Archive Health
What the BGPKIT MRT archive is doing right now: which RouteViews and RIPE RIS collectors published a file in the last 24 hours, which went quiet, and which alert episodes are open. Every number below is read from the public v4 API when this page loads, with no build step and no stored copy.
loading…
Status signals
The API's own verdicts per data product: what each signal last observed, how old that is,
and whether its freshness rule calls it stale. overall summarises the
self signals; upstream covers the remote archives. Below that,
dataset freshness from the storage layer.
Open and recovering alerts
Alert episodes that are open or recovering now, one row per condition and collector. Age is wall-clock time since onset. An episode is a derived condition on the file index, not a BGP event.
Alert history, last 72 hours
Collectors inside an alert in each hourly bucket, split by project and stacked by severity: critical at the base, warning above it. One row per project, shared vertical scale.
Collector coverage
RIB coverage per collector, quietest first: hours since the newest RIB snapshot, days of snapshots kept in the ledger, and rows stored. Rows older than 24 hours are highlighted.
Method & limits
This page has no build step and no data pipeline of its own. Each panel fetches the public v4 API at request time; nothing is baked into the HTML. A number that looks wrong is an API or source problem, not a stale build: reload, then compare against the endpoint named under each panel.
- Alerts are derived episodes. A condition (for example a RIB that stopped advancing) is evaluated per collector from the broker's file index. An open alert means the archive stopped producing what that rule expects; it does not mean routing changed, a session dropped, or a prefix disappeared.
- "Published in 24 h" counts distinct collectors with any file whose
ts_startis within 24 hours of page load, from the broker read model. It says nothing about file size, completeness, or whether the file parses. - File age, not publish lag. The catalog records no availability time, so
age_secondsand hours-behind are measured from the file's own timestamp. A collector can be "new" and still have published late. - RIB coverage comes from the
bgp.rib_latesthot copy plus thebgp.rib_publishledger. "Days stored" and "rows" are ledger aggregates per collector; hours behind is wall-clock now minusnewest_rib_ts, not the time of the last publish attempt. - Alert retention is 30 days in the API, which is why the history chart is capped at 72 hours on this page. Each bucket counts collectors in alert, so a collector that flaps adds one per bucket, not one per transition.
- Not measured here: peer/session state, prefix counts inside a file, parse success rate, per-file latency, or upstream archive availability beyond what the health signals report. Those live in other BGPKIT Labs experiments.
- Endpoints read:
/v4/meta/health,/v4/meta/freshness,/v4/mrt/collectors/latest,/v4/meta/alerts?state=open,recovering,/v4/meta/alerts/series?hours=72,/v4/mrt/collectors/rib-coverage.
Data sources: RouteViews and RIPE RIS, indexed by the BGPKIT broker. BGPKIT is not affiliated with the collectors. Corrections: contact@bgpkit.com.