Skip to content
LakeBench
ProblemsCommunityPricing
Sign inStart practicing
Back
  1. Home
  2. Interview prep
  3. Upstream data is late

Pipelines & scenarios · Production Scenarios

Upstream data is late

Mediumpipelines-43
scenarioslalate-dataincidentcommunication

Question

The upstream team's data hasn't arrived and your SLA is in one hour. What do you do?

Solution

First confirm that the data is really late, then decide what to publish, and tell the people who depend on you before they find out themselves. The technical part is small. Most of the work is judgement and communication.

Confirm (first 5 minutes)

Check the freshness signal you have: the sensor or file check in your orchestrator, the latest partition in the source table, the upstream job's status page. Make sure it is not your own sensor failing, a wrong path, or a timezone mistake. Then message the upstream owner with facts: which table, which date, what you expect, and your deadline.

Decide with the information you have

You usually have three options, and the right one depends on the consumers.

  • Wait. Fine if the upstream owner gives you an ETA that still fits your SLA, or if the report is not time critical.
  • Run with partial data and flag it. Publish what you have, mark the tables or dashboards "incomplete, missing source X", and rerun when the data lands. Good for reports where approximate numbers beat nothing, but dangerous for finance numbers that people copy into other documents.
  • Publish yesterday's data with a visible banner ("data as of March 1"). The safest choice for executive dashboards, because it is stale but consistent.

Do not run the full pipeline with the missing source treated as empty and say nothing. That produces wrong numbers that look right, and is the worst outcome.

Communicate

Post in the channel your consumers watch: what is late, what you are doing, what they will see, and when you will update them next (even if the update is "no news"). One clear message at the start saves twenty questions.

Afterwards

  • Rerun everything affected when the data lands, and check the numbers against the partial version.
  • Talk to the upstream team about an explicit agreement: a delivery time, a way to signal delay early, and a named contact.
  • Move your own alert earlier. If you only learn about lateness at the SLA deadline, you have no time to react. An alert at 80 percent of the window gives you room.
  • Write a short note if it happens more than once, with the count and the impact, so the conversation about an upstream SLA has numbers.

In an interview, state the three options, say how you choose between them, and stress the communication step.

PreviousNext