docs/platform/concepts/trigger.mdx - next - githubmirror/novu

docs/platform/concepts/trigger.mdx
80 строк12 KB
Dima Grossman
feat(api-service): support context data upsert on workflow trigger fixes NV-8301 (#11944)
16 июл 2026, 13:33
16 июл 2026, 13:3368a900e
---
title: 'Trigger'
description: "Understand how Novu processes a trigger event from initial API request to fan-out, digest, delay, and channel delivery inside a workflow run."
---

## Trigger Request

A trigger request is the initial step in handling a trigger event. It contains crucial details such as the template identifier, a list of subscribers who will receive the notification, the payload of the notification, and any overrides that need to be applied.

## Trigger Endpoint

Upon sending the request to the [event/trigger](/api-reference/events/trigger-event) endpoint, a series of essential steps are initiated:

1. **Subscriber Mapping and Validation:** The first step involves mapping and validating the subscribers for the specified event. This ensures that notifications are sent to the correct recipients.

2. **Workflow Validation:** Following subscriber validation, the workflow associated with the event is validated. This validation process considers factors such as the active status to determine if it meets the necessary criteria for processing.

3. **Attachment Upload:** Once the validation process is successfully completed, any attachments associated with the event are uploaded to the designated storage service.

4. **Event Queuing:** The trigger event, now enriched with mapped subscribers and attachment links, is appended to the **trigger event queue**. This queuing mechanism optimizes response times, ensuring efficient event processing.

## Trigger Event Processing

When an event is picked up by the **trigger queue worker**, the processing phase begins. Here's what happens:

- **Subscriber Upsert:** The worker validates the subscribers associated with the event and either creates or updates the subscriber with the information passed in the `to` object.
- **Context Upsert:** When the trigger includes context entries with inline `data` objects, the worker creates missing contexts or replaces existing context `data`. String-only context references do not modify stored data. Inbox and other subscriber-facing APIs do not perform context data updates.
- **Notification Entity Creation:** For each subscriber listed in the trigger event, a corresponding notification entity is created. This entity contains essential data related to the organization, template, subscriber, and event payload.
- **Job Creation:** Based on the notification template's defined steps, jobs are generated. These jobs are responsible for carrying out specific tasks related to the event notification. Additionally, the notification entity is updated with a "channels" field generated from these steps, indicating the communication channels through which notifications will be sent.

## Jobs

Jobs play a crucial role in the trigger event lifecycle. They are created based on the steps outlined in the workflow.

### Job Statuses

| Status        | Description                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                          |
| ------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| **PENDING**   | This status is assigned to a job before it is added to the worker queue. It indicates that the job is waiting to be processed.                                                                                                                                                                                                                                                                                                                                                                                                                       |
| **QUEUED**    | After the initial validation and just before adding a job to the worker queue, it is set to `QUEUED`. This status signifies that the job is ready for processing but is awaiting its turn in the queue.                                                                                                                                                                                                                                                                                                                                              |
| **RUNNING**   | When a job is picked up by a worker from the queue, its status is changed to `RUNNING`. This indicates that the job is currently being processed by a worker.                                                                                                                                                                                                                                                                                                                                                                                        |
| **COMPLETED** | Once a job has been successfully executed and processed, its status is changed to `COMPLETED`. This signifies that the job has been successfully completed.                                                                                                                                                                                                                                                                                                                                                                                          |
| **FAILED**    | If a job encounters an issue during processing or execution, its status is changed to `FAILED`. This indicates that the job has not been successfully completed, and there may have been errors or problems during processing.                                                                                                                                                                                                                                                                                                                       |
| **DELAYED**   | The `DELAYED` status is applied to specific types of jobs, such as `digest` or `delay` jobs, to indicate that they are delayed and not immediately processed. For `digest` jobs, it means that the digesting process is running or scheduled for a later time. For `delay` jobs, it signifies that the job is set to be executed at a specified delay time.                                                                                                                                                                                          |
| **CANCELED**  | When a job is canceled for any reason, its status is set to `CANCELED`. This indicates that the job will not be processed further and is effectively removed from the processing queue.                                                                                                                                                                                                                                                                                                                                                              |
| **MERGED**    | The `MERGED` status is assigned to events that are part of a `digest`. It indicates that an event will be merged into the digesting event. In a digesting process, there is typically a primary or initial event that serves as the digesting event, and subsequent events are merged into it. Instead of having a separate `COMPLETED` status for these merged events, they are marked as `MERGED` to indicate their specific role in the digesting process.                                                                                        |
| **SKIPPED**   | The `SKIPPED` status is used in the context of backoff versions of digesting. In this scenario, the first event's digesting is skipped, and the second event takes on the digesting role. The `SKIPPED` status is applied to the first event's digesting, indicating that it was intentionally skipped in the digesting process. Subsequent events may be merged into the second event's digesting process, as explained with the `MERGED` status. The `SKIPPED` status helps differentiate the skipped event from others in the digesting sequence. |

## Frequently asked questions

<AccordionGroup>
  <Accordion title="What is an event?">
    An event is a request (for instance, an API call to [/v1/events/trigger](/api-reference/events/trigger-event)) that starts off an action in the Novu Workflow Engine. Events can make many different types of actions, including digests, delays, and sending notifications to various channels, as well as filters and user preference checks.
  </Accordion>

  <Accordion title="How events are billed?">
    If a workflow is triggered to a [subscriber](/platform/concepts/subscribers), then it is billed as 1 event (one workflow run). If workflow is triggered to a group of 15 subscribers, then it is billed as 15 events.

    If a workflow is triggered to a [topic](/platform/concepts/topics), Novu fans out to matching [topic subscriptions](/platform/subscription/manage-topic-subscriptions). You are billed for **one workflow run per subscriber whose subscription passes** Context matching and JSON Logic conditions, not for every subscriber on the topic. A topic with 100 subscriptions where only 20 pass creates **20** billable workflow runs. See [Billing and workflow runs](/platform/subscription/manage-topic-subscriptions#billing-and-workflow-runs).
  </Accordion>

  <Accordion title="If notifications were filtered out, are they still billed?">
    It depends on **where** the filter runs:

    - **Topic subscription filters** ([Context](/platform/concepts/contexts) mismatch or [JSON Logic conditions](/platform/subscription/manage-topic-subscriptions#json-logic-conditions)): **Not billed.** Filtering happens during fan-out, before a workflow run is created.
    - **[Workflow channel preferences](/platform/concepts/preferences#workflow-channel-preferences)**, **[global channel preferences](/platform/concepts/preferences#subscriber-global-preferences)**, or **[step conditions](/platform/workflow/add-and-configure-steps/step-conditions)**: **Still billed.** The workflow run is already created; only delivery or individual steps are skipped.
  </Accordion>

  <Accordion title="How events are billed for digested events?">
    As a workflow triggered to one subscriber is billed as 1 event. if events are digested 15 times and a single digested notification is sent, then 15 events will be billed.
  </Accordion>

  <Accordion title="How are duplicate trigger requests with the same idempotency key billed?">
    Repeated `POST /v1/events/trigger` requests that include the same `Idempotency-Key` header are halted at the API layer. Once the first request completes, duplicates return the cached response and do not queue the trigger again, so they count as **one** workflow run for billing. While the first request is still processing, duplicates receive a `409 Conflict` response. See [Idempotency](/api-reference/idempotency) for header requirements and retry behavior.
  </Accordion>

  <Accordion title="Does transactionId deduplicate triggers the same way as an idempotency key?">
    No. The optional `transactionId` field is returned for tracing and can be used to [cancel a triggered event](/api-reference/events/cancel-triggered-event), but duplicate `transactionId` values are **not** deduplicated at the API layer the way `Idempotency-Key` is. Uniqueness is enforced during trigger processing, so retried or concurrent requests with the same `transactionId` do not get the same billing protection as an idempotency key. For safe retries and atomic deduplication, use an `Idempotency-Key` header on every trigger request.
  </Accordion>
</AccordionGroup>