Resources and limits
The question "how much CPU does this application have?" has four answers, and showing only one hides why it is what it is. The console shows all four side by side.
This page is about the size of each instance. How many instances exist is a different question, and lives in Scale and availability — on the same console screen, just above.
| Answer | What it is |
|---|---|
| Platform default | What applies when nobody declared anything |
| Declared | What you asked for |
| Applied | What the active version actually applied |
| Current usage | The most recent sample measured, with its age |
The vocabulary
| Term | Meaning |
|---|---|
| Reserved | What the service is guaranteed, always |
| Maximum | The ceiling it may reach at peaks |
| Current usage | How much it is consuming right now |
Reserving too much wastes capacity; reserving too little makes the application compete for resources exactly when it needs them most. Observed usage is what lets you calibrate without guessing.
Declaring
- Open the project and go to Scale and availability.
- Choose the service.
- Enter CPU and memory — reserved and maximum.
- Save.
Values outside what the platform accepts are refused immediately, with the valid range in the message.
Saving is not applying
With an active version, saving redeploys the service — same image, new configuration — and the screen follows the operation to its outcome.
Saying "applied" because the form accepted it would be lying about the half that matters: the change only holds once the new version is live.
When nothing is declared
The service uses the platform default, and the screen says so explicitly rather than showing a number with no origin. A service running on the default is not a misconfigured service — it is a service that has not needed tuning yet.
Reading observed usage
Usage comes from periodic samples. Two readings matter:
- The sample's age. A number from two hours ago does not describe now, and the screen shows the age instead of pretending it is live.
- Absence is not zero. With no measurement, the screen says "not measured" — not "0%". A dashboard that answers zero when it measured nothing is indistinguishable from one that answers zero because everything is quiet.
Common errors
| Symptom | Likely cause | What to do |
|---|---|---|
| The application restarts on its own under load | Maximum memory too low | Raise the maximum and watch usage |
| Slow responses at peaks | Reserved CPU too low | Raise the reservation |
| I saved and nothing changed | The redeployment has not finished | Follow the operation on the same screen |
| "Not measured" under usage | Collection has not produced a sample yet | See What exists today |
Next steps
Was this page helpful?
Report a problem on this pageDo not send passwords, keys, tokens, or customer data.