Health | IMI Life Science

Health

Status check.

Application Request Performance

Purpose: Identify slow or failing application routes without health checks and background polling dominating the measurements.

  • Choose a gateway: All gateways combines current reporting sources. Admin and Storefront isolate their traffic. Missing or stale sources are shown explicitly; expired sources do not contribute to current combined totals.
  • Read the window: Metrics cover a rolling minute, with a shorter warm-up after a gateway starts. Background requests are excluded before calculation and counted separately. Trend range changes the chart history, not the summary window.
  • Compare latency: Average is total gateway elapsed time divided by request count. P50, P95, and P99 are histogram upper-bound estimates. Low-volume windows are labelled; no traffic is not evidence of healthy performance. These are HTTP request timings, not browser page-load measurements.
  • Separate failures: Server failures show HTTP 5xx. Client errors show HTTP 4xx, including authentication, validation, missing routes, and rate limits. Both use all application requests in the window as their denominator. Neither is uptime availability.
  • Find costly routes: Endpoint impact defaults to total time. Search a route or sort by average latency, request count, or error count. Search does not change the summary metrics. Numeric and GUID identifiers are grouped, and query values are omitted.
  • Review slow requests: The slow count includes every request above 500 ms. Recent slow samples are bounded examples, not the full count. Open Logs for investigation.
  • Inspect a stable view: Pause freezes the displayed snapshot only. Resume shows the latest received data. Refresh retries the metrics request; failed refreshes retain the previous sample and show an error.
  • Check freshness: Live indicates a fresh measurement and connected subscription. Polling fallback continues when the subscription is unavailable. Stale measurements are not current traffic. Legacy collector warnings require updated gateways and ControlPlane before application-only metrics can be trusted.

Host Resources

Purpose: Check CPU, RAM and root filesystem usage for the host serving the monitoring request.

  • Read utilization: CPU covers all host CPUs. RAM is total minus available memory, including reclaimable cache as available. RAM and disk show used/total capacity in GiB; disk covers the host root filesystem, not every mounted volume or container limits.
  • Review pressure: Usage bars turn amber at 75% and red at 90%. These are utilization thresholds, not a service-health verdict.
  • Refresh: Readings refresh every ten seconds and with Refresh Status. Samples older than thirty seconds or retained after a refresh failure are labelled stale. Unavailable measurements are never shown as zero.
  • Missing readings: Ask an administrator to check the Control Plane host mounts. Linux Docker deployments require read-only host procfs and root filesystem mounts. Other operating systems report unavailable measurements.
  • Identify scope: The host label identifies the local Control Plane node. This overview is not an aggregate across cluster nodes.

Container Resource Details

Purpose: Identify which Docker containers consume CPU and memory without leaving System Health.

  • Open details: Expand Container details below the host meters. The list includes infrastructure, application and stopped containers from the configured Docker daemon. Collapse it to stop statistics polling.
  • Compare usage: Sort by memory or CPU, or search a container name. CPU follows Docker accounting: 100% represents one logical CPU, so a container using multiple CPUs can exceed 100%. The host CPU meter instead uses all host CPUs as 100%.
  • Read memory: Memory is Docker usage minus reclaimable inactive-file cache. Limit is the effective limit reported by Docker; without an explicit container limit it can be host capacity. Values use MiB or GiB. Container memory totals are not expected to equal host memory usage.
  • Refresh: The open table refreshes every fifteen seconds. Use its refresh icon for an immediate retry. Failed refreshes retain the previous readings with a stale warning.
  • Missing statistics: Stopped containers and failed or timed-out readings show Unavailable, never a fabricated zero. A failed Docker inventory request is distinct from an empty daemon. Collection is read-only and does not start or restart containers.