Data Retention

Path: /operate/ttl. Verb: ttl:read (granted by maintainer, operator, admin).

The Data Retention page shows how long the connected OAP keeps records and metrics. It is read-only; change retention in OAP configuration, then refresh Horizon.

What You See

The page has two sections:

Section What it means
Records Event-style data such as Normal records, traces, Zipkin traces, logs, and browser error logs.
Metrics Aggregated metric tiers: minute, hour, day, and metadata when the OAP backend exposes it.

Values are shown in days.

Stages

Retention is reported in two stages:

  • HOT + WARM — the default view, the collapsed value OAP returns for each data class.
  • COLD — a separate cold-storage stage, present only on a BanyanDB backend. When the backend exposes it, the page renders a Data lifecycle bar; otherwise the page shows a single uniform stage and no cold pane.

On a BanyanDB backend, a cold-stage toggle appears in the topbar. Enabling it switches reads to cold-only, so recent windows go empty — pick an older time range when the toggle is on. no cold stage next to a value means the connected OAP has no cold storage stage for that data class.

The toggle applies to traces, logs and metrics, which are the classes a deployment is advised to age into cold storage. Alarms, events, the instance and endpoint pickers and everything from profiling are always read hot: they are small, and they are the first things you reach for during an incident, so emptying them would buy nothing. Those pages keep answering while the toggle is on.

When the toggle is on and the time range still overlaps hot+warm, a warning says so and names a window that clears the boundary. The window it suggests is the deepest of the classes above, not the shallowest — clearing only the shallowest brings traces back and leaves every metric widget exactly as empty. Where the connected OAP has no cold stage at all, no window can work, and the warning says that instead of suggesting one.

Requirements

  • OAP query port reachable from Horizon.
  • The logged-in user has ttl:read.
  • OAP supports getRecordsTTL and getMetricsTTL.

If the page reports OAP unreachable, check oap.queryUrl and the network path to the query port.

During Operations

Use this page before changing dashboard time windows, alert retention, or storage sizing. If a user expects old traces or metrics and Horizon cannot find them, compare the requested time range with this page first.