Skip to content

ACQUISITION_REDEPLOY.md

Latest commit

 

History

History
83 lines (61 loc) · 4.08 KB

ACQUISITION_REDEPLOY.md

File metadata and controls

83 lines (61 loc) · 4.08 KB
Raw

Restore acquisition tracking — redeploy checklist (Nan)

Root cause (verified 2026-10-19)

Production is deploy drift, not a code regression on main and not a Durable Object rename.

Evidence:

Check Result
main worker/analytics.mjs report() Still returns acquisitionEnabledAt + rowsV2
DO name in code Still site-v1 (getByName('site-v1'))
Snapshot 2026-10-12 Had acquisitionEnabledAt + rowsV2 (20 rows); startedAt = 2026-09-26T10:06:12.427Z
Snapshot 2026-10-15+ Same startedAt, but omits both acquisition fields → Worker script rolled back
Live https://git4data.ai/assets/js/analytics.js Byte-identical to pre-acquisition commit bc6007d (PR #4), no UTM capture
Live /privacy Missing acquisition disclosure from PR #5/#6

Same Durable Object is still live (startedAt unchanged). Redeploying current main reattaches the v2 Worker class to site-v1 and does not wipe totals / totals_v2.

wrangler.jsonc: Worker git4data-ai, DO binding SITE_ANALYTICS → class SiteAnalytics, migration tag site-analytics-v1. Deploy path is Workers static assets (npx wrangler deploy), not a separate Cloudflare Pages project.

What this PR also fixes (on top of redeploy)

  1. Client allowlists in assets/js/analytics.js were never updated by PR #6 — synced with worker/acquisition-tracking.mjs (mo-blog, Slack/Discord community sources, profile medium, etc.).
  2. Campaign slugs used in promos (agent-db-write, four-verbs, playground, evergreen, …) already match the campaign pattern; covered by tests.
  3. Server no longer rewrites client referrer buckets via https://{bucket}.com (which mapped direct/search → other).
  4. Cache buster analytics.js?v=1 → ?v=2 so browsers pick up the restored client after deploy.

Deploy steps (requires Cloudflare auth — this agent cannot run them)

From a clean checkout of the merged PR (or main after merge):

cd /path/to/git4data.ai
git pull
npm ci
npx wrangler whoami          # must show the account that owns git4data.ai
npx wrangler deploy          # deploys Worker + static assets together

Secrets already on the Worker stay put. Do not run:

  • wrangler delete / recreate the Worker
  • DO migration renames / new class names
  • anything that creates a new DO name other than site-v1
  • SQL deletes against totals / totals_v2

If wrangler deploy asks about migrations: keep existing tag site-analytics-v1; do not add a migration that drops or recreates SiteAnalytics.

Verify after deploy

# 1) Client is the acquisition build (not the 2.9KB pre-UTM file)
curl -sS 'https://git4data.ai/assets/js/analytics.js?v=2' | head -c 200
# Expect: mentions extractAcquisition / ALLOWED_SOURCES / utm_source

# 2) Report includes acquisition fields again
curl -sS 'https://git4data.ai/api/analytics/report?days=7' \
  -H "Authorization: Bearer $ANALYTICS_READ_TOKEN" | jq '{
    startedAt,
    acquisitionEnabledAt,
    rowsV2_len: (.rowsV2|length),
    has_rowsV2: (.rowsV2!=null),
    keys: keys
  }'

Success criteria:

  • JSON includes acquisitionEnabledAt (should remain 2026-10-01T12:44:38.833Z if the same DO metadata row survived; if missing in metadata, Worker inserts on next cold start — empty rowsV2 is OK).
  • JSON includes rowsV2 (array; may be empty for recent days until traffic arrives with the new client).
  • startedAt still 2026-09-26T10:06:12.427Z → confirms DO was not wiped.

Optional smoke: open
https://git4data.ai/?utm_source=x&utm_medium=social&utm_campaign=agent-db-write
then re-fetch the report and look for a rowsV2 row with those dimensions (respect DNT/GPC).

If fields are still missing after deploy

  1. Confirm deploy targeted Worker name git4data-ai in the correct Cloudflare account.
  2. Confirm custom domains git4data.ai / www.git4data.ai still bind to that Worker (see wrangler.jsonc routes).
  3. Confirm you did not accidentally point DNS at a stale Pages project.
  4. Re-check live analytics.js?v=2 content; CDN/browser cache should miss with the new query string.