Appearance
Data delay and freshness
Two separate lags sit between a probe taking a measurement and you reading a number on a page. They have different causes and different symptoms, and confusing them is the most common way to misdiagnose a stale-looking chart.
Lag 1 — RIPE's ingestion delay
This is what the Data delay view (/statistics/data-delay) measures. RIPE Atlas publishes a metric describing how long results coming off probes take to land in its own storage cluster. Atlas polls it every minute and keeps every sample forever, which is what makes the chart useful during an incident: you can go back and see exactly how bad it got and for how long.
The view shows the current delay alongside the mean, median and the 5th, 75th, 90th, 95th and 99th percentiles, plus counts of how many results are old, current and new in the buffer. The percentiles are the interesting part — a median of three minutes with a p99 of six is a healthy pipeline; a p99 in the hundreds of seconds while the median stays flat means a subset of probes is struggling, not the whole system.
This delay applies to everyone, including RIPE's own interface. It is not something Atlas can be behind or ahead of.
A compact version of this widget sits at the top of your dashboard, refreshing on its own 60-second loop, so the "is the pipeline healthy right now?" answer is always one glance away.
Lag 2 — Atlas's archive lag
Most Atlas views do not read RIPE live. Background workers ingest on their own schedules and pages read what has landed:
| Data set | Cadence |
|---|---|
| Connection events (measurement 7000) | every 15 minutes |
| Probe uptimes (measurement 7001) | recurring, change-compressed |
| Probe archive snapshots | daily, one snapshot per day |
| Anchor mesh matrix and mesh quality | daily |
| PeeringDB, GeoNames, geofeeds | daily |
| Anycast prefixes | weekly |
So a connection event is minutes old, a probe-archive figure is up to a day old, and an anycast record can be a week old. None of that is a fault — it is the cost of having history at all — but it does mean "today" on a daily series is often incomplete or absent.
Ingestion is designed to tolerate gaps rather than hide them: imports are idempotent, overlap windows are generous, and missed periods can be backfilled afterwards. A gap you see today may be filled in tomorrow.
Which views are live
Lookups that hit the upstream API directly — probe lookup, measurement detail, RIS Live, and the personal dashboard — are exempt from lag 2 entirely. They are only ever subject to lag 1, plus whatever RIPE's API is doing at that moment.
Diagnosing a stale-looking view
In order:
- Check the data-delay view. If RIPE's own delay is elevated, everything downstream is late and there is nothing to fix on this side.
- Check the cadence in the table above. A daily series that stops at yesterday is behaving normally.
- Check for a gap rather than an end. A break in the middle of a series is a missed ingestion window; a series that simply stops is either cadence or an ongoing problem. See Gaps versus zeroes.
- Still wrong? Raise an issue — that is what flags are for.