Running PostHog?

PostHog events missing or flags evaluating wrong? We fix it.

PostHog promises product analytics, session replay, feature flags and experimentation in one place, which is appealing until your capture rate drops, an autocapture change makes every button a new event, or the self-hosted ClickHouse runs out of disk on a Sunday. We get the pipeline honest again and leave the dashboards sensible.

  • Since 2015Fixing technology
  • FreeEstimates
  • US teamReal engineers

posthog-capture.log

still failing

What it is

PostHog at a glance

PostHog is an open-source product analytics suite, founded by James Hawkins and Tim Glaser and first released in 2020. It covers event tracking, session replay, feature flags, A/B experiments, surveys and a lightweight data warehouse, with SDKs for the usual web, mobile and server runtimes.

There are two deployment paths: PostHog Cloud (hosted in the US or EU) and a self-hosted option built on ClickHouse, Kafka, Postgres and Redis. The self-hosted edition is powerful and genuinely free, and also non-trivial to operate. The hosted product is billed on events, replays and flag requests, which can surprise teams whose event volume quietly doubled.

Type
Product analytics, session replay, feature flags and experiments
First released
2020
License
MIT for the OSS core; proprietary for some paid features
Maintained by
PostHog Inc.
Typically used for
Funnels, retention, replay, flagged rollouts, conversion experiments

The honest version

PostHog pros and cons

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

Where PostHog is strong

  • Four products in one SDKAnalytics, replay, feature flags and experiments all run off the same install, so you avoid the usual triangle of Segment, LaunchDarkly and FullStory tying together.
  • Honest open-source coreThe repo is actually the product, not a hobbled teaser. A self-hosted deployment gets the same analytics engine as the cloud version.
  • Autocapture lowers the activation costClicks, pageviews and form submits land in PostHog without an engineer wiring every event, so product people can explore the funnel on day one.
  • Feature flags evaluate server-sideFlags can be fetched locally in Node, Python and Go SDKs, so you avoid the round-trip latency and the typical client-flash problem of SPA experiments.
  • Session replay with maskingThe recorder masks inputs by default and lets you redact selectors, so replay is usable without immediately failing a privacy review.
  • SQL insights for power usersHogQL lets analysts write real SQL against event and person tables, which closes the gap between a product analytics tool and a warehouse query.

Where PostHog bites

  • Autocapture becomes noise fastWithout a disciplined taxonomy, every dashboard is dominated by $autocapture events pointing at the wrong CSS selectors. Cleaning it up means defining explicit events and then weaning the team off the autocaptured ones.
  • Self-hosted is a real operations projectClickHouse, Kafka, Zookeeper or KRaft, Postgres, Redis and the plugin server all need monitoring, upgrades and tuning. A single-node Docker demo is not what production looks like.
  • Billing scales with event volumeThe free tier is generous and the next tier is reasonable, but a chatty client-side instrumentation can push a growing app into the five-figure range per month without the team noticing.
  • Session replay is heavyA replay payload can balloon on complex apps and interact with your own performance work. Masking, selective capture and sampling are not optional once you have real traffic.
  • Feature flag flicker without server evaluationSPAs that fetch flags over HTTP will render one variant for a split second before hydration, which is good enough for internal tools and visible in production experiments.
  • Cohort calculation can be slowLarge behavioral cohorts over long windows run as ClickHouse queries, and complex definitions on big datasets can time out or burn a lot of memory before completing.

Sound familiar?

Common PostHog problems we fix

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

Request failed with status code 429

The client-side SDK is being rate limited

A deployment started capturing an event in a tight loop, or a bot network found the endpoint. The symptom is lost events and a pile of 429s. The fix is throttling at the source and allowlisting known traffic.

Capture clean and inside the limit

returned undefined, default applied

Feature flags are returning the wrong variant

A property that is not sent on the server but is on the client, a flag evaluated before the user is identified, or a release targeting rule that excluded the test cohort.

Flags deterministic across both sides

ClickHouse: Memory limit exceeded

Insights or cohorts time out

A query touching too many rows, often a cohort built with nested behavioral rules, or a dashboard set to the last 365 days when the last 30 was enough.

Queries fast and inside memory

Ingestion lag: 6m23s

Events are arriving late in the UI

Backpressure on the plugin server, a slow downstream integration, or a Kafka consumer lagging behind. The data is not lost, but dashboards disagree with reality for ten minutes.

Ingestion caught up and watched

snapshot too large, dropped

Session replays are missing or broken

A complex app producing enormous rrweb payloads, a mask selector that was removed, or a user on a slow network dropping chunks of the recording.

Replays complete and privacy-safe

distinct_id detected: anonymous merged

Users are showing up twice

Identify called after the first event, a mobile install that lost its distinct_id after an update, or a cross-domain flow that never aliased the anonymous and logged-in IDs.

One person per person

GDPR delete request pending 48h

A deletion request has not completed

A backlog in the deletion worker, a shard that cannot be reached, or a plugin that duplicates events into a destination the delete cannot reach.

Deletes applied within policy

What we do

PostHog help, Cloud or self-hosted

Reach out for a single broken insight, or let us own the whole instrumentation.

Rescue
Events not landing, flags returning the wrong thing, replays blank, or a self-hosted node refusing to boot. We stabilise the pipeline and write up what changed.
Event taxonomy
Autocapture tamed, a core event set defined per surface, properties standardised and the old chaos archived instead of lingering in the sidebar.
Feature flags and experiments
Server-side evaluation wired in, release rules reviewed, experiment exposure modelled honestly, and a shutdown plan for flags that have been live for six months.
Self-hosted operations
ClickHouse sizing, Kafka tuning, disk and memory budgets, upgrades rehearsed on a copy, and backups that have been restored at least once.
Cost review
Event volume traced to specific code paths, replay sampling tuned, retention windows set to what the product actually uses, and the bill brought back into range.
Managed Tech
Weekly health checks on ingestion, replay payload trends, flag hygiene and billing. A named engineer rather than another ticket queue.

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 PostHog alongside something else?

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

Questions

PostHog questions

Something else? Call+1 737 727 7327

Should we use PostHog Cloud or self-host it?

Below a few million monthly events, PostHog Cloud is almost always cheaper than running ClickHouse, Kafka, Postgres and the plugin server yourself. Self-hosting starts to win above that, and only if you already have people who operate those systems. We will run the numbers against your real event volume before suggesting either.

Why do my PostHog feature flags return the wrong value?

The common cause is calling the flag before identify, so PostHog evaluates against an anonymous user who does not match your targeting rule. The other common cause is a property set on the client that is missing on the server, and the two environments then disagree. Both are fixable by evaluating locally with bootstrapped properties.

How do we stop autocapture from flooding the dashboards?

Define a small, explicit set of events that matter to the product and wire them. Keep autocapture on as a safety net, but filter it out of the main insights and only use it to discover events you forgot. Over time the explicit set becomes the canonical one, and the noise fades.

Our ClickHouse node ran out of disk. What do we do first?

Free enough space to let the server start, then look at TTLs on the events tables, retention settings in PostHog and whether session replay payloads are being kept longer than they are useful. The long-term answer is larger disks and a monitored growth curve, not a weekend of manual cleanup.

Can we use PostHog alongside an existing analytics tool?

Yes, teams often run PostHog next to Google Analytics or Mixpanel for a while. The trick is deciding what is the system of record for which decision, and not instrumenting the same event to two tools with slightly different definitions. We help draw that line before the duplication becomes a problem.

How much does PostHog help cost?

Our quote is free. Tell us the plan you are on, the volume you capture and what is misbehaving, and we return a scope and a fixed fee before we look at the account.

Free estimate

PostHog giving you trouble?

Capture, replay, flags, experiments and the stack underneath. We handle both the hosted edition and the self-hosted one.

  • 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.