Skip to content

Troubleshooting

Infinitnet edited this page Nov 17, 2026 · 1 revision

Troubleshooting

Common problems, what causes them and how to fix them. Most answers start on the Data page: its "Not recorded" table counts every request that was not stored, by reason.

How to use it

  1. Find the symptom below.
  2. Check the cause in the place named (Data page, Settings, the browser's network panel, the Worker logs).
  3. Apply the fix and test again in a private browser window.

No data appears

Cause Fix
The layout does not render EmDashHead, so no script reaches the page. Render EmDashHead in <head>. View the page source: it must hold <script defer src="/_/e.js">.
The integration is missing or comes before emdash(). Add cookielessAnalyticsIntegration() to integrations after emdash(). Open https://example.com/_/e.js: it must answer 200.
injectScript is off and the theme adds no tag. Turn injectScript on, or add the tag with the path from scriptPath.
You test while signed in, or your browser sends Global Privacy Control. Test in a private window without those (excludeEditors, signals).
No Cron Trigger: older ranges show no summaries. Add "triggers": { "crons": ["* * * * *"] } to the Wrangler config.

Check: the network panel shows POST /_/e answered 204, and Overview replaces its "No visits yet" panel with the report.

Editor visits missing

Signed-in editors and administrators are not recorded while excludeEditors is true (the default). This is intended. Test in a private window, or turn the setting off for a moment.

Blocked by an ad blocker

Content blockers look for words in paths. The plugin serves its script and event path from your own origin under neutral paths (scriptPath /_/e.js, collectPath /_/e). If you change them, avoid such words: a path that contains one falls back to its default. Client sites load the script from your install, a third party host for them; proxy /_/e.js and /_/e on the client's own zone if that matters.

Search not syncing or throttled

Cause Fix
Bing shows Throttled ("Bing limits shared server addresses. The sync retries on the next scheduled run."). Nothing to do: Bing limits calls from shared addresses such as Cloudflare's. The stored rows stay; the sync retries.
Google stops with invalid_grant after 7 days. The consent screen is in Testing mode. Publish it (In production), then connect again.
No property mapped for a client site. Set the Google property and Bing site URL on Definitions, Sites.
A run stops early on Workers Free. Keep syncSubrequestBudget at 40 on Free; on Paid raise it to about 900.
Backfills move slowly on Small. Search writes are limited to 21,250 rows per UTC day on small; a long backfill takes several days.
Errors after 429 or 5xx answers. Runs wait 5 minutes, doubling up to 2 hours. Check the status strip on the Search page.

Encryption key missing

Saving a Search secret fails without EMDASH_ENCRYPTION_KEY. Generate and store one:

npx emdash secrets generate
wrangler secret put EMDASH_ENCRYPTION_KEY

A lost or changed key makes the stored secrets unreadable: connect the Search sources again.

Database busy or overloaded

The Data page lists overload, write_error or daily_limit under "Not recorded".

Reason Cause Fix
overload D1 refused the write as overloaded, or a Postgres database did not answer in time (after a timeout the server instance does not connect again for 20 s). The page script retries for about a minute. Use the Queue-Worker, or an own database for analytics.
write_error Any other failed write, for example a binding named in database that does not exist. Check the binding name in the wrangler config and the plugin options.
daily_limit The account used the Workers Free plan's 100,000 rows written per day. Set capacityProfile to small; it stops recording before the limit.
budget_blocked The Small profile's write guard paused recording for the rest of the UTC day. Wait for 00:00 UTC, or move to Workers Paid and standard.

A refused write is answered 503. On Free, a background run that hits the request limit makes the Data page say "This install appears to run on the Workers Free plan": set the profile to Small.

Numbers differ from other tools

  • Visitors are counted per day and summed over days; there are no cross-day unique visitors.
  • Bots, browsers with Global Privacy Control, signed-in editors, excluded paths and IP ranges are not counted.
  • No cookies: a visitor whose network or browser changes mid-visit starts a new session; two people behind one address with the same browser can be joined.
  • A page view counts on the day its visit started, also after midnight.
  • A reload of the same path within reloadDedupeSeconds (default 2) is not counted again.
  • Nothing is sampled.

Days start at the wrong hour

Days and hours follow siteTimezone (default UTC). Set it to your IANA timezone, for example Europe/Berlin. The Small profile's write guard and the Search rows budget use the UTC day regardless. Visitor keys rotate at saltRotationHour in saltTimezone.

Client site visits refused

The Data page counts them under these reasons:

Reason Cause Fix
origin_rejected The page's host is not listed for the site, or the page is not on https. Add the exact host name (with and without www) on Definitions, Sites.
unknown_key The key is wrong, or was replaced with New key. Copy the current snippet from the Sites tab.
site_paused The site is paused. Resume it on the Sites tab. Changes reach every isolate within 60 seconds.

Also check that the middleware is installed: the Sites tab warns when it has not been seen for a day.

Queue not consuming

Cause Fix
The producer binding is not named CA_QUEUE. Rename it. Without it the setting has no effect and the Data page shows "Direct to the database."
The consumer's @infinitnet/cookieless-analytics version differs from the site's. Set the same version and deploy the consumer again.
The consumer lacks the CA_DB binding or the database option. Add both; it retries every message until then.
Messages failed for about 2.5 hours. They are in ca-ingest-dlq. Add it as a second consumer queue to apply them.
late_after_close counts after an outage. Messages that waited past 00:15 of the day after their own day are not written.

Older detail is gone

Retention deleted it. Detailed rows are kept for rawRetentionDays (default 760), shorter when the storage budget of the capacity profile requires it; the Data page then shows a notice. Older days answer from daily summaries, with fewer filters; journeys show only their top steps and paths there, and Search Satisfaction shows a notice. Deleted detail cannot be restored by raising the setting. See Retention.

Limits

  • The page script retries a refused page view up to 3 times over 65 to 97.5 s, only while the page stays open.
  • Drop counts are a lower bound: an instance that stops loses up to 10 seconds of counts.

Gotchas

  • Bots are excluded and never leave a stored row; a high bot count is normal.
  • Test with a private window, not a signed-in session.
  • Without Cloudflare in front (a local dev server) the country stays empty.

Related

Quick-Install, Data-Page, Queue-Worker, Capacity-Profiles-and-Cloudflare-Plans, Search-Integrations, Multi-Site-and-Sites-View, Retention, Upgrade.

Clone this wiki locally