Mobile App Observability: Beyond Crash-Free Sessions

Q: Why is crash-free sessions a misleading mobile metric?

It counts only one kind of failure, process crashes, and it counts it per session. Users leave over freezes, slow starts, hung network calls and battery drain, none of which are crashes. And a session percentage dilutes harm: a heavy user who freezes once a day across ten sessions looks 90% healthy. Measure users affected, per device model, for crashes, ANRs and hangs together.

Q: What are the Google Play Android vitals bad behaviour thresholds?

Google's Android vitals documentation lists a user-perceived crash rate threshold of 1.09% overall and 8% per phone model, and a user-perceived ANR rate threshold of 0.47% overall and 8% per phone model. Exceeding them can reduce your app's visibility on Google Play. The page also lists memory and code optimisation thresholds, with possible visibility impact starting from February 2027.

Q: What is an ANR on Android?

An ANR, Application Not Responding, is raised when the app's main thread is blocked for too long. Android's documentation lists several conditions, including the app not responding to an input event such as a key press or screen touch within 5 seconds. ANRs are usually caused by slow work on the main thread such as disk or network access, heavy computation or lock contention, and users experience them as a frozen app.

Q: How does iOS define an app hang?

Apple's documentation on understanding hangs explains that most of its developer tools start reporting issues when the main run loop is unresponsive for more than 250 milliseconds, and that delays under 100 milliseconds in a discrete interaction are rarely noticeable. Hangs are reported through Xcode Organizer, Instruments and MetricKit, which delivers daily metric and diagnostic reports from real users' devices.

Q: Why does backend observability not work well for mobile apps?

Phones are offline or constrained much of the time, so telemetry arrives late or never. The failure often kills the reporter, such as a hung request that never logs or a frozen app the user swipes away. The fleet spans thousands of device models you do not control, telemetry costs users battery and data, and privacy constraints are stricter. Mobile needs store-and-forward delivery and per-device-model analysis.

Q: Can I use OpenTelemetry for mobile apps?

Yes. The OpenTelemetry Android project and its Swift counterpart let you emit spans and events from the device in OTLP, the same format your backend uses, so one pipeline can receive both. The most valuable use is propagating trace context from the user action on the phone into backend requests, so a slow journey can be split into device time, network time and server time for the same request.

Q: Which metrics should gate a mobile release rollout?

Compare each release with the previous one at the same rollout stage on a short list: users affected by crashes, ANRs and hangs overall and for your top device models; cold start time at the 75th percentile per device tier; completion rates for three to five critical journeys such as sign-in and payment; and memory per device tier. Halt and flag off the offending feature if any regresses on a major model.

Q: Should every mobile app build full client observability?

No. Small apps with modest traffic and one team should use a crash SDK, read Play vitals and Xcode Organizer every release, and fix the slow screens they already know about. Full client instrumentation, store-and-forward delivery and trace propagation pay off when you have many device models, several teams, staged rollouts and revenue journeys where a regression costs real money within days.

Satyam Kumar

Назад к блогу

architecture

Mobile App Observability: Crash-Free Sessions Is the Wrong Number, and the Phone Is a Server You Cannot SSH Into

By Satyam KumarOctober 26, 202611 min read

mobile app observability crash-free sessions android vitals anr rate app hangs ios metrickit google play bad behavior threshold mobile performance monitoring app start time opentelemetry android mobile distributed tracing staged rollout gating mobile release quality app store rating drop client telemetry memory vitals

Mobile App Observability: Crash-Free Sessions Is the Wrong Number, and the Phone Is a Server You Cannot SSH Into

Frequently Asked Questions

Поделиться статьёй

Twitter LinkedIn WhatsApp

Satyam Kumar

Founder & AI Architect, AppScale LLP

AI & Cloud архитектор. Помогаю командам строить системы, масштабируемые до миллионов.

LinkedIn GitHub

Mobile App Observability: Crash-Free Sessions Is the Wrong Number, and the Phone Is a Server You Cannot SSH Into

Frequently Asked Questions

Поделиться статьёй

Comments

Leave a comment