Compute Pressure API

In real-time applications, such as video conferencing web apps, the Compute Pressure API lets you detect which pressure the system is currently facing. The system will handle any stress as well as it can, but a collaboration between system and app is useful to handle the pressure best. This API notifies you of high-level pressure state changes, so you can adjust your workloads and still offer a pleasant user experience. The signal is proactively delivered when the system pressure trend is either rising or easing to allow timely adaptation.

You can use these pressure change signals, for example, to reduce or increase the video quality or the number of video feeds shown simultaneously to avoid dropping video frames, audio cuts, or delaying other critical parts of the application. The quality of service of your web app can vary, also due to pressure from external factors and apps at unexpected times, but ideally that does not lead to a total system failure, input delay, or unresponsiveness. Instead, the set of enabled features and their quality level is balanced against the resource pressure of the end-user device. It is similar to network pressure in which case a streaming app adapts to the available bandwidth.

More use cases are:

• Web games, for which you could balance the quality and amount of 3D assets, change the framerate, resolution, depth of field etc., to ensure low latency and stable frame rate.
• User interfaces, for which you could render placeholders instead of real data while the system is under pressure, and render the real content once the pressure has eased.

Fast and delightful web applications should balance workloads when the system's computing resources are used at (near) full capacity. The Compute Pressure API's goal is to prevent, rather than mitigate, bad user experience in the web app itself and also for the user's device to not become too hot, too loud, or to drain the battery at an unacceptable rate. Therefore, it is advised to prefer this API over feedback mechanisms or singular performance adjustments (for example, by lowering the frequency of window.requestAnimationFrame) where bad user experience might be mitigated, but not proactively avoided. For measuring and segmenting the performance of user sessions after the fact, the PerformanceLongTaskTiming API is better suited to analyze tasks that occupy the UI thread for 50 milliseconds or more (see also Performance API for additional performance measurement APIs).

In your web app or website, different tasks are fighting for the processing time of different processing units (CPU, GPU, and other specialized processing units). The current version of the Compute Pressure API specification defines two main source types that can be queried to gather pressure information:

• "thermals" represents the global thermal state of the entire system.
• "cpu" represents the average pressure of the central processing unit (CPU) across all its cores. This state can be affected by other apps and sites than the observing site.

The list of supported sources varies per browser, operating system, and hardware, and is evolving. Use the static PressureObserver.knownSources hint to see which source types are available to your browser. Note that availability can also vary by your operating system and your hardware. Call observe() and check for a NotSupportedError to see if pressure observation is possible.

The Compute Pressure API is available in the following contexts:

• Window (main thread)
• Worker
• SharedWorker