Performance for Everyone

In mobile apps, performance is not an optional feature but a basic feature. Pinterest has measured, protected, and improved the performance of core surfaces like Home Feed and Search Result Feed, using as its central metric user perceived latency (= Visually Complete) — the time from when a user takes an action to when they actually see content.

The problem is that the criteria for Visually Complete differ by surface. On Video Pin Closeup, the criterion is the start of full-screen video playback; on Home Feed, it's the completion of image rendering and video playback; on the Search Auto Complete Page, it's the completion of rendering recommended text and avatar images. As a result, custom measurement logic had to be built separately for each surface, and attaching a single metric to a product on Android took an average of 2 engineer-weeks.

The solution was to embed the performance measurement logic into BaseSurface, the common base UI class, rather than into individual features. Starting from RootView, it traverses the view tree, examining only visible views, and automatically calculates Perceived Latency by determining whether an image has been drawn, whether text has been rendered, and whether a video has started, through common interfaces such as PerfImageView, PerfTextView, and PerfVideoView.

With this approach, Android now continuously visualizes User Perceived Latency across 60+ surfaces. Multiple product teams can fairly compare surface performance using the same criteria, and even short-lived features like a Christmas landing page get performance metrics from the moment of launch without any separate instrumentation work.

The key effect is that performance metrics are provided to product engineers essentially for free, making performance more visible and encouraging anyone to optimize their own surface. This model has since been extended to iOS and web as well.