Metrics

Open Monitoring → Metrics to understand what the server was doing before and during a problem. The backend records samples even while your browser is closed. In 0.8.4 the default sample interval is 15 seconds and the default retention is seven days; JD_METRICS_INTERVAL and JD_METRICS_RETENTION change those values. The identity line shows the interval and retention in effect, or history off when nothing is recorded.

Investigate a spikeCopy Anchor Link

1. Choose a window: Live, 1h, 6h, 24h, or 7d. Drag across a chart to select a narrower span. The selected from and to times go into the URL, so you can reopen or share that exact window with another dashboard user.
2. Read Resources for CPU (with a bar per core), memory, load, network, and disk I/O. Each carries the window on screen as its trend and, on a recorded window, a change against the previous one.
3. Open Notable moments. Peaks, stalls, failed deployments, and reboots that arrived together are grouped into one incident, led by its heaviest signal. Press an incident to pin its instant on every chart; press it again to release it. The zoom control narrows the charts to that incident and keeps the pin.
4. While a moment is pinned, step through it with Previous sample and Next sample, or open Logs around this moment to compare it with Logs. Release moment or Escape unpins it. You can also click any chart to pin an instant.
5. Use Top processes (by CPU, memory, or I/O) and the Processor, Memory, Network, Storage, and Saturation sections to locate the source. Select Export CSV to download the window on screen when you need the numbers outside the dashboard.

A missing sensor or unsupported pressure reading means the host could not report it; it is not a zero. Memory is read as available memory, which accounts for reclaimable Linux page cache. The page no longer gives a health verdict or findings; it shows the measurements and leaves the judgement to you. For a container's own history, open it in Docker.

Read saturation as well as useCopy Anchor Link

High CPU use alone does not tell you whether work is waiting. Check CPU modes, especially I/O wait and steal on a VPS; Saturation for pressure and load; disk operations, latency, and busy time; Inodes; and Sockets with TCP, TIME_WAIT, and open file handles. Network also charts resent segments, connection RTT, and failed connections. A disk can be saturated by small random writes while transferring few bytes. Storage is shown per filesystem so a change in which mount is fullest cannot look like one disk suddenly freed space. Interfaces shows the host's real devices by default; choose Everything to include Docker's virtual pairs and bridges.

Use Live for the current state and Pause live feed when you need to inspect a moment. Recorded history remains available until retention removes it. Setting JD_METRICS_RETENTION=0 disables recording and leaves only live readings; missing historical data cannot be reconstructed later.