Skip to main content
Nerve Centre KPIs · Audit Profile · Sentiment Settings Databricks-specific health audit for the lakehouse database archetype. Answers six questions: (1) is workspace authentication valid and is the service principal / PAT scoped to read warehouses, clusters and jobs; (2) are the workspace, SQL warehouses and clusters reachable and active; (3) is SQL warehouse query latency (p95 / p99) and slow-query rate within band; (4) are DLT / Workflows data pipelines streaming without lag (the operational stand-in for replication); (5) are clusters and SQL sessions right-sized against capacity, with idle DBU burn bounded; (6) is Delta Lake durability covered via Time Travel retention and recent OPTIMIZE / VACUUM runs. Designed to surface the three top audits from the manifest: failed-job spikes, DBU burn anomalies, and clusters that never auto-terminate.

​
What this audit checks

​
Authentication & access

  • Workspace URL host resolves to a valid Databricks region (.cloud.databricks.com / .azuredatabricks.net / .gcp.databricks.com)
  • Personal Access Token authenticates and is not expired against /api/2.0/clusters/list
  • Token / service principal has READ permission on warehouses, clusters and jobs (no 403 on probe calls)
  • HTTP Path resolves to a real warehouse (/sql/1.0/warehouses/<id>) or clusters endpoint

​
Connection & availability

  • Workspace responds and at least one cluster or SQL warehouse is in a RUNNING state
  • Active SQL warehouses reachable via /api/2.0/sql/warehouses with no STOPPED warehouse expected to be serving
  • SQL query error rate over the last 5m below 1% (no widespread query failures)
  • Rate-limit headroom healthy - not hitting 429 on /api/2.1/jobs/runs/list against the 600/min workspace ceiling

​
Query performance

  • SQL warehouse query latency p95 below 5000ms over the rolling 5m window
  • SQL warehouse query latency p99 below 30000ms over the rolling 5m window
  • Slow-query rate below 5% from /api/2.0/sql/history/queries (duration over the slow threshold)
  • Top 10 slowest queries identified by query_id with duration_ms and rows_examined for tuning

​
Replication & lag

  • DLT / Workflows pipeline lag (time since last successful run) below the pipeline’s expected_interval
  • No data pipeline in BROKEN or STOPPED state from /api/2.0/pipelines status distribution
  • Scheduled job success rate over the last 24h at or above 95% from /api/2.1/jobs/runs/list
  • No failed-job burst - fewer than 5 runs with result_state FAILED or TIMEDOUT in the last 1h

​
Storage & capacity

  • SQL warehouse saturation below 90% (active sessions vs max_num_clusters x concurrency)
  • Avg cluster CPU utilisation between 30% and 90% - flag under-utilised (right-size down) or saturated (scale up)
  • Idle cluster DBU wasted over 24h below 10% of total DBU (auto-termination firing on idle clusters)
  • DBU burned over 24h within band - not +50% week-over-week without proportional workload growth

​
Backups & durability

  • Last Delta Lake OPTIMIZE / VACUUM on the largest tables ran within 72h
  • Delta Time Travel retention at or above the default 7 days for recovery coverage
  • Long-running jobs (over 1h past expected duration) flagged as cost-runaway / stuck-write risk
  • Cross-platform: SQL slow-query or pipeline-lag burst correlated with sibling commerce checkout drop (revenue at risk)

​
Severity thresholds

​
Data sources

  • GET https://{workspace}.cloud.databricks.com/api/2.0/clusters/list - Cluster inventory, state, DBR version, region - auth probe + capacity
  • GET https://{workspace}.cloud.databricks.com/api/2.0/sql/warehouses - SQL warehouse list, size, sessions, max concurrency - saturation
  • GET https://{workspace}.cloud.databricks.com/api/2.0/sql/history/queries - Query history with duration, read_rows, error state - latency + slow-query rate
  • GET https://{workspace}.cloud.databricks.com/api/2.1/jobs/runs/list - Job / workflow runs - success rate, failures, pipeline lag (replication)
  • GET https://{workspace}.cloud.databricks.com/api/2.0/pipelines - Delta Live Tables pipeline status distribution + lag
  • GET https://{workspace}.cloud.databricks.com/api/2.0/billable-usage/download - DBU burn by cluster / user / workflow - cost anomaly (account-level auth)