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
Running GrowthBook?
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.
What it is
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.
The honest version
No technology is all upside. This is where GrowthBook earns its place and where it tends to cause trouble.
Sound familiar?
The GrowthBook problems that bring people to us, and what you have once each one is fixed.
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
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
The SDK was integrated without a trackingCallback, so the warehouse never receives exposure events and the experiment analysis cannot compute metrics.
Exposure pipeline complete
A metric definition joins across large tables, the query exceeds the timeout, and the experiment dashboard never populates.
Queries tuned, metrics loading
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
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
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
Hire us for one stuck experiment, or let us run the whole self-hosted platform.
How it works
Tell us what's going wrong. We look at it, then tell you plainly what's wrong and what it will cost.
We fix the cause, not the symptom. If something is down, getting it back up comes first.
Backups, monitoring, updates and access, so the same problem can't quietly come back.
You get a plain-language write-up. Or we stay on for hosting, upkeep and new work.
Often found together
We work on the whole stack, so the problem can be wherever it actually is.
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.
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.
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.
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.
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.
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
Tell us what you're running and what's going wrong. A person reads it and comes back with an estimate.
Your message is with a real person, and we'll get back to you soon.
Rather talk now? Call us: