At a glance
A real-time donut showing how your Delta Live Tables (DLT) pipelines are split across Running, Idle, Failed, and Stopped states. It is the one-glance answer to “are my declarative pipelines healthy right now?” Any slice in the Failed segment means a pipeline is not producing fresh data and a downstream table is going stale.
What it tracks
The card reads the pipeline inventory from the Databricks DLT API (GET /pipelines/list, the Lakeflow Declarative Pipelines list endpoint) and buckets every pipeline by its current state: Running (an update is actively processing), Idle (healthy but between scheduled or continuous updates), Failed (the last update ended in error), and Stopped (paused or never started). The donut shows the proportion in each state in real time, refreshed every polling cycle, so a pipeline flipping into Failed appears within one cycle.
A healthy distribution is mostly Idle and Running with an empty Failed segment. The states that demand attention are Failed (a broken pipeline, stale downstream tables) and unexpectedly Stopped (a pipeline that should be scheduled but is not). Because DLT pipelines are distinct from classic Jobs, this card is the streaming / declarative counterpart to Job Success Rate (24h); read both for full pipeline coverage. When a slice turns Failed, pair with Pipeline Lag (since last success) to size how stale the data has become and Failed Jobs (24h) to check whether the breakage spans both pipeline types.
This card has no hard alert threshold; it is a live status overview rather than a trigger. For the alerting view on data freshness, use the lag and failed-job cards.
Reconciling against the source
Open Workflows → Delta Live Tables (Pipelines) in the Databricks workspace: the pipeline list shows each pipeline’s latest update state, which should match the donut’s segments. For an exact count, querysystem.lakeflow.pipelines in a SQL warehouse. Brief differences are normal because the API reflects state within one polling cycle while the UI updates on page refresh.