the whole self-hosted platform.

Rescue
Proxy out of sync, sticky bucket store misbehaving, or an experiment that has been running on broken selectors for a week. We stabilize the feature delivery first and take the analysis apart next.
Self-host operation
Deploying GrowthBook in a form your platform team can live with: real database sizing, real monitoring, and a documented upgrade path.
SDK integration
Tracking callbacks that match your analytics, sticky bucket stores wired to Redis or Postgres, and fallback values that fail safely under network trouble.
Warehouse connection
Snowflake, BigQuery, Redshift or ClickHouse connections tuned so metric queries return in seconds rather than timing out.
Experiment analysis review
Reading the results the platform produces, checking for sample ratio mismatch and guardrail violations, and recommending whether the test should ship or hold.
Managed Tech
A standing engagement that keeps the self-hosted install upgraded, the warehouse queries fast, and the experiments defensible when a stakeholder asks.

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 GrowthBook alongside something else?

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

Questions

GrowthBook questions

Something else? Call+1 737 727 7327

Should we self-host GrowthBook?

Only if you want the data to stay on your side of the network and you have the people to operate another MongoDB-backed service. For most teams the hosted tier is faster to adopt and cheaper overall when you include the cost of running it yourself.

Why are our GrowthBook experiments showing unstable results?

In our experience the top causes are a sticky bucket store that was not actually configured, a tracking callback that drops exposures during high traffic, and a visual-editor experiment with a selector that no longer matches the DOM. We verify each piece and rerun the analysis on the known-good exposure data.

How do we keep our warehouse bill from exploding with GrowthBook queries?

Materialise metric sources into pre-aggregated tables, cap the time window a query will scan, and schedule heavy metrics to refresh on a cadence rather than on each dashboard open. We review the SQL GrowthBook generates and rewrite the hot queries.

Can we move from LaunchDarkly to GrowthBook?

Yes, and the cost argument usually makes the project worthwhile at scale. The work is mapping each flag and segment into the GrowthBook model, adapting the tracking callback to the analytics you already have, and running both platforms in parallel until you trust the GrowthBook outputs.

Why does our visual editor experiment show no effect?

Because it is probably not running on the visitors you think it is. Usually the DOM selector stopped matching after a release, and GrowthBook is quietly serving the control. We rerun the experiment against a stable selector or move the change into code.

How much does GrowthBook help cost?

The quote is free. Send over the self-hosted error log, the experiment that is behaving oddly, or the SDK integration step, and we price the work before making any changes.

Free estimate

GrowthBook 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.