Observability Examples¶

A salmon point and no anomaly¶

What you see. The orders card has a 1 Day Absolute Change check with a Min of 1. Tuesday's point is salmon: the count stayed at 25,000, below the Min line. The heatmap square for Tuesday is cyan, and the Anomalies list is empty.

What it means. The hourly measurement recorded the count, and the chart compares it with the check's limits. Salmon means the count is outside them but no anomaly exists, because no Scan evaluated the check. The measurement alone never raises one.

What to do. Schedule a Scan with the Metadata category on the container, after the load. From then on, the same situation turns the point orange and creates an anomaly. See When the Check Is Evaluated.

A grey square in a cyan row¶

What you see. The heatmap is cyan every day except Wednesday, which is grey, and the chart has no point on Wednesday.

What it means. No row count was recorded for Wednesday: the hourly measurement could not read the container that day, and no Scan or Profile counted it. A Volumetric change check needs a count on every day of its Measurement Period, so it is Not Asserted on Scans whose period includes Wednesday.

What to do. Look at why the measurement failed (credentials, network, a dropped table). The history fills again from the next successful count. See The History the Check Needs.

A container that became Inaccessible¶

What you see. The card's status shows Inaccessible, and the heatmap has been grey for several days.

What it means. Three consecutive measurements failed, so the container was marked Inaccessible and the hourly cycle stopped measuring it. It does not retry on its own.

What to do. Fix the access problem, then run a Sync on the datastore. A successful Sync that finds the container marks it Available, and the next hourly cycle measures it again. A Sync does not discover a Computed Table, Computed File, or Computed Join, so for those, open the container, click Edit, then Validate and Save its definition.

A bar past the Maximum Age line¶

What you see. The customers_snapshot card shows its Freshness check with a Maximum Age of 6 hours. Thursday's bar rises above the yellow line and is salmon. Hovering it shows Staleness 9h 12m in red, and Last modified on Wednesday evening.

What it means. The daily measurement found the table had not changed for over nine hours, past the Maximum Age. The bar is salmon rather than orange because no Scan evaluated the check that day, so no anomaly exists.

What to do. Schedule a Metadata Scan as often as the delay you can tolerate. With an hourly Scan, the same staleness raises an anomaly within the hour, and the bar turns orange. See Best Practices.

A day with no bar¶

What you see. The freshness chart has a gap on Monday, and the Monday tooltip shows Staleness — and Last modified —, although Freshness Tracking is on.

What it means. Each bar uses the most recent measurement of its day. On Monday the last one came from a Profile, which records a row count but no last modification time, so the day has nothing to show for freshness. The earlier hourly reading is still in the history.

What to do. Nothing is broken. The gap disappears on days whose last measurement carries a last modification time, such as a day whose last reading came from the hourly cycle or from a Scan that evaluated the Freshness check.

No measures on a new check¶

What you see. A Metric check with Percentage Change was created two days ago and has run on two Scans, but its card shows No measures.

What it means. The chart needs three measurements for a change comparison, and two for Absolute Value. The measurements exist, and the check started comparing on its second Scan.

What to do. Wait for the next Scan. The chart draws from then on, starting at the second measurement because the first has no previous value to compare with.

The points stop appearing¶

What you see. A Metric check on the daily_revenue computed table recorded a point every morning, then nothing for three days.

What it means. A change to the computed table's query made it return one row per region instead of a single row. The check needs exactly one row, so every Scan since then marked it Not Asserted and recorded no measurement.

What to do. Restore the single-row query, or narrow the check with a filter clause such as region = 'EMEA'. See Exactly One Row.