Pipeline Status

이 페이지는 아직 귀하의 언어로 번역되지 않았습니다. 현재 작업 중이므로 곧 다시 확인해 주십시오.

조기 액세스

이 기능은 Early Access 버전입니다. 이 기간 동안에는 기능의 일부가 계속 개발될 수 있으며, 이 문서는 항상 최신 상태가 아닐 수 있습니다. 질문이 있으시면 다음으로 문의하십시오: Amplitude Support.

Pipeline Status tracks events from collection through processing and into Amplitude. Use it to confirm that data arrived, understand why event volume changed, and separate expected data controls from problems that need investigation.

Use Pipeline Status to investigate aggregate volume across a project. To inspect the event stream for one user, use the Amplitude Debugger. To debug individual requests, use the Ingestion Debugger.

Pipeline Status is available to all Amplitude customers. Metrics are estimates and may update late. Use them to understand processing trends.

Open Pipeline Status

  1. In Amplitude, go to Data > Pipeline Status.
  2. Select the project you want to inspect.

Choose the data to inspect

Use the controls at the top of the page to choose:

  • Source: Switch between realtime ingestion and batch imports. Availability to switch depends on your configured sources.
  • Time range: Review the last 24 hours, 7 days, or 30 days. Where a comparison appears, it compares the selected period with the previous period of the same length.

Pipeline health is available for realtime ingestion and batch imports. Request health applies only to realtime ingestion.

Read the overview

The overview summarizes three points in the event pipeline:

  • Events entering processing: Events that passed initial request and record checks and entered Amplitude processing.
  • Dropped during processing: Events removed after processing began, with the percentage of entering events that the pipeline dropped.
  • Events available in Amplitude: Processed events ready for analysis, with the percentage of entering events that became available.

The ingestion funnel includes standard tracked events only.

Counts can differ while metrics update. Events rejected before processing don't enter the first count, and Amplitude tracks attribution holds separately.

Understand realtime request health

Request health shows whether realtime ingestion traffic reaches Amplitude before event processing starts. Use it when the number of events entering processing drops unexpectedly.

  • Check whether the success rate changed when event volume fell.
  • Review the response-code breakdown to narrow down the cause of failed requests.
  • If requests remain healthy, continue to the event pipeline to check what happened during processing.

Request counts and event counts use different units because one request can contain multiple events. When Amplitude rejects a request, its events don't enter the pipeline or contribute to Events entering processing. Request health also excludes failures that happen before Amplitude can identify the API key.

Understand pipeline health

Pipeline health reports broader processing delays for the selected source. It reflects overall system health rather than the health of only your project, so a degraded status doesn't always mean that the delay affects your data.

A Healthy status doesn't mean that every request succeeded or that your project has no processing drops. Review request health and the ingestion funnel separately to understand those outcomes.

  • Healthy: Realtime ingestion or batch imports are processing as expected.
  • Degraded: Amplitude has detected a broader delay that may affect your data.
  • Major delay: Amplitude has detected widespread delays and may direct you to check the official status page.
  • Unavailable: Amplitude can't determine pipeline health because the required telemetry is incomplete or unavailable.

Follow events through processing

The pipeline shows where records enter, leave, and become available. Select a drop stage to add it to the Events Over Time chart. The chart displays hourly or daily UTC totals and can plot up to eight stages at once. It may use a separate scale for small drop series so they remain visible beside total event volume.

The exact path varies by source, record type, and project configuration. A stage can show zero because no events met its condition or because that stage doesn't apply to the selected source.

Before processing

These records don't contribute to Events entering processing:

  • Rejected before processing: Input records that Amplitude can't convert into valid events, such as records with missing fields or invalid formats.
  • Skipped before processing: Batch import rows that the import mapping chooses not to convert into events.

During processing

Pipeline Status can show these removal stages when they apply:

  • Duplicate events: Events that repeat an insert_id that Amplitude has already processed. Learn more about event deduplication.
  • Sampled events: Events removed by a configured sampling policy.
  • Schema validation rejects: Events rejected by your tracking plan or taxonomy rules. Learn how to configure your schema.
  • Blocked or invalid events: Events that Amplitude removes because of project controls or input restrictions. The stage can show a reason breakdown, including blocked or deleted event types and device ID filters.
  • Silenced devices: Events from devices Amplitude has identified as producing unreliable volume.
  • Group resolution: Events that Amplitude can't connect to the required group information.

Some removals, including sanctioned-region controls, happen after processing begins but don't appear as a separate reason yet. Because of this, Events available in Amplitude may not equal entering events minus the listed drops.

Attribution handling

Amplitude keeps attribution events separate from pipeline drops because it may return them later:

  • Temporarily held: Attribution events waiting for a matching device ID.
  • Returned later: Previously held attribution events added back to the available event stream.

The hold and return can happen in different time periods, so these counts don't behave like a normal drop stage.

Understand what happens during processing

Pipeline Status summarizes counts, but Amplitude performs additional work on accepted events before they become available for analysis. Not every step applies to every event.

Prepare data

Amplitude converts incoming records into a consistent event format, normalizes fields, and adds context that later analysis depends on. Depending on the source and project configuration, this work can include:

  • Adding geographic properties when Amplitude has location information. Learn how Amplitude derives location data.
  • Applying configured transformations and custom tracking rules.
  • Expanding source-specific records into the events and properties used in Amplitude.
  • Preparing experiment exposure and commerce data for analysis.

Protect and validate data

Amplitude applies project controls before it stores an event. These controls can remove restricted data, enforce your tracking plan, block unwanted event types, deduplicate repeated events, and apply sampling rules.

These steps explain why a successful request doesn't guarantee that every submitted event becomes available. Pipeline Status separates configured controls from malformed input where the current metrics allow it.

Connect identities and attribution

Amplitude resolves identifiers, aliases, and attribution context so activity connects to the right user over time. This can include matching anonymous activity with a known user and linking an attribution event with its related conversion. Learn how Amplitude tracks unique users.

Sync properties and groups

Amplitude combines event data with the user and group context needed for analysis. Depending on your configuration, this can include syncing user properties, applying global or cross-project properties, resolving group membership, and preparing nested properties. Learn more about user and event properties and account-level reporting.

Finalize events

Amplitude finishes any remaining source-specific processing, records metadata used to discover events and properties in the product, and writes accepted events so they become available for analysis.

Share feedback

Under Was this page helpful?, click the thumbs-up or thumbs-down control to share feedback. Your response helps Amplitude prioritize the metrics and explanations that matter most during Early Access.

이 내용이 도움이 되었나요?

For AI agents: a documentation index is available at /docs/llms.txt. Append .md to any page URL for markdown, or send Accept: text/markdown.

이 페이지에서