Features

• Add Session.State.Unhandled for unhandled errors that do not terminate the process (#5919)

Improvements

• Move ANR profiling out of experimental (#6042)

Fixes

• Keep dropped tombstone and ANR events dropped, instead of reporting the same app exit again at every app start (#6002)
• Apply Sentry.withScope and Sentry.withIsolationScope data to events captured inside the callback when globalHubMode is enabled (#6004)
  • globalHubMode is enabled by default on Android, where tags, extras, contexts and level set inside the callback were silently dropped
  • Scopes that are explicitly made current, e.g. via Sentry.setCurrentScopes or the SentryContext coroutine integration, are now also honoured when globalHubMode is enabled
  • Sentry.pushScope, Sentry.pushIsolationScope and Sentry.popScope remain no-ops when globalHubMode is enabled
• Drop the profiler_id from transactions and spans when no Perfetto profile covers them, e.g. when Android's ProfilingManager rate limits the profiling request (#6015)
• Prevent events from being dropped when feature flags are added while an event is being captured (#5989)

Internal

• Add InternalSentrySdk.captureEnvelopeNonTerminating for hybrid SDKs (e.g. Flutter) so unhandled exceptions that don't terminate the process no longer end the session as crashed (#5921)
• Add InternalSentrySdk.updateSessionForDroppedEventNonTerminating so hybrid SDKs can still update the session when an error is dropped by sampling (#5990)

Dependencies

• Bump Native SDK from v0.16.4 to v0.16.5 (#6034)
  • changelog
  • diff

Features

• Set app.vitals.start.screen and app.vitals.start.type on standalone app.start children (#6005)
• Add screenshot attachment button to the Android user feedback widget (#5828)
  • Users can now attach a screenshot when submitting feedback. Enabled by default; can be disabled via SentryFeedbackOptions.setEnableAttachScreenshot(false) or the io.sentry.feedback.enable-attach-screenshot manifest flag.
  • Requires the androidx.activity >=1.8.2 dependency
• Add manual Session Replay controls through Sentry.replay() (#5978)
  • Explicit start() and startBuffering() calls bypass the configured replay sample rates; sampling still controls automatic startup.
  • start() starts a full-session replay and does nothing if one is already recording.
  • startBuffering() keeps a rolling buffer that is sent on flush() or an error, then continues in session mode.
  • stop() ends the current replay; the next start() creates a new replay session.
  • pause() suspends recording until resume() and remains paused across background and foreground transitions and automatic replay restarts in the same process.
  • resume() continues the same manually paused replay.
  • flush() sends the current replay data, or starts a full-session replay when recording is stopped.

Fixes

• Prevents inclusion of null. prefix before default-package class names when parsing Java and JNI frames from Android ANR thread dumps (#5979)
• Prevent duplicated breadcrumbs on tombstone-merged native crash events (#5888)
• Prevent a class of Session Replay deadlocks by confining lifecycle state changes to Android's main thread (#5965)
• Symbolicate tombstone native frames for libraries loaded directly from APKs (#5992)
• Prevent a deadlock between the app start extension and the Android performance event processor (#6007)

Performance

• Defer starting Session Replay off the SDK initialization critical path (#5965)
• Use manifest metadata resolved at build time to reduce Android SDK initialization overhead (#5976)

Dependencies

• Bump Native SDK from v0.16.2 to v0.16.4 (#5962, #5996)
  • changelog
  • diff

Features

• Allow child spans to use explicit start timestamps through ISpan (#5929)
• Make ISpan.startChild overloads with SpanOptions public (#5927)

Improvements

• Remove ApiStatus.Experimental annotation from SentrySQLiteDriver (#5938)

Fixes

• Clear contexts when calling Scope.clear() (#5902)
• Preserve custom Throwable identities when R8 optimizes Android apps (#5881)
• Report the correct cpu usage for the first performance sample of a transaction, which was measured against the time since device boot (#5926)
• Prevent an ANR when the Session Replay video encoder gets stuck (#5842)
  • Some hardware encoders never signal end-of-stream, which made the replay worker spin forever while holding the encoder lock. The app's lifecycle callbacks then blocked on that lock and the app froze until the system killed it. The encoder now gives up instead of spinning, and closing the replay cache no longer waits indefinitely for a wedged encoder.

Performance

• Read the clock once per performance collection round instead of once per in-flight transaction (#5934)
• Reduce allocations while collecting cpu usage during transactions by reading the process cpu time via Process.getElapsedCpuTime() instead of parsing /proc/self/stat (33.6kB to 16 bytes per sample on a Pixel 3) (#5926)
• Store performance measurements as primitives, removing a boxed allocation per measurement per performance sample (#5935)

Dependencies

• Bump Native SDK from v0.16.1 to v0.16.2 (#5910)
  • changelog
  • diff

Fixes

• Restore the interrupt flag when cached envelope processing is interrupted between files (#5884)
• Reduce false-positive SDK crash attribution for host app SQLite cursor crashes (#5883)
• Prevent inflated cold app start when the OS spawns the process in the background (e.g. FCM push) on API 35+ (#5841, #5880)
• Preserve single-sample ANR profile chunks so profiles remain available on ANR events (#5872)
• Avoid a CPU busy-loop when recording discarded log or metric envelopes under rate limiting (#5835)
  • ClientReportRecorder now reads the item count from the envelope item header instead of deserializing the payload, which under sustained rate limiting could pin CPU cores while repeatedly throwing exceptions
• Report tasks handed to a no-op ISentryExecutorService as cancelled (#5874)
  • NoOpSentryExecutorService previously returned a Future that was never run and never cancelled, so callers could not tell a dropped task from a queued one and get() would block until its timeout

Performance

• Defer use of reflection by SentryFrameMetricsCollector during Sentry.init (#5886)
• Avoid waiting up to shutdownTimeoutMillis when closing the SDK with a pending transaction timeout or session-end task (#5851)
• Use RGB_565 instead of ARGB_8888 for screenshot and replay capture bitmaps, halving per-frame memory usage (#5821)
• Remove an unused lock from SentryPerformanceProvider, which was allocated on every cold start in ContentProvider.onCreate without ever being acquired (#5871)
• Reduce main-thread allocations when parsing the app start profiling config (#5867)
• Batch and coalesce scope-persistence disk writes to reduce startup cost (#5791)
  • Scope mutations are now coalesced (latest value per field) and breadcrumbs are appended in batches behind a single fsync, instead of one synchronous disk write per mutation.
• Reduce the number of SDK threads: the HostnameCache worker thread now times out while idle instead of staying alive for the whole process lifetime (#5817)

Dependencies

• Bump Native SDK from v0.16.0 to v0.16.1 (#5879)
  • changelog
  • diff

Features

• Use Android's ProfilingManager (Perfetto) for continuous profiling on API 35+ devices (#5251)
  • On API 35+ devices, continuous profiling now automatically uses Android's system ProfilingManager with Perfetto-based stack sampling, providing lower-overhead and more accurate profiles. No configuration change is required.
  • Devices below API 35 keep using the legacy Debug-based profiler.
  • Added an enableLegacyProfiling option (default true) to disable the legacy Debug-based profiler. Setting it to false disables continuous profiling on API < 35 devices as well as transaction-based profiling (profilesSampleRate/profilesSampler) on all devices, since transaction-based profiling is not supported by Perfetto.
  • It can also be configured via the io.sentry.profiling.enable-legacy-profiling manifest flag.
  • See the Android profiling docs for details.

Behavioral Changes

• The outbox and cache directories are no longer created by Sentry.init (#5792)
  • They are now created lazily by whichever component first writes into them, off the init thread. As a result, the directories at SentryOptions.getOutboxPath() and SentryOptions.getCacheDirPath() are not guaranteed to exist once Sentry.init returns.
  • If you write envelopes into the outbox path yourself instead of going through the SDK — as hybrid SDKs do for captureEnvelope — create the directory first, e.g. new File(outboxPath).mkdirs().

Improvements

• Skip building Android manifest metadata debug log messages when debug logging is disabled, reducing allocations during SDK init (#5790)

Fixes

• Use the original app build's ProGuard UUID for ANR profile chunks (#5852)
• Fix potential ANR/deadlock in Session Replay when checkCanRecord runs on the replay executor thread (#5837)
• Prevent concurrent PixelCopy access during Session Replay masking and bitmap cleanup (#5808)
• Release MediaMuxer when the replay video encoder fails to start to avoid a resource leak (#5607)
• Set the correct platform (android instead of java) on ANR profile chunks so they are billed as UI Profile Hours rather than Continuous Profile Hours (#5836)
• Skip encoding and capturing buffered session replay segments while rate-limited, so we don't waste resources on envelopes the transport will drop (#5813)
  • These skipped replays are now reported as ratelimit_backoff discarded events in client reports, so they no longer disappear from drop statistics. One event is recorded per buffer flush rather than per segment.
  • Buffer mode is also kept while rate-limited instead of switching to session mode, so the rolling buffer stays warm and the next error after the rate limit expires can send a complete replay.

Performance

• Create the outbox and cache directories lazily in their consumers instead of during SDK init, moving the mkdirs() calls off the init (main) thread (#5792)
• Reduce the number of SDK threads: LifecycleWatcher now schedules the session-end task on the shared timer executor instead of creating a dedicated java.util.Timer thread (#5819)
• Reduce the number of SDK threads: RateLimiter now schedules its rate-limit-lifted notifications on the shared timer executor instead of creating a dedicated java.util.Timer thread (#5814)
• Speed up deserialization of arbitrary JSON objects by typing numbers without throwing exceptions (#5783)

Dependencies

• Bump Native SDK from v0.15.4 to v0.16.0 (#5845)
  • changelog
  • diff

Fixes

• Pin the published Sentry Android SDK's AAR metadata minCompileSdk to our minSdk (21) instead of AGP 9's new default of the SDK's own compileSdk (37), so apps that depend on the SDK aren't forced to raise their compileSdk (#5823)

Android 17 support

• We've put Android 17 through a set of rigorous tests. We're now officially giving it the Sentry stamp of compatibility .(#5796)

Fixes

• Reduce main-thread work during Sentry.init by resolving the shake-detector accelerometer off the main thread (~1.75ms on a Pixel 10) (#5784)
• Backfill release, environment, distribution, tags, and app version/build—and use the matching replay-on-error sample rate—for ApplicationExitInfo ANR and native crash events captured before SDK initialization, without reusing options cached by a later app update (#5762)
• SentryTagModifierNode.isImportantForBounds now matches the default behavior and returns true (#5789)
• Prevent a StackOverflowError when a beforeSend, beforeBreadcrumb, beforeSendLog, or beforeEnvelope callback triggers another capture (directly or through a logging integration such as Timber) (#5737)
  • Captures made from within a user callback (event, transaction, breadcrumb, log, envelope, or check-in) are now dropped while that callback runs, instead of recursing. Captures made by event processors are unaffected.
• Replace deprecated ThrowableProxy with LogEvent#getThrown() in sentry-log4j2 (#5751)

Dependencies

• Bump Native SDK from v0.15.3 to v0.15.4 (#5793)
  • changelog
  • diff
• The SDK is now compiled with Android Gradle Plugin 9.2.1 (#5779)

Features

• Session Replay: Record segment names (transaction names) (#5763)

• Add io.sentry:sentry-opentelemetry-bom to align Sentry OpenTelemetry modules with tested OpenTelemetry dependencies (#5629)

  • Spring Boot Gradle plugin: add the Sentry BOM to dependencyManagement; explicit imports are applied after Spring Boot's implicit BOM

    dependencyManagement {
      imports {
        mavenBom("io.sentry:sentry-opentelemetry-bom:<sentry-version>")
      }
    }

  • Gradle: import it as a platform and omit versions from Sentry OpenTelemetry and OpenTelemetry dependencies

    implementation(platform("io.sentry:sentry-opentelemetry-bom:<sentry-version>"))

  • Maven: import it before Spring Boot's BOM in the same <dependencyManagement> block, or in the child POM when using spring-boot-starter-parent

    <dependency>
      <groupId>io.sentry</groupId>
      <artifactId>sentry-opentelemetry-bom</artifactId>
      <version>${sentry.version}</version>
      <type>pom</type>
      <scope>import</scope>
    </dependency>

Fixes

• Session Replay: Fix first recording segment missing for replays in buffer mode (#5753)
• Session Replay: Fix error-to-replay linkage in buffer mode (#5754)
• Prevent logs and metrics from remaining queued after a flush scheduling race (#5756)
• Fix main thread identification for tombstone (native crash) events (#5742)
• Prevent malformed JDBC URLs, which may contain credentials, from being printed to stdout (#5656)
• Restrict JVM-global proxy authentication credentials to challenges from the configured proxy host (#5656)
• Sanitize Spring 7 and Spring Jakarta WebClient span descriptions to prevent embedded URL credentials from being sent to Sentry (#5656)
• Respect tracePropagationTargets when injecting Sentry tracing headers through the OpenTelemetry OTLP propagator (#5656)

Performance

• Schedule transaction idle/deadline timeouts on a shared, dedicated executor instead of spawning a Timer thread per transaction (#5670)

Dependencies

• Bump OpenTelemetry to support Spring Boot 4.1 (#5573)
  • If this causes issues for you because you are also using Spring Boot Dependency Management Plugin (io.spring.dependency-management),
    which may downgrade the OpenTelemetry SDK, please have a look at the changelog entry above that explains how to use sentry-opentelemetry-bom.
  • OpenTelemetry to 1.63.0 (was 1.60.1)
  • OpenTelemetry Instrumentation to 2.29.0 (was 2.26.0)
  • OpenTelemetry Instrumentation Alpha to 2.29.0-alpha (was 2.26.0-alpha)
  • OpenTelemetry Semantic Conventions to 1.42.0 (was 1.40.0)
  • OpenTelemetry Semantic Conventions Alpha to 1.42.0-alpha (was 1.40.0-alpha)
• Bump Native SDK from v0.15.2 to v0.15.3 (#5728)
  • changelog
  • diff

Features

• Add Sentry.extendAppStart(), Sentry.finishExtendedAppStart(), and Sentry.getExtendedAppStartSpan() to extend the app start measurement past the first frame for extra launch-time work on Android (#5604)

  • Requires standalone app start tracing (options.isEnableStandaloneAppStartTracing). Call extendAppStart() in Application.onCreate after SDK init and finishExtendedAppStart() when done:

  Sentry.extendAppStart()
  
  // Optionally, retrieve the extended app start span to attach your own child spans
  val child = Sentry.getExtendedAppStartSpan()?.startChild("preload", "Preload resources")
  // ... extra launch-time work ...
  child?.finish()
  
  Sentry.finishExtendedAppStart()

• Add trace_metric_byte data category and record byte-level client reports when trace metrics are discarded (#5626)

• Support the io.sentry.tombstone.report-historical manifest option to enable historical tombstone reporting via AndroidManifest.xml <meta-data> (#5683)

Fixes

• Fix NoSuchMethodError from using Math.floorDiv/Math.floorMod overloads that are unavailable on Java 8 (#5743)
• Fix main thread identification parsing for ApplicationExitInfo ANRs (#5733)
• Do not send threads without stacktraces for ApplicationExitInfo ANRs (#5733)
• Record byte-level client reports when event processors discard logs or trace metrics (#5718)
• Name the device-info caching thread SentryDeviceInfoCache so all threads spawned by the SDK are identifiable (#5684)
• Apply byte-category rate limits to log and trace metric envelope items (#5716)

Performance

• Skip Hint allocation in Scope.addBreadcrumb when no beforeBreadcrumb callback is set (#5689)
• Speed up scope persistence by detecting the Sentry executor thread via a marker instead of a Thread.getName() name scan on every scope mutation (#5691)
• Remove executor prewarm during SDK init (#5681)
  • The single-threaded SentryExecutorService queued the prewarm work ahead of the first useful task, so it could only delay init work, never speed it up; the thread and class loading it warmed are paid identically by the first real task submitted right after.

Behavioral Changes

• SentryOkHttpInterceptor::intercept now throws IOException. This is a source-only and Java-only breaking change (#5654)

Fixes

• Don't start a redundant UI interaction transaction when a transaction is already bound to the Scope (#5491)
  • Previously, SentryGestureListener always started a UI transaction and only afterwards skipped binding it to the Scope when a manually-bound transaction already existed, leaving the new transaction to be dropped as an idle transaction without children.
• Fix potential NPE within Scope.endSession() (#5657)
• Fix memory leak in ReplayIntegration due to persisting executor not being shut down (#5627)
• Fix AbstractMethodError when compose-ui 1.11+ is used in combination with Modifier.sentryTag() or the Sentry Kotlin compiler plugin (#5672)

Performance

• Speed up touch gesture target detection on deeply nested view hierarchies by hit-testing in local coordinates instead of calling getLocationOnScreen per view (#5595)
• Probe class availability without initializing the class during SDK init (#5635)
• Avoid constructing an exception per view when resolving view ids during view-hierarchy and gesture capture (#5631)
• Start the frame metrics thread lazily on first collection instead of during SDK init (#5641)
• Reduce SentryId and SpanId allocation overhead by replacing their per-instance LazyEvaluator (and its lock) with a lightweight lazily-generated String. (#5645)
• Lazily allocate the ReentrantLock backing AutoClosableReentrantLock to avoid eager lock allocations for SDK objects that never contend during SentryAndroid.init (#5643)