Zoom webinar automation — findings & next steps

remove  webhook          → wait_5m
add     webhook          → zoom_validation
remove  webhook_response → clean
add     webhook_response → wait_5m
remove  skip mint? [false] → zoom_validation
add     skip mint? [false] → clean
add     skip mint? [true]  → execution_data_pulse_skip   (new node)

const rows = $input.all().map(i => i.json).filter(r => r && r.status);
const hit = rows.length > 0;
return [{ json: {
  ...$('webhook').first().json,   // body.payload.object.* for clean
  skip_mint: hit,
  map_status: hit ? rows[0].status : null,
  webinar_id: $('map_check').first().json.webinar_id,
} }];

clean keeps reading $json.body.payload.object.* unchanged.

Replace the node in place (same name, same position) with IF typeVersion 2.2,
matching zoom_validation:

{ "conditions": {
    "options": { "caseSensitive": true, "leftValue": "",
                 "typeValidation": "strict", "version": 2 },
    "conditions": [{
      "leftValue": "={{ $json.skip_mint }}", "rightValue": "",
      "operator": { "type": "boolean", "operation": "true", "singleValue": true }
    }],
    "combinator": "and" },
  "options": {} }

checked with n8n's validator: proposed = valid;
today = invalid ("conditions.conditions is missing"), so the node treats the check as passed:
{ "conditions": { "boolean": [{ "value1": "={{ $json.skip_mint }}", "operation": "isTrue" }] } }

a. validate_node_config on skip mint? and map_skip, then validate_workflow
b. test run with pinned data:
     endpoint.url_validation → replies instantly with plainToken + encryptedToken
     webinar.created (Pulse-made, id in pulse_event_map) → ends at pulse skip
     webinar.created (web_portal, e.g. Deal Detective)       → reaches clean with real fields
c. publish → Zoom Marketplace app → Event Subscriptions → re-validate the endpoint
d. backfill with 322c: Deal Detective (10/1) and Co-Living (10/9–13)
rollback: restore the version before the fix

Sep 29 results:
  a ✓ valid
  b ✓ all three cases routed correctly with every write pinned; no WEBINARS rows written
  c ✓ published; live validation ping answered in 0.33 s; a real webinar
      (Business Buying with Pace Morby, 12:49 PM) was added end to end.
      Zoom is delivering events, so re-validating is optional.
  d ✕ skipped: no backfill of webinars created before Sep 29

- What was the Sep 24 4:37 PM save? Is it tied to 922h NR Registry API?
- Was disconnecting append_registrant intentional (is Zapier or another flow
  feeding the SMS Verified tab)?
- Should the switch stay per-webinar row flags, with REG_CUTOVER left unused?

settings.errorWorkflow = shared-error-handler

# only if Matt confirms the SMS roster should come from the hub:
add     outbound_comms? [true] → append_registrant      (alongside confirm_sms?)
remove  append_registrant      → append_all_registrants (already fed by
                                  log_webinar_registration; avoids a double row)
keep    append_registrant      → zoom_count → count_ok? → update_reg_count

test_workflow with pinned "call" input (all row flags empty = shadow):
  a. matched tag (e.g. "10/3/26 Gator Live Registrant") → claim → phone verify
     → stamp_completed; no Close/AC/Zoom calls
  b. same person again → duplicate_stop
  c. unknown tag → log_unmatched_registration
  (the existing hub end-to-end test script covers these paths)

publish_workflow 323a
rollback: unpublish_workflow 323a

-- which AC tags arrived with no webinar row (marketing database)
SELECT tag, brand, count(*) AS registrants, min(created_at), max(created_at)
FROM unmatched_registrations
WHERE created_at > now() - interval '21 days'
GROUP BY tag, brand
ORDER BY registrants DESC;

Updated Sep 29: webinars created before Sep 29 (the Sep 26 Gator webinar and the 2 that 322a dropped) won't be backfilled. From Sep 29 on, 322a adds new webinars itself, but it leaves ac_tag empty. That field has to exactly match the AC tag before the hub can match registrants.

On that webinar's WEBINARS row, one flag at a time, checking runs after each:
  1. reg_platform = true   (Zoom add + join link)
  2. crm_writes   = true   (Close, round robin, AC)
  3. brand_log    = true   (brand log + All Registrants sheet)
  4. confirm_sms  = true   (confirmation text)
then pause that webinar's Zapier zap.
Don't enable brand_log without crm_writes; log rows would have a blank lead/rep.
rollback: set the flags back to empty

# add to local.n8n_env (workers inherit it through worker_env)
    # n8n's own DB pool. The default of 2 per process starved under AC lead
    # surges on 9/25 ("timeout exceeded when trying to connect").
    # 1 main + up to 10 workers x 10 = 110 connections max.
    { name = "DB_POSTGRESDB_POOL_SIZE", value = "10" },
    { name = "DB_POSTGRESDB_CONNECTION_TIMEOUT", value = "60000" },

# pre-check (n8n database): must comfortably exceed 110 + other clients
SHOW max_connections;
# CloudWatch: DatabaseConnections, CPUUtilization, CPUCreditBalance
#             around 2026-09-25 11:56 AM Phoenix

# apply ONLY the n8n tasks/services, never the whole directory
terraform plan  -target=aws_ecs_task_definition.main -target=aws_ecs_task_definition.worker \
                -target=aws_ecs_service.main -target=aws_ecs_service.worker
terraform apply -target=... (same four targets)

# rollback: remove the two lines and re-apply the same four targets

node every_2s → rule.interval: [{ field: "seconds", secondsInterval: 30 }]
(keep the node name; only the interval changes)
rollback: set secondsInterval back to 2

disable: ledger (subto), ledger (zdb), ledger (gator), ledger (tc), ledger (sus)
(no renames; execution_data → universal_router path is untouched)
rollback: re-enable the five nodes

CREATE TABLE IF NOT EXISTS registrant_queue (LIKE write_queue INCLUDING ALL);
-- same columns, unique event_id, defaults. Note: shares write_queue's id sequence.

-- node: enqueue_registrant  [postgres · executeQuery · marketing database]
-- wired: Registrant tag? (true) → enqueue_registrant
-- keep the old Execute node, disabled, for rollback
WITH t AS (SELECT $1::jsonb AS j)
INSERT INTO registrant_queue (event_id, platform, account, action, priority,
  coalesce_key, request_cost, payload, status, attempts, last_error)
SELECT
  'reg|' || (j->>'brand') || '|' || (j->>'contact_id') || '|' || (j->>'tag'),
  'n8n', j->>'brand', 'webinar_register', 'realtime',
  'reg|' || (j->>'brand'), 1, j, 'queued', 0, ''
FROM t
WHERE coalesce(j->>'contact_id', '') <> ''
ON CONFLICT (event_id) DO NOTHING
RETURNING event_id;

-- queryReplacement (same 14 fields 323a gets today):
={{ JSON.stringify({ ac_id: $json.contact_id, contact_id: $json.contact_id,
  email: $json.email, phone: $json.phone, fname: $json.first_name,
  lname: $json.last_name, tag: $json.tag, brand: $json.brand,
  utm_source: $json.utm_source, utm_medium: $json.utm_medium,
  utm_campaign: $json.utm_campaign, utm_content: $json.utm_content,
  utm_term: $json.utm_term, impact_click_id: $json.impact_click_id,
  impact_partner_id: $json.impact_partner_id }) }}
-- settings: retryOnFail true, maxTries 3, waitBetweenTries 2000

The event id is built from brand, contact and tag, so if AC resends the same registration it's ignored (today's random id can't catch that).

every_30s [schedule]
  → claim [postgres]
  → Loop Over Items (batch 1)
      → register [executeWorkflow 323a, waitForSubWorkflow: true,
                  onError: continueErrorOutput]
          ok  → mark_done  → execution_data (name, phone, email)
          err → mark_retry → execution_data (name, phone, email)
settings: executionTimeout 600, errorWorkflow shared-error-handler,
          saveDataSuccessExecution none

-- claim: 20 rows, skips if a batch is still in flight
WITH inflight AS (
  SELECT count(*) AS n FROM registrant_queue
  WHERE status = 'working' AND updated_at > now() - interval '2 minutes'
), cand AS (
  SELECT id FROM registrant_queue
  WHERE attempts < 5
    AND (status = 'queued'
         OR (status = 'working' AND updated_at < now() - interval '10 minutes'))
    AND (not_before IS NULL OR not_before <= now())
    AND (SELECT n FROM inflight) = 0
  ORDER BY id
  LIMIT 20
  FOR UPDATE SKIP LOCKED
)
UPDATE registrant_queue q SET status = 'working', updated_at = now()
FROM cand WHERE q.id = cand.id
RETURNING q.id, q.event_id, q.payload, q.attempts;

-- mark_done
UPDATE registrant_queue
SET status = 'done', finished_at = now(), updated_at = now()
WHERE id = $1;

-- mark_retry: exponential backoff, gives up after 5 attempts
UPDATE registrant_queue
SET attempts   = attempts + 1,
    status     = CASE WHEN attempts + 1 >= 5 THEN 'failed' ELSE 'queued' END,
    last_error = left($2, 500),
    not_before = now() + (power(2, least(attempts + 1, 5)) || ' seconds')::interval,
    updated_at = now()
WHERE id = $1;