Runtime health

The runtime page is one row per service, environment, and deployment. A deployment is a Kubernetes deployment name, a container name, or a host name, whichever the resource attributes include.

Each row shows the release, the last check-in, restarts, disk, CPU, memory, and the log error count for the last hour. Open a row to filter issues to that resource.

Resource attributes

The Collector stamps these on every log, metric, and check-in:

• service.name, service.version, deployment.environment.name
• k8s.cluster.name, k8s.namespace.name, k8s.deployment.name, k8s.pod.name, k8s.container.name, k8s.node.name
• container.name, container.image.name, host.name

Logs

POST /otlp/v1/logs accepts OTLP JSON and protobuf, gzip, and a project API key with logs:write. Lines keep trace_id and span_id. An error line opens an issue when that trace does not already have an SDK exception.

search_logs reads lines by trace, service, severity, and time.

Gauges

POST /otlp/v1/metrics accepts container.cpu.usage, container.memory.usage, container.filesystem.usage, and the k8s.* gauges the Collector emits. The key scope is metrics:write.

A gauge alert uses kind gauge and a query such as metric:container.filesystem.usage AND tag.container.name:web. A log-rate alert uses kind logRate and optional service:web. The threshold is the error count in the window.

k8s.job.failed_pods above zero opens an issue. A rising k8s.container.restarts opens one issue. Filesystem usage at or above 90% of the limit opens a disk issue and resolves below 85%.

See pricing for how logs and metrics are billed.