Container metrics
CPU, memory, and network charts for a single project or across all of your projects, over a time window you choose.
Every project deployed on App Platform reports container-level metrics automatically. View them per-project, or roll them up across every project in your account, over a rolling time window.
View a single project
Open a project and go to its Monitoring tab. Pick a time window — 1 hour, 6 hours, 24 hours, or 7 days — and the CPU, memory, and network charts redraw for that period.
View all projects at once
From the top-level Monitoring section in the sidebar, switch to the all-projects view for an aggregate picture across your whole account, on the same time-window control.
Reading the charts
- CPU — percentage of the allotted vCPU in use. Sustained readings near 100% mean requests are queuing behind compute, not just briefly bursting — a signal to move up a resource preset or enable auto-scaling.
- Memory — usage against the container's limit. A curve that climbs and never falls back usually points to a leak; hitting the limit gets the container restarted, which shows up as a gap in the chart.
- Network — traffic in and out over the window. Useful for spotting traffic spikes, unexpected outbound calls, or confirming a deploy actually shifted load.
Healthy vs. concerning patterns
| Metric | Healthy pattern | Concerning pattern |
|---|---|---|
| CPU | A sawtooth that rises with traffic and drops back between bursts, with headroom below 100% even at peak. | A flat line pinned near 100% for extended stretches — requests are queuing behind compute, not just spiking briefly. |
| Memory | Rises after a deploy or under load, then falls back once requests finish or a garbage collection runs — a repeating sawtooth. | A slope that climbs steadily and never drops back to baseline, or repeated gaps in the chart from the container being restarted after hitting its limit. |
| Network | Traffic that tracks your known usage pattern — busier during expected peak hours, quiet overnight. | Sudden, sustained outbound traffic with no matching deploy or known traffic source — often the first visible sign of a bug or misuse. |
The most useful habit is comparing a chart to its own recent history rather than judging a single number in isolation — a CPU chart sitting at 60% is unremarkable on its own, but is worth a look if that project has never gone above 20% before.
Time windows
| Window | Good for |
|---|---|
| 1h / 6h | Watching a deploy or an active incident in near-real time. |
| 24h | Spotting a daily traffic pattern or an overnight anomaly. |
| 7d | Capacity planning — is the baseline trending up week over week? |