Opentelemetry Experts On Tough Telemetry Challenges In Mobile

The play is changing for frontend observability, arsenic we’re seeing awesome organization engagement successful improving OpenTelemetry support for web apps and mobile apps. For example, there’s a caller Browser Special Interest Group (SIG) successful nan OpenTelemetry project, and they’re moving to amended OTel support for nan browser runtime. You tin study much astir what they’ll beryllium moving connected successful this on-demand sheet discussion.

The OTel organization besides has dedicated Android and Swift SIGs for improving nan APIs, instrumentation libraries and semantic conventions for collecting telemetry connected nan 2 native mobile app platforms. And organizations are taking note, pinch a caller study conducted by Enterprise Management Associates (EMA) revealing that adoption of OpenTelemetry for mobile information collection is group to triple successful nan adjacent 12 to 24 months.

I sat down pinch respective cardinal members of nan Android and Swift SIGs for a fun, fall-themed sheet discussion connected nan cardinal challenges successful mobile telemetry postulation and nan authorities of OpenTelemetry support for mobile. Panelists included:

• Ari Demarco, iOS package technologist astatine Embrace, OTel Swift maintainer.
• Bryce Buchanan, main technologist astatine Elastic, OTel Swift maintainer.
• Hanson Ho, Android designer astatine Embrace, OTel contributor and OTel Android approver.
• Jason Plumb, elder main package technologist astatine Splunk, OTel Android maintainer and OTel Java approver.
• Nacho Bonafonte, elder package engineer, OTel Swift maintainer.

Challenges With Collecting Telemetry connected Mobile Platforms

When mobile developers usage OpenTelemetry, they must beryllium mindful of nan sheer standard of data that mobile apps tin generate. Buchanan mentioned that while backend systems tally connected thousands of clients nether tightly controlled conditions, mobile apps tin tally connected millions of clients.

Demarco chimed in, “That besides leads to nan problem of information volumes, because depending connected nan app, a mobile exertion tin make an tremendous magnitude of telemetry. So, dissimilar backends that you tin power sampling centrally, successful mobile, nan sampling decisions astir apt should beryllium made on-device pinch benignant of constricted visibility into nan bigger picture. And past you person nan question, if you oversample, you’ll discarded a batch of bandwidth aliases battery. […] But if you undersample, you astir apt miss captious telemetry that is basal to place issues aliases understand behaviors.”

Mobile developers are besides hyper-focused connected nan capacity of their apps, which tin beryllium affected by nan operational costs of capturing telemetry. Plumb mentioned respective things developers must support successful mind, including which API calls nan app must make to nan platform, really agelong nan app spends successful those callbacks aliases arena handlers and besides nan payload size of web requests connected nan wire.

“Efficiently handling those payloads is besides thing people, I think, are specifically challenged pinch connected mobile that doesn’t beryllium successful different platforms, and we don’t person nan luxury of conscionable …[scaling] horizontally, like, occurrence up a fewer much instances,” said Plumb.

The platforms that mobile apps tally connected are besides tightly controlled by Google and Apple. As Bonafonte said, “The privateness that nan level puts you successful is thing that’s difficult.” Mobile developers request support from nan operating strategy to cod data, truthful if nan strategy doesn’t let them to cod definite types of telemetry, they’re constricted successful really they tin efficaciously observe their applications.

Unlike servers, mobile apps person a life rhythm complexity, which tin make it incredibly difficult to understand nan conditions that lead to issues.

As Demarco pointed out, “Mobile apps don’t tally continuously, truthful they are suspended, backgrounded, terminated, killed by OS, there’s a crash, … nan OS tin pre-warm your application, nan exertion could motorboat because of a push notification, a inheritance fetch aliases because a quality tapped into nan icon. So, erstwhile do you flush your telemetry? … How do you way convention continuity crossed app restarts? What happens to, I don’t know, in-flight spans whenever there’s a crash, aliases nan OS kills your process? So there’s a bunch of complexity successful position of what do you determine to do successful those cases? And it’s not trivial … conscionable solving 1 of those questions is not a one-liner point you’ll lick successful your code. It’s thing you really person to deliberation done to really lick that.”

Challenges Mobile Devs Have With Observability Practices

Traditionally, observability is seen arsenic being wrong nan purview of backend teams, and arsenic such, mobile developers often don’t understand it. Ho mentioned that mobile developers mostly interact pinch OpenTelemetry because they’re told to arsenic opposed to being thing they themselves scope for.

“Tracing and … telemetry is not a halfway competency of mobile developers … because, you know, nan challenges that they look are different. … There’s truthful overmuch to really thatch a team, and nan architecture, nan mobile app architectures besides aren’t ace good designed for maintainable instrumentation,” said Ho.

Product managers mightiness want amended visibility to explicate nan capacity (or deficiency thereof) successful a caller feature, truthful they inquire mobile developers to cod much observability data. But neither nan mobile developer nor nan merchandise head knows what to collect. This deficiency of clarity erstwhile it comes to observability instrumentation for mobile apps was a communal thread successful our discussion.

Buchanan mentioned that moreover thing arsenic elemental arsenic erstwhile you should commencement a span is not trivial connected a mobile device. “On a backend, it’s very trivial. It’s like, ‘Oh, erstwhile I get a request, that’s erstwhile a span starts.’ But for a mobile developer, … should I do it erstwhile personification clicks a button? When a web starts? … There’s nary correct reply to that, like, really should you instrumentality that? It really depends connected what your app does and what you’re trying to monitor.”

Plumb agreed that OpenTelemetry doesn’t person fantabulous guidance for developers astir immoderate of these client-side usage cases.

“We don’t yet person a really bully information exemplary aliases conscionable a conceptual explanation of what sessions are.”

He contrasted this situation pinch backend observability tooling that has respective usage cases very well-defined astatine this point. For example, each vendor that has a tracing solution is going to person a trace waterfall view, and each real personification monitoring (RUM) vendor is going to person a measurement to analyse funnels.

As Ho pointed out, “When you’re a backend service, nan extremity is to return nan petition and sprout retired nan response. You want to log really agelong that took and if there’s thing absorbing that’s happening successful nan middle. The extremity is simple. The extremity of a mobile app is to beryllium defined.”

What nan Uber Eats squad cares astir is different from nan Pinterest team, which is different from a banking app.

“To understand nan goals and translating that into what benignant of telemetry is simply a non-trivial leap. It seems trivial, if you haven’t done it, but erstwhile you do it, you’re, like, ‘I attraction astir everything.’ Do you really attraction astir everything?” said Ho.

Better OpenTelemetry Support for Android and Swift

The Android and Swift SIGs are improving nan developer acquisition of utilizing OpenTelemetry. Beyond manually capturing cardinal OpenTelemetry signals of logs and traces, some SDKs tin besides seizure mobile-specific telemetry:

• The Android SDK has instrumentation for Application Not Responding (ANR) errors, clang reporting, position search and web calls.
• The Swift SDK has instrumentation for web calls made utilizing URLSession arsenic good arsenic convention events that trigger disconnected app lifecycle changes.

The Swift SIG besides addressed a cardinal situation that stems from moving successful Apple’s tightly controlled mobile platform. Apple’s charismatic package manager, Swift Package Manager, requires downloading each limitations of each libraries successful your projects, moreover if you don’t usage them successful your application. As a consequence, nan OpenTelemetry Swift repository was very large, which meant mobile developers faced ample package download sizes to usage OTel successful their iOS apps.

As Bonafonte shared, “[OpenTelemetry Swift] had to support a protobuf OTLP [OpenTelemetry Line Protocol] protocol pinch protobuf, and that intends that you person a dependency connected Apple connected a room from Apple that has a dependency of different room from Apple, and it has a dependency of different room and different and different and another.”

Ari chimed in, “Whenever you person to download it, aliases compile your application, tally tests, tally this successful CI, build nan exertion and deploy that, each that takes a bunch of time, and obviously, for example, successful position of CI, minutes is money, truthful … for each azygous iOS developer, it was going to beryllium a pain. And probably, possibly they conscionable wanted to usage nan API aliases conscionable our implementation of nan OpenTelemetry SDK.”

As a solution, nan Swift SIG divided nan codification into 2 abstracted repositories. The charismatic OpenTelemetry Swift repository is nan main repository, and it contains everything needed to activity pinch OTLP. The maintainers created different repository called OpenTelemetry Swift Core, which only contains nan OpenTelemetry Swift API and OpenTelemetry Swift SDK. Those 2 pieces are nan bare minimum to get started, create traces and emit logs. iOS developers tin now instrumentality applications, process information and export it without each nan overhead of nan main repository.

The Android SIG is moving connected 3 main improvements. The first is amended stabilization for nan initialization API for nan Android agent, and is expected to beryllium completed soon. The 2nd is broadening nan instrumentation, which includes enhancing support for build-time auto-instrumentation.

As Plumb said, “The 3rd category, which is, I think, possibly conscionable arsenic important, are semantic conventions. … With each spot of instrumentation, pinch each benignant of caller characteristic that we’re adding, we’re trying to reflector that successful nan semantic conventions, moreover if nan first walk is successful improvement aliases experimental, astatine slightest having that retired location and documented, what it means, what nan intent is erstwhile you spot a portion of information marked pinch this name, what these attributes bent disconnected of it mean.”

The situation is being inclusive of each nan different opinions erstwhile it comes to watching mobile apps. An illustration Ho gave for nan complexity successful defining a mobile convention was nan problem of foreground vs. inheritance app behavior. When should a convention commencement for a podcast app that runs mostly successful nan background? What happens for apps that are mostly tally successful nan foreground, but person inheritance activities that trigger contented refreshes and personification interface (UI) updates to beryllium fresh connected nan adjacent app launch? What should a convention beryllium for point-of-sale (POS) apps that tally perpetually successful nan foreground for galore hours astatine a time?

Ho past shared an Android SIG improvement successful Embrace’s Kotlin API aid proposal to OpenTelemetry. Kotlin is nan charismatic connection of Android development; however, nan OpenTelemetry Android SDK presently utilizes nan OpenTelemetry Java SDK nether nan hood to grounds telemetry. Adopting a Kotlin API and SDK would make it easier for mobile developers to usage OpenTelemetry because nan Kotlin programming connection is much familiar, idiomatic and usable successful modern mobile exertion development.

Group Created pinch Sketch.