DOCS · CORE WORKFLOWS
Monitoring & Retries
The activity trail, per-attempt run receipts, retry semantics, and failure alerts that hit your own endpoints.
Scheduled work that nobody can see is nobody’s job. Every delivery in Recurrence Cloud leaves a receipt.
The activity trail
GET /api/activity (session) returns the org-wide stream:
curl "https://api.recurrence.cloud/api/activity?status=failed&days=30" \
-H "Authorization: Bearer <session>"
Filters: status (success|failed|retrying|exhausted|skipped), scheduleId, webhookId, subject, days (1–90), page, pageSize. Each entry carries the URL, method, status code, request and response excerpts, duration, attempt number, and batch id. The dashboard’s Activity page is the same data with UI filters.
Per-schedule history: GET /api/schedules/:id/runs — one record per attempt, with trigger (scheduled / retry / manual).
Retry semantics
Configure per schedule (see retry policy):
- Attempt 1 fires normally.
- Non-2xx responses (or timeouts/connection failures) schedule a retry:
backoffSeconds × 2^(attempt-1)— 60s → 120s → 240s… for amaxRetriesof 3. - Retries reuse
batchIdand setX-Recurrence-Attemptto the new number. - Final failure → run status
exhausted; the batch showsretryingwhile attempts remain.
Because the receiver gets the same batch body again, the receiver-side contract is: idempotent on (batchId), fast 2xx when the work is already done.
Failure alerts (dogfooded)
When a batch is exhausted, Recurrence Cloud fires a failure alert to a webhook of your choice — your own ops endpoint, Sentry-style. The alert delivery is itself signed with the alert webhook’s secret, so your alert receiver can verify it the same way. Point alerts at your alerting stack:
curl -X POST https://api.recurrence.cloud/api/billing/orgs/<orgId>/accounts \
-H "Content-Type: application/json" -H "Authorization: Bearer <session>" \
-d '{ … alertWebhookUrl: "https://ops.yourcompany.dev/alerts" }'
(Sentry/Datadog-style integrations typically just point this endpoint at a generic inbound webhook.)
Skipped runs
A schedule with no complete delivery target (no webhook and no MCP tool) is recorded as skipped in the run history — visible, not silent.
Health & debugging recipes
- “Did the 3am job run?” →
GET /api/activity?scheduleId=…&days=1— read the status and response excerpt. - “Why is it always timing out?” → response excerpt +
timeoutMson the webhook; consider fanning out into a queue. - “Is my cron right?” →
POST /api/schedules/test-dueshows what would fire when, and why. - “Which attempt is this?” →
X-Recurrence-Attemptheader; correlate chains viaX-Recurrence-Batch-Id.
Next: MCP integration — giving AI agents a scheduling primitive.