Incident drill: a broken upstream caught by freshness and lineage

At a glance

Learning objectives

• Confirm a freshness breach from a schedule diff in minutes
• Walk upstream with get_lineage to find the source that slowed
• Report the impact set — owner, criticality, earliest SLA — as the incident note

What you need

• [object Object]
• A lineage store reached through mcp-datahub or the lineage-graph plugin

Steps

1. Confirm the trigger — a cron changed from hourly to daily, and a 6h freshness SLA will be violated. (from hook-freshness-pre-merge )
2. Show the arithmetic — the check compares the schedule in the diff with the declared SLA and warns with the computed worst-case lag. (from hook-freshness-pre-merge )
3. Name the guard that should have caught it — a hook that warns when a change would silently widen the freshness SLA of a source. (from hook-freshness-pre-merge )
4. Walk upstream — get_lineage from the stale asset finds the source that stopped feeding on time. (from mcp-datahub )
5. Walk downstream for impact — from the broken node to the published marts; report owner, criticality, and the earliest SLA that will break. (from skill-lineage-explainer )
6. Ask the opening question out loud — "what breaks if I change this column?" is how the drill starts. (from skill-lineage-explainer )
7. Run it from commands — /lineage-up to find the culprit, /impact to size the blast radius. (from plugin-lineage-graph )

Check yourself

• You can state the breach, the worst-case lag, the broken source and the first SLA that breaks — in that order, with evidence.
• The incident note names an owner and a criticality, not a dashboard link.

Worked example

1. breach:  cron hourly→daily vs 6h SLA — worst-case lag 24h
2. source:  get_lineage(stale mart) — the upstream source slowed
3. impact:  walk downstream, stop at published marts —
            owner, criticality, earliest SLA that breaks
4. page:    owner + evidence (the diff), not a screenshot

/lineage-up <stale asset>    # find what feeds it, spot the slow source
/impact <broken column>      # list downstream consumers a change breaks

Sourced from

• Freshness Pre-Merge Check
• DataHub MCP Server
• Lineage Explainer
• Lineage Graph

Security scan

Last sweep 2026-10-24: 0 finding(s), 0 marked as examples.

Common gotchas

• The diff is evidence — the schedule change and the SLA it breaks are provable before anyone is paged.
• An upstream that slowed rather than stopped still breaches a tight SLA — worst-case lag catches it.
• Stop the walk at published marts — that is where owners and SLAs live, and where the walk ends.

Production considerations

• Wire the freshness-pre-merge check into CI so this incident class dies before merge.
• Keep the drill to four questions — breach, source, impact set, owner.

Diagram

incident drill
   |
   |- 1. breach:  cron hourly→daily vs 6h SLA ──▶ worst-case lag 24h
   |
   |- 2. source:  get_lineage(stale mart) ──▶ upstream source slowed
   |
   |- 3. impact:  walk downstream ──▶ mart_revenue (published) ──▶ owner, criticality, earliest SLA
   |
    --> 4. page: owner + evidence (the diff), not a dashboard screenshot

Registry stats