Add Video Analytics in call-join flow - #1716
Merged
aleksandar-apostolov merged 35 commits intoJun 12, 2026
Merged
Conversation
Contributor
PR checklist ✅All required conditions are satisfied:
🎉 Great job! This PR is ready for review. |
rahul-lohra
changed the title
Contributor
SDK Size Comparison 📏
|
Contributor
WalkthroughThis PR adds comprehensive analytics instrumentation across the video call lifecycle. New analytics state machines track join progression, SFU WebSocket connection, peer connection ICE states, and media frame rendering. A centralized ChangesCall Analytics Instrumentation
Estimated code review effort🎯 4 (Complex) | ⏱️ ~60 minutes Suggested labels
Suggested reviewers
🚥 Pre-merge checks | ✅ 4 | ❌ 1❌ Failed checks (1 warning)
✅ Passed checks (4 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. ✨ Finishing Touches📝 Generate docstrings
🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
Contributor
There was a problem hiding this comment.
Actionable comments posted: 16
Caution
Some comments are outside the diff and can’t be posted inline due to platform limitations.
⚠️ Outside diff range comments (2)
stream-video-android-core/src/main/kotlin/io/getstream/video/android/core/Call.kt (1)
693-785:⚠️ Potential issue | 🟠 Major | ⚡ Quick winPropagate
joinAnalyticsModelinto SFU connect.At Line 773,
_joincallsconnectInternal()withoutjoinAnalyticsModel, so SFU websocket failure events in initial join retries always emitretryCount=0.Suggested fix
- when (val result = session.value?.connectInternal()) { + when ( + val result = + session.value?.connectInternal( + joinAnalyticsModel = joinAnalyticsModel, + ) + ) {🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@stream-video-android-core/src/main/kotlin/io/getstream/video/android/core/Call.kt` around lines 693 - 785, _join currently calls session.value?.connectInternal() without passing joinAnalyticsModel, so SFU websocket connect attempts don't receive the analytics model and retry events show retryCount=0; update the call in _join to pass joinAnalyticsModel into connectInternal (i.e., session.value?.connectInternal(joinAnalyticsModel)), and if RtcSession.connectInternal does not accept that parameter, add a parameter to RtcSession.connectInternal (and any other callers) to accept JoinAnalyticsModel and thread it through to the SFU websocket connection logic so initial join retries emit correct analytics.stream-video-android-core/src/main/kotlin/io/getstream/video/android/core/call/RtcSession.kt (1)
691-706:⚠️ Potential issue | 🟠 Major | ⚡ Quick winAvoid hardcoded
retryCount = 0on successful SFU WS completion.At Line 702, success completion is emitted with a constant retry count, so successful rejoin/migrate attempts are reported with incorrect retry metadata.
Suggested fix
@@ - call.callAnalytics.sfuAnalytics.onSfuWsCompleted( - success = true, - retryCount = 0, - ) } @@ is SfuSocketState.Connected -> { + call.callAnalytics.sfuAnalytics.onSfuWsCompleted( + success = true, + retryCount = joinAnalyticsModel?.retryAttempt ?: 0, + ) sendConnectionTimeStats(reconnectDetails?.strategy) SfuConnectionResult.Connected }Also applies to: 887-911
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@stream-video-android-core/src/main/kotlin/io/getstream/video/android/core/call/RtcSession.kt` around lines 691 - 706, Replace the hardcoded retryCount = 0 in the SfuSocketState.Connected success handler with the actual SFU websocket retry counter used by the session (the same counter incremented during rejoin/migrate attempts), e.g. use the session's sfuReconnectAttempts or sfuRejoinAttemptCount variable instead of 0; update both places where onSfuWsCompleted(success = true, retryCount = 0) is called (the SfuSocketState.Connected block and the similar block around lines 887-911) and ensure the chosen counter is reset or read consistently when emitting the success event.
🧹 Nitpick comments (1)
stream-video-android-core/src/main/kotlin/io/getstream/video/android/core/analytics/call/observer/PeerConnectionAnalyticsStateHolder.kt (1)
73-76: ⚡ Quick winMake
PeerConnectionAnalyticsStateimmutable to preserve flow snapshot semantics.Using
varhere allows in-place mutation ofstate.valuefields without emitting a newStateFlowvalue. Converting these tovalkeeps updates funneled throughcopy(...)andMutableStateFlow.update.♻️ Suggested change
internal data class PeerConnectionAnalyticsState( - var peerConnectionObserverJob: Job? = null, - var publisherJob: Job? = null, - var subscriberJob: Job? = null, + val peerConnectionObserverJob: Job? = null, + val publisherJob: Job? = null, + val subscriberJob: Job? = null, val publisherStage: Stage = Stage.NOT_STARTED, val subscriberStage: Stage = Stage.NOT_STARTED, )🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@stream-video-android-core/src/main/kotlin/io/getstream/video/android/core/analytics/call/observer/PeerConnectionAnalyticsStateHolder.kt` around lines 73 - 76, The PeerConnectionAnalyticsState data class currently uses mutable vars (peerConnectionObserverJob, publisherJob, subscriberJob) which permits in-place mutation of state.value and breaks StateFlow snapshot semantics; change those properties to val so the class is immutable, and update call sites to modify the state via PeerConnectionAnalyticsState.copy(...) inside MutableStateFlow.update { current -> current.copy(...) } (or assign a new instance) so all changes emit new StateFlow values and preserve snapshot semantics.
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In `@stream-video-android-core/api/stream-video-android-core.api`:
- Around line 14728-14729: SfuSocketConnection's public deprecated constructor
changed parameter order causing binary-compat breaks; add a new public
deprecated constructor overload that preserves the original parameter ordering
(the version with CoroutineScope as the last positional parameter) which
delegates to the internal primary constructor (or to the existing public
delegating constructor) so existing compiled callers continue to work, and add a
short KDoc on the deprecated overload explaining the migration to the new
constructor signature and recommending named-argument usage. Ensure the new
overload's signature and `@Deprecated` marker match the previous public API and
delegate without duplicating initialization logic (refer to SfuSocketConnection
constructors and the internal primary <init>).
In
`@stream-video-android-core/src/main/kotlin/io/getstream/video/android/core/analytics/call/observer/AudioAnalytics.kt`:
- Around line 46-59: The isEnabled flag in AudioAnalytics is hardcoded to false,
preventing observeParticipantsForFirstRemoteAudioFrame and related logic from
ever running; change isEnabled to be configurable at runtime (e.g., accept a
constructor parameter, inject a FeatureFlag/Config, or read from a runtime
setting) and update any call sites that construct AudioAnalytics to pass the
correct flag or flag provider; ensure the
observeParticipantsForFirstRemoteAudioFrame, recordedFirstFrame, trackSinks, and
observeJob behavior remains identical when enabled, and remove the early return
guard that checks the hardcoded isEnabled so the observer can run when the
runtime flag is true.
In
`@stream-video-android-core/src/main/kotlin/io/getstream/video/android/core/analytics/call/observer/JoinAnalyticsStateHolder.kt`:
- Around line 84-90: JoinTelemetryState currently has mutable properties (vars:
joinStageAttemptId, stageId, joinStage, callSessionId) which allow in-place
mutation of state.value and can break StateFlow emission semantics; change these
mutable properties to immutable vals so the data class is fully immutable, and
ensure any updates to the held state are done via the state holder's updater
(e.g., _state.update { it.copy(...) }) rather than mutating fields directly;
locate the JoinTelemetryState data class and replace the var declarations with
val for joinStageAttemptId, stageId, joinStage, and callSessionId.
In
`@stream-video-android-core/src/main/kotlin/io/getstream/video/android/core/analytics/call/observer/VideoAnalytics.kt`:
- Around line 51-54: The Log.d call in VideoAnalytics (tag "VideoObserver")
exposes sensitive identifiers (videoTrackId, videoSessionId, callSessionId);
update the logging in the method containing this debug (the
firstVideoFrameRendered/observer log block inside VideoAnalytics) to remove or
redact those identifiers (e.g., replace with fixed "[REDACTED]" or a
non-reversible short hash) or gate the detailed log behind a development-only
flag, and keep only non-sensitive info (trackType, width, height) in production
logs.
- Around line 62-65: The event reporting is using
joinAnalyticsStateHolder.state.value.callSessionId (which can be stale) instead
of the function parameter callSessionId; update the reporter call in
VideoAnalytics (the method that builds the join event) to pass the method
argument callSessionId into the event payload (e.g., where joinReason and
trackId are set) rather than reading callSessionId from
joinAnalyticsStateHolder.state.value so the reported session ID reflects the
current callSessionId parameter.
- Around line 45-50: The analytics misses screen-share because getTrack is
hard-coded to TrackType.TRACK_TYPE_VIDEO; update the track lookup in
VideoAnalytics (the branch handling TrackType.TRACK_TYPE_VIDEO,
TrackType.TRACK_TYPE_SCREEN_SHARE) to pass the dynamic trackType variable to
rtcSession?.subscriber?.value?.getTrack(videoSessionId, trackType) (instead of
TrackType.TRACK_TYPE_VIDEO) so screen-share tracks resolve and
FIRST_VIDEO_FRAME_RENDERED gets reported; keep the existing videoSessionId vs
callSessionId check and asVideoTrack()/video?.id() usage.
In
`@stream-video-android-core/src/main/kotlin/io/getstream/video/android/core/analytics/coordinator/CoordinatorAnalytics.kt`:
- Around line 53-71: In the VideoSocketState handling (branches for
VideoSocketState.Connected and
VideoSocketState.Disconnected.DisconnectedPermanently) call
eventReporter.reportCoordinatorWSCompleted as you do, then immediately reset the
stageId to an empty value (e.g., stageId.value = "") to avoid duplicate
completion reports on later terminal-state emissions; update the branches that
reference CoordinatorSocketStateService.Companion.lastRetryAttempts to report
first, then clear stageId.value so the session is no longer considered
in-flight.
In
`@stream-video-android-core/src/main/kotlin/io/getstream/video/android/core/analytics/reporting/ClientEventReporter.kt`:
- Around line 440-515: The three reporter methods reportFirstAudioFrameRendered,
reportFirstVideoFrameRendered, and reportMediaPermissionStatus only emit
EventType.INITIATED and return a stageId but have no matching completion events,
so create paired completion APIs that send EventType.COMPLETED with the same
stageId and relevant outcome data: add functions like
reportFirstAudioFrameRenderedCompleted(stageId: String, sfuId: String, callId:
String, callType: String, joinStageAttemptId: String, callSessionId: String,
joinReason: JoinReason, success: Boolean, optional error/reason: String?),
reportFirstVideoFrameRenderedCompleted(stageId: String, trackId: String, ...
success/error?), and reportMediaPermissionStatusCompleted(stageId: String,
callId: String, callType: String, joinStageAttemptId: String, joinReason:
JoinReason, cameraAllowed: Boolean, microphoneAllowed: Boolean), each calling
clientEventFactory.buildRequest with the same EventStage
(EventStage.Call.FIRST_AUDIO_FRAME_RENDERED, FIRST_VIDEO_FRAME_RENDERED,
MEDIA_DEVICE_PERMISSION), eventType = EventType.COMPLETED, and the original
stageId so elapsed/outcome can be recorded.
- Around line 107-111: The PreCallInFlightSession is being created with the
wrong stage constant: when opening a coordinator websocket pre-call session (the
entry stored in postCallFlightSessions inside ClientEventReporter), replace
EventStage.Call.COORDINATOR_JOIN with EventStage.CoordinatorWs so the in-flight
metadata reflects a coordinator WS session; update the PreCallInFlightSession
instantiation (the call that sets stage = ...) to use EventStage.CoordinatorWs.
- Around line 134-142: The COMPLETED coordinator WS event built in
clientEventFactory.buildRequest is missing the stageId, preventing correlation
with the INITIATED event; update the call in ClientEventReporter (the
buildRequest invocation where stage = EventStage.CoordinatorWs and eventType =
EventType.COMPLETED) to pass the stageId argument (use the same stage.id or
stageId value used when building the INITIATED event) so the COMPLETED event
contains stageId, keeping all other fields (outcome, retryCountAttempt,
retryFailureCode, elapsedTime, retryFailureReason) unchanged.
In
`@stream-video-android-core/src/main/kotlin/io/getstream/video/android/core/analytics/reporting/datasource/FileBasedPendingEventDataSource.kt`:
- Around line 35-39: The constructor of FileBasedPendingEventDataSource must
validate batchSize to prevent non-positive values that cause loadAndClear() to
never drain files; add a guard in the FileBasedPendingEventDataSource
initializer (or primary constructor) that checks if batchSize <= 0 and either
throw an IllegalArgumentException with a clear message or coerce it to
DEFAULT_BATCH_SIZE, so any call creating this class (and code paths using
batchSize in loadAndClear()) will never operate with a non-positive batch size.
- Around line 60-70: The current loadAndClear in FileBasedPendingEventDataSource
reads and removes lines from the NDJSON file before deserializing them, which
can permanently drop events if parsing fails; change the logic in loadAndClear
to first read the batch lines (use file.readLines().filter { it.isNotBlank() }
and take(batchSize)), attempt to deserialize those lines via adapter.fromJson
(using runCatching) and collect results and parse failures, and only if parsing
succeeded for all intended lines mutate the file (delete if no remaining lines,
or write remaining.joinToString(...)); if any line in the batch fails to parse,
do not delete/overwrite the file (leave it intact) and return only the
successfully parsed items or handle according to existing failure semantics —
update references in this method (FileBasedPendingEventDataSource, loadAndClear,
adapter.fromJson, file.delete, file.writeText) accordingly.
In
`@stream-video-android-core/src/main/kotlin/io/getstream/video/android/core/analytics/reporting/datasource/PendingEventDataSource.kt`:
- Line 45: The loadAndClear() implementation takes a snapshot and clears in two
non-atomic steps which lets a concurrent save() slip in and be lost; make
loadAndClear() and save() mutually exclusive by protecting accesses to the
shared queue with the same synchronization mechanism (e.g., synchronized(this)
block or a dedicated Mutex) so that loadAndClear() performs snapshot + clear
atomically and save() appends under the same lock; update functions
loadAndClear() and save() to acquire/release that lock around all queue
operations.
In
`@stream-video-android-core/src/main/kotlin/io/getstream/video/android/core/analytics/reporting/dispatcher/ImmediateEventDispatcher.kt`:
- Around line 58-62: ImmediateEventDispatcher currently keys the "don't requeue"
decision on e.message == WONT_RETRY but retryInternal never returns that
sentinel; instead make the control flow based on exception type: introduce/use a
dedicated NonRetryableException (or reuse an existing domain exception) thrown
by retryInternal when events should not be retried, and change the onFailure
handler in ImmediateEventDispatcher to check for that exception type (e.g., if
(e is NonRetryableException) skip dataSource.save(events) else
dataSource.save(events)); update retryInternal to throw NonRetryableException
where appropriate so non-retryable failures are not re-queued.
In
`@stream-video-android-core/src/main/kotlin/io/getstream/video/android/core/analytics/reporting/model/AnalyticsFailureCodes.kt`:
- Around line 25-26: The enum entries REQUEST_TIMEOUT and SFU_REQUEST_TIMEOUT in
AnalyticsFailureCodes.kt are colliding because both use the same code string
"REQUEST_TIMEOUT"; change the SFU_REQUEST_TIMEOUT code string to a unique
identifier (for example "SFU_REQUEST_TIMEOUT") while keeping/describing the
human-readable message ("SFU connection timed out") intact so backend analytics
can differentiate the two failure classes; update only the code string for the
SFU entry in the AnalyticsFailureCodes enum (refer to the symbols
REQUEST_TIMEOUT and SFU_REQUEST_TIMEOUT) and run tests/compile to verify no
other references need adjusting.
In
`@stream-video-android-core/src/main/kotlin/io/getstream/video/android/core/socket/coordinator/CoordinatorSocketStateService.kt`:
- Around line 34-37: The companion-object counters retryAttempts and
lastRetryAttempts are process-global and must be moved to per-instance,
thread-safe state to avoid cross-client corruption; update
CoordinatorSocketStateService to make these fields instance properties (e.g.,
private var retryAttempts / lastRetryAttempts on the class) or use
confined/thread-safe primitives (AtomicInteger or coroutine-confined vars) and
replace all companion references with the instance properties, then adjust
CoordinatorAnalytics interactions to accept and record per-instance retry
snapshots (pass the instance counters or a Snapshot data class) so telemetry is
derived from the specific CoordinatorSocketStateService rather than a shared
companion object.
---
Outside diff comments:
In
`@stream-video-android-core/src/main/kotlin/io/getstream/video/android/core/Call.kt`:
- Around line 693-785: _join currently calls session.value?.connectInternal()
without passing joinAnalyticsModel, so SFU websocket connect attempts don't
receive the analytics model and retry events show retryCount=0; update the call
in _join to pass joinAnalyticsModel into connectInternal (i.e.,
session.value?.connectInternal(joinAnalyticsModel)), and if
RtcSession.connectInternal does not accept that parameter, add a parameter to
RtcSession.connectInternal (and any other callers) to accept JoinAnalyticsModel
and thread it through to the SFU websocket connection logic so initial join
retries emit correct analytics.
In
`@stream-video-android-core/src/main/kotlin/io/getstream/video/android/core/call/RtcSession.kt`:
- Around line 691-706: Replace the hardcoded retryCount = 0 in the
SfuSocketState.Connected success handler with the actual SFU websocket retry
counter used by the session (the same counter incremented during rejoin/migrate
attempts), e.g. use the session's sfuReconnectAttempts or sfuRejoinAttemptCount
variable instead of 0; update both places where onSfuWsCompleted(success = true,
retryCount = 0) is called (the SfuSocketState.Connected block and the similar
block around lines 887-911) and ensure the chosen counter is reset or read
consistently when emitting the success event.
---
Nitpick comments:
In
`@stream-video-android-core/src/main/kotlin/io/getstream/video/android/core/analytics/call/observer/PeerConnectionAnalyticsStateHolder.kt`:
- Around line 73-76: The PeerConnectionAnalyticsState data class currently uses
mutable vars (peerConnectionObserverJob, publisherJob, subscriberJob) which
permits in-place mutation of state.value and breaks StateFlow snapshot
semantics; change those properties to val so the class is immutable, and update
call sites to modify the state via PeerConnectionAnalyticsState.copy(...) inside
MutableStateFlow.update { current -> current.copy(...) } (or assign a new
instance) so all changes emit new StateFlow values and preserve snapshot
semantics.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: Repository UI
Review profile: CHILL
Plan: Pro
Run ID: e0a542a1-4360-4ebc-b3c2-d3730848a964
⛔ Files ignored due to path filters (1)
stream-video-android-core/src/main/kotlin/io/getstream/android/video/generated/models/ClientEvent.ktis excluded by!**/generated/**
📒 Files selected for processing (36)
stream-video-android-core/api/stream-video-android-core.apistream-video-android-core/src/main/kotlin/io/getstream/video/android/core/Call.ktstream-video-android-core/src/main/kotlin/io/getstream/video/android/core/ClientState.ktstream-video-android-core/src/main/kotlin/io/getstream/video/android/core/StreamVideoClient.ktstream-video-android-core/src/main/kotlin/io/getstream/video/android/core/analytics/call/CallAnalytics.ktstream-video-android-core/src/main/kotlin/io/getstream/video/android/core/analytics/call/observer/AudioAnalytics.ktstream-video-android-core/src/main/kotlin/io/getstream/video/android/core/analytics/call/observer/JoinAnalytics.ktstream-video-android-core/src/main/kotlin/io/getstream/video/android/core/analytics/call/observer/JoinAnalyticsStateHolder.ktstream-video-android-core/src/main/kotlin/io/getstream/video/android/core/analytics/call/observer/MediaPermissionObserver.ktstream-video-android-core/src/main/kotlin/io/getstream/video/android/core/analytics/call/observer/PeerConnectionAnalytics.ktstream-video-android-core/src/main/kotlin/io/getstream/video/android/core/analytics/call/observer/PeerConnectionAnalyticsStateHolder.ktstream-video-android-core/src/main/kotlin/io/getstream/video/android/core/analytics/call/observer/SfuAnalytics.ktstream-video-android-core/src/main/kotlin/io/getstream/video/android/core/analytics/call/observer/SfuAnalyticsStateHolder.ktstream-video-android-core/src/main/kotlin/io/getstream/video/android/core/analytics/call/observer/VideoAnalytics.ktstream-video-android-core/src/main/kotlin/io/getstream/video/android/core/analytics/call/observer/model/JoinAnalyticsModel.ktstream-video-android-core/src/main/kotlin/io/getstream/video/android/core/analytics/call/observer/model/Stage.ktstream-video-android-core/src/main/kotlin/io/getstream/video/android/core/analytics/coordinator/CoordinatorAnalytics.ktstream-video-android-core/src/main/kotlin/io/getstream/video/android/core/analytics/reporting/ClientEventFactory.ktstream-video-android-core/src/main/kotlin/io/getstream/video/android/core/analytics/reporting/ClientEventReporter.ktstream-video-android-core/src/main/kotlin/io/getstream/video/android/core/analytics/reporting/datasource/FileBasedPendingEventDataSource.ktstream-video-android-core/src/main/kotlin/io/getstream/video/android/core/analytics/reporting/datasource/PendingEventDataSource.ktstream-video-android-core/src/main/kotlin/io/getstream/video/android/core/analytics/reporting/datasource/SynchronizedPendingEventDataSource.ktstream-video-android-core/src/main/kotlin/io/getstream/video/android/core/analytics/reporting/dispatcher/EventDispatcher.ktstream-video-android-core/src/main/kotlin/io/getstream/video/android/core/analytics/reporting/dispatcher/ImmediateEventDispatcher.ktstream-video-android-core/src/main/kotlin/io/getstream/video/android/core/analytics/reporting/model/AnalyticsCallAbortReason.ktstream-video-android-core/src/main/kotlin/io/getstream/video/android/core/analytics/reporting/model/AnalyticsFailureCodes.ktstream-video-android-core/src/main/kotlin/io/getstream/video/android/core/analytics/reporting/model/EventOutcome.ktstream-video-android-core/src/main/kotlin/io/getstream/video/android/core/analytics/reporting/model/EventStage.ktstream-video-android-core/src/main/kotlin/io/getstream/video/android/core/analytics/reporting/model/EventType.ktstream-video-android-core/src/main/kotlin/io/getstream/video/android/core/analytics/reporting/model/InFlightSession.ktstream-video-android-core/src/main/kotlin/io/getstream/video/android/core/analytics/reporting/model/PeerConnectionRole.ktstream-video-android-core/src/main/kotlin/io/getstream/video/android/core/call/RtcSession.ktstream-video-android-core/src/main/kotlin/io/getstream/video/android/core/internal/module/SfuConnectionModule.ktstream-video-android-core/src/main/kotlin/io/getstream/video/android/core/socket/coordinator/CoordinatorSocketStateService.ktstream-video-android-core/src/main/kotlin/io/getstream/video/android/core/socket/sfu/SfuSocket.ktstream-video-android-core/src/main/kotlin/io/getstream/video/android/core/socket/sfu/SfuSocketConnection.kt
rahul-lohra
force-pushed
the
feature/rahullohra/error-reporting-impl
branch
from
June 11, 2026 06:01
759d588 to
e1462ec
Compare
Contributor
|
Two |
PratimMallick
approved these changes
Jun 12, 2026
Base automatically changed from
feature/rahullohra/error-reporting-introduce-call-leave-reason
to
develop
June 12, 2026 13:23
rahul-lohra
changed the title
fix: remove persistance and audio analytics
rahul-lohra
force-pushed
the
feature/rahullohra/error-reporting-impl
branch
from
June 12, 2026 14:50
dac2d12 to
0ae3774
Compare
aleksandar-apostolov
approved these changes
Jun 12, 2026
|
Collaborator
|
🚀 Available in v1.27.0 |
This was referenced Jun 17, 2026
24 tasks
This was referenced Jul 10, 2026
27 tasks
PratimMallick
added a commit
that referenced
this pull request
Aug 10, 2026
* Upgrade to `m145` webrtc and noise-cancellation lib `v3.0.0` (#1678) * upgrade to m145 webrtc and noise-cancellation * Update to webrtc m145 and corresponding noiseCancellation lib * Remove the snapshot repo resolution * [skip ci] Update SDK sizes * AUTOMATION: Version Bump * Honor SFU-provided degradation preference for adaptive video publishing (#1699) * feat(core): pick up SFU DegradationPreference and add WebRTC mapper Regenerate SFU protos to pick up the new DegradationPreference enum and its degradation_preference fields on PublishOption and VideoSender (plus TrackInfo.self_sub_audio_video), and refresh the public API dump. Add toRtcDegradationPreference() converting the SFU enum to org.webrtc.RtpParameters.DegradationPreference, returning null for UNSPECIFIED so callers can keep the current value. Includes unit tests covering every enum variant. Co-authored-by: Cursor <cursoragent@cursor.com> * publisher changes for applying degradation preferences * Remove duplicate handling of ChangePublishQualityEvent from callState. This event directly gets handled by RtcSession handleEvent method * Added test for the two call sites where degradation Preference is getting set to test for the case where sfu sends the same degrdation preference which is already set in the transcevier sender param --------- Co-authored-by: Cursor <cursoragent@cursor.com> * fix(core): initialize Call.events before CallState to avoid NPE in sorted-participants init (#1701) * AUTOMATION: Version Bump * [skip ci] Update SDK sizes * Add ability to add custom user in demo-app (#1700) * demo: Add logic to add custom user * chore(demo-app): gate add user dialog to development flavor Hide the new add-user button and popup outside the development flavor so the production demo app doesn't expose internal user injection. --------- Co-authored-by: Aleksandar Apostolov <apostolov.alexandar@gmail.com> * fix(core): send only the changed track in UpdateMuteStates (#1706) When toggling a single track (e.g. muting the mic while the camera stays on, or turning the camera off while the mic stays on), RtcSession sent the full declarative mute-state map for all track types. Re-asserting an unchanged track as un-muted made the SFU re-emit a redundant TrackPublishedEvent for that track. That event carries a potentially stale published_tracks snapshot which re-enabled the just-disabled track, freezing the local self-view on the last frame / showing the avatar while no frames arrive. Send only the mute state of the track that actually changed, matching the web SDK. On SFU (re)connect/migration each enabled track is re-signalled individually via listenToMediaChanges, so the full state is still restored. Co-authored-by: Cursor <cursoragent@cursor.com> * fix(core): clear stored push device even when deleteDevice API fails (#1705) deleteDevice previously only purged the cached Device on API success. When DELETE /devices returned 404 or the network failed, the stale token sat in deviceTokenStorage. The next createDevice for a different user short-circuited on token equality (when autoRegisterPushDevice=true) and returned Success without calling POST /devices, silently leaving the new user without a device row server-side and breaking incoming-call push. Local cleanup now runs unconditionally and is guarded so a storage failure doesn't mask the API outcome. CancellationException is re-thrown to preserve structured concurrency. Three regression tests added. AND-1214 * [skip ci] Update SDK sizes * fix(core): make apiCall wait for guestUserJob to avoid anonymous push device registration (#1703) Guest user setup runs asynchronously: StreamVideoBuilder.build returns immediately while setupGuestUser kicks off a background createGuest call to fetch the JWT. Any authenticated request that fires in that window goes out with stream-auth-type "anonymous" and no Authorization header, so the backend silently registers it against the wrong identity. The customer-visible effect is push device registration succeeding under !anon and incoming-call pushes never reaching the guest user. apiCall now awaits guestUserJob before invoking the request block, with a self-job guard so createGuestUser — which also goes through apiCall — does not await its own enclosing job and deadlock. Adds two regression tests: one for the wait, one for the deadlock guard. AND-1202 Co-authored-by: Rahul Kumar Lohra <tgunix@gmail.com> * [skip ci] Update SDK sizes * Adopt createGuest's server-issued identity for guest users (#1704) * fix(core): adopt response.user from createGuest to keep guest identity in sync The createGuest endpoint returns the server-resolved user (which may differ from what was passed in — e.g. normalized id). The SDK previously kept only the access token and left its in-memory user as the builder's input, so the WS auth payload and the JWT user_id claim could disagree. setupGuestUser now also updates client.user from response.user (matching the JS SDK's connectUser(response.user, response.access_token) semantics). userId becomes a computed property so every existing reader of client.userId picks up the new identity automatically. CoordinatorSocketConnection.user turns into a var so its onCreated() auth payload reads the latest user. Adds three regression tests: userId reactivity, the var update inside the socket connection's connect path, and a full setupGuestUser flow with the api mocked to return a different user id than the input. AND-1202 * fix(core): mirror adopted guest user into ClientState.user ClientState._user was snapshotted from the integrator-supplied user at construction, so observers of state.user kept the old id after setupGuestUser adopted the server-issued one. Propagate the adopted user via a new internal ClientState.setUser. * [skip ci] Update SDK sizes * Introduce UserRepository as single source of truth for SDK user (#1708) * refactor(core): introduce UserRepository as single source of truth for SDK user Replaces the parallel `var user` fields in `StreamVideoClient` and `CoordinatorSocketConnection` (and the `_user` mirror in `ClientState`) with a single `UserRepository`: - `UserRepository` — public, read-only access via `user` / `userFlow`. - `WritableUserRepository` — internal sub-interface with `setUser`. Only `StreamVideoClient` holds a write reference, so identity updates go through one path. - `StreamUserRepositoryImpl` — in-memory impl backed by a `MutableStateFlow`. `StreamVideoBuilder` constructs one instance and shares it between the client (writer) and the coordinator socket / `ClientState` (readers). `setupGuestUser` writes the adopted user to the repo once; readers pick it up automatically without any local copy to keep in sync. `connect()`/`reconnect()` no longer mutate a snapshot of the user on the socket — they only forward the call to `internalSocket`, and `onCreated()` reads from the repository when building the WS auth payload. * test(core): add direct unit tests for StreamUserRepositoryImpl Covers seed-from-constructor, user/userFlow reads, setUser write, emission to active StateFlow collectors, and replacement semantics. Lifts coverage on the new repository from indirect-only (via StreamVideoClient tests) to full coverage of the impl. * Auto-connect and register push device for guest users (#1707) * feat(core): auto-connect and register push device for guest users StreamVideoBuilder previously only ran the auto-register-push and auto-connect block for UserType.Authenticated. Guest users fell through, forcing every Guest integrator to write the same boilerplate (manual registerPushDevice + connect after build) — boilerplate the iOS and JS SDKs don't require. Widen the gate to include UserType.Guest. registerPushDevice() and connectAsync() inside StreamVideoClient already await guestUserJob, so both are safe to fire from the builder block before /video/guest completes. Anonymous users still don't have an identity to register a device against, so they remain excluded. AND-1202 * fix(core): wait for guestUserJob before registering push device StreamNotificationManager.createDevice() goes straight to api.createDevice() without the apiCall {} wrapper, so the guestUserJob await guard added in #1703 doesn't cover it. registerPushDevice() now waits for guest setup itself before delegating, so the push generator can't fire createDevice() before the coordinator's auth headers flip from anonymous to JWT. * Include internal audio switch to fix concurrency issue (#1710) * fix: include internal audio switch to fix concurrency issue * fix: include aar * chore(core): deprecate StreamVideo.logOut (#1709) The KDoc claimed `logOut` clears internal user state, removes push notification devices, and clears call state. The actual implementation only writes null to the local DeviceTokenStorage — no `DELETE /devices`, no socket disconnect, no in-memory clear. The name and the historical doc invite a customer to ship broken user-switching: anyone reading the API surface would reasonably assume a clean slate. Surfaced while diagnosing a customer integration where push delivery silently failed across user transitions. Annotate the interface declaration and the StreamVideoClient override with `@Deprecated`. Update the KDoc to describe current behavior accurately. Point `ReplaceWith` at `StreamVideo.removeClient()`, which triggers a real `cleanup()` and uninstalls the singleton. Customers who need to remove the server-side device row should call `deleteDevice()` before `removeClient()`. No binary signature change — `@Deprecated` is annotation-only, so the public `.api` file is unchanged. AND-1217 Co-authored-by: Rahul Kumar Lohra <tgunix@gmail.com> * fix(core): prevent DevicePreferences ClassCastException in updateDevice (#1711) Assign the DataStore.updateData() result to a local in DeviceTokenStorage.updateUserDevice so the suspend function is compiled as a state machine that returns Unit. Without it the compiler tail-call-optimizes the call and propagates the DevicePreferences result up the updateDevice suspend chain, which can surface as "DevicePreferences cannot be cast to kotlin.Unit" at the caller once R8 inlines the chain. Co-authored-by: Cursor <cursoragent@cursor.com> * AUTOMATION: Version Bump * fix(core): stop coordinator events from clobbering SFU participant count (#1712) Post-join, the SFU healthcheck delivers the authoritative participant count. Coordinator session events (participant_joined/left, counts_updated, anything carrying a CallSessionResponse) carry a smaller, stale snapshot that disagrees at scale. The previous guard checked only !is RealtimeConnection.Joined — a transient state immediately replaced by Connected — so every coordinator session event re-wrote the count, producing wild swings during livestreams (e.g. 25k -> 32k -> 42k -> 28k in seconds). Broaden the guard to cover the entire in-call lifetime (Joined, Connected, Reconnecting, Migrating). Pre-join, the session-derived path now uses max(byRoleCount, participants.size) for monotonicity during fast joins, matching the stream-video-js SDK. AND-926 * Prevent "MediaSource has been disposed" crash when leaving a call (#1718) * fix(core): prevent "MediaSource has been disposed" crash on leave Guards the lazy audio/video source and track creation/disposal in MediaManagerImpl with a single reentrant lock, and adds a terminal `released` flag so the mic/camera mute paths no-op after cleanup instead of lazily resurrecting native objects. The crash occurred when a call was left while the first AudioSwitch setup was still in flight: cleanup() disposed the audio source on one thread while the deferred mic-disable callback recreated the audio track from that disposed source on stream-audio-thread. Co-authored-by: Cursor <cursoragent@cursor.com> * test(core): stub runOnAudioTrackIfAvailable in MicrophoneManager tests enable()/disable() now route the track toggle through mediaManager.runOnAudioTrackIfAvailable instead of the audioTrack getter, so the test helper stubs the new helper to invoke its block with the mock track. Fixes the 5 failing MicrophoneManagerTest verifications. Co-authored-by: Cursor <cursoragent@cursor.com> --------- Co-authored-by: Cursor <cursoragent@cursor.com> * [skip ci] Update SDK sizes * Revert audio switch local AAR (#1710) and reset version to allow re-release of 1.26.0 (#1719) * Revert " Include internal audio switch to fix concurrency issue (#1710)" This reverts commit 9a8da36. * chore(release): reset version to 1.25.0 to allow re-release of 1.26.0 The 1.26.0 Maven publish failed due to the local AAR introduced in #1710. Resetting gradle.properties to 1.25.0 so the release workflow can re-tag and publish 1.26.0 cleanly from the reverted develop state. * AUTOMATION: Version Bump * fix(core): set small icon on setting-up-call notification (#1720) The "setting up call" foreground-service notification was built without a small icon for any non-incoming trigger (outgoing/ongoing/livestream). The non-deprecated getSettingUpCallNotification(trigger, callId) delegated its else branch to the deprecated no-arg overload, which never called setSmallIcon. A small icon is mandatory for foreground-service notifications, so Android 13+ rejected it with CannotPostForegroundServiceNotificationException ("Bad notification for startForeground") when the call foreground service started. Extract a non-deprecated buildSettingUpCallNotification() helper that always sets setSmallIcon(R.drawable.stream_video_ic_call), and have both the non-deprecated else branch and the deprecated overload delegate to it. Add a regression test covering the non-incoming (outgoing) trigger path. Co-authored-by: Cursor <cursoragent@cursor.com> * Update open api generated models in Video SDK(1/4) (#1713) * update open api generated models * update code gen script * Fix self cancelling coroutine code (2/4) (#1714) * update open api generated models * update code gen script * fix: self cancelling coroutine code * fix: fix self cancelling coroutine code * Send standardized call leave reason instead of string when leaving the call (3/4) (#1715) * update open api generated models * update code gen script * fix: self cancelling coroutine code * internal: add call leave reason * internal: update Call Leave Reason LLC * fix: fix unit tests * 📝 CodeRabbit Chat: Implement requested code changes * Update stream-video-android-core/src/main/kotlin/io/getstream/video/android/core/CallLeaveReason.kt * chore: send correct leave reason from StreamCallActivity.kt --------- Co-authored-by: coderabbitai[bot] <136622811+coderabbitai[bot]@users.noreply.github.com> Co-authored-by: Aleksandar Apostolov <apostolov.alexandar@gmail.com> * [skip ci] Update SDK sizes * Add Video Analytics in call-join flow (#1716) * update open api generated models * internal: add call leave reason * internal: Add analytics * internal: renaming * internal: spotless * chore: fix tests, remove logs * chore: fix retry exception * chore: fix binary compatibility * chore: add capping to analytics file * chore: use isolate coroutine scope to interact with data source * chore: renaming * chore: renaming * chore: fix paparazzi tests * chore: write analytics tests * chore: fix tests * chore: Enable Audio Analytics * chore: update audio analytics unit-test * chore: update PeerConnectionAnalyticsState * chore: add temp commit * fix: fix incorrect events * fix: fix incorrect events in pc and video * fix: fix pc publish analytics * fix: remove persistance and audio analytics, update tests fix: remove persistance and audio analytics * fix: fix error code and reason * fix: fix tests * fix: remove comments * fix: spotless * fix: Correctly send sfu ws terminal * fix: spotless * fix: spotless * fix: fix failed permission * fix: fix failed permission * fix: add missing class * fix: run only if user id is not null * fix: fix tests * AUTOMATION: Version Bump * [skip ci] Update SDK sizes * Do not show error message while video is loading (#1721) * fix(compose): show loading spinner in CallLobby while camera warms up Replace the misleading track-failure fallback with a neutral loading placeholder and overlay it in CallLobby until the first video frame renders. Removes the redundant demo-app workaround. Fixes #1656 Co-authored-by: Cursor <cursoragent@cursor.com> * use DefaultBadNetworkFallbackContent with a CircularProgressBar instead of Text --------- Co-authored-by: Cursor <cursoragent@cursor.com> * feat(ui-compose): add participantLabelContent slot to CallLobby (#1726) The previous CallLobby rendered the lobby participant label unconditionally with no way to hide or override it. iOS and React do not render a participant label on the local preview, so Android was the outlier and customers had no escape hatch. Add a participantLabelContent slot (matches React VideoPreview's slot-based overrides). Pass `{}` to hide the label or override to supply custom content and positioning. The previous overload (labelPosition: Alignment) is deprecated and delegates to the new one. * [skip ci] Update SDK sizes * Chore(deps): Bump actions/checkout from 6 to 7 (#1728) Bumps [actions/checkout](https://github.com/actions/checkout) from 6 to 7. - [Release notes](https://github.com/actions/checkout/releases) - [Changelog](https://github.com/actions/checkout/blob/main/CHANGELOG.md) - [Commits](actions/checkout@v6...v7) --- updated-dependencies: - dependency-name: actions/checkout dependency-version: '7' dependency-type: direct:production update-type: version-update:semver-major ... Signed-off-by: dependabot[bot] <support@github.com> Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com> * Move @Preview composables out of published artifact to fix R8 builds (#1730) * fix(compose): move @Preview composables out of published artifact to fix R8 builds Compose-generated ComposableSingletons classes referenced the compileOnly stream-video-android-previewdata module (StreamPreviewDataUtils, previewCall, etc.) from production code, breaking consumers' minified (R8) release builds with "Missing class" errors. - Move all @Preview composables from src/main into a new debug-only source set (one <Name>Preview.kt per file), so preview-data is never referenced from the release artifact. - Change previewdata dependency from compileOnly to debugImplementation. - Declare kotlinx-datetime explicitly (used by the livestream countdown UI) so it no longer leaks in transitively, fixing the related kotlinx.datetime R8 missing-class errors. Fixes #1448 Co-authored-by: Cursor <cursoragent@cursor.com> * api dump --------- Co-authored-by: Cursor <cursoragent@cursor.com> * [skip ci] Update SDK sizes * Fix `runCallServiceInForeground` handling for incoming PN and non-ringing call flows (#1729) * temp: force-commit * fix: invoke updateRingingState() when setting active call * fix: add canRunService on incoming calls * demo-app: add screen to configure call settings * fix: refactor * fix: refactor * fix: remove unused classes * fix: fix unit-test * Improve video analytics by correctly sending ice state (#1724) * improve: fix sending correct ice state * improve: correctly update analytics stage * improve: refactor * fix: remove grace period to wait for ice state * fix: refactor * fix: refactor --------- Co-authored-by: Aleksandar Apostolov <apostolov.alexandar@gmail.com> * Code refactor for Video Analytics (#1727) * improvement: video analytics refactor * fix: fix unit test * Fix a crash when application put to background restriction (#1731) * temp: force-commit * fix: refactor * fix: remove unused classes * fix: Fix a crash when application put to background restriction by rendering normal notification instead of CallStyle notification * fix: add safe null-check * fix: moved file to utils * [skip ci] Update SDK sizes * Fix notification dismissal for Leave/Hang Up notification actions (#1732) * temp: force-commit * fix: invoke updateRingingState() when setting active call * fix: add canRunService on incoming calls * demo-app: add screen to configure call settings * fix: refactor * fix: refactor * fix: remove unused classes * fix: Fix a crash when application put to background restriction by rendering normal notification instead of CallStyle notification * fix: add safe null-check * fix: fix unit-test * fix: fix notification cleaning from leave button in notification * fix: add todos for future fixes * feat(ui-compose): add videoPreviewModifier slot to CallLobby (#1733) The lobby video preview Box was hardcoded to a responsive height (180/280/200dp by screen size and orientation), fillMaxWidth, and a 12dp rounded corner clip with no way to override size or shape. iOS sizes the preview via the parent's GeometryReader; React drives size via className on VideoPreview. Android was the outlier. Add a videoPreviewModifier parameter so callers can override the box modifier directly. The default preserves the previous responsive height, full width, and 12dp clip, so existing callers see no change. * AUTOMATION: Version Bump * fix: fix tracer timestamp and remove additional lock (#1734) * Report SFU WebSocket connection and join timeouts accurately (#1740) * fix(core): report SFU WebSocket connection/join timeouts accurately Splits the SFU socket connect into a transport-open phase (bounded by OkHttp's connect timeout) and a distinct join-response phase (bounded by a dedicated timer via a new WebSocketConnected state), so a silent SFU no longer hangs the join. Surfaces real transport failure messages and HTTP status codes on the resulting NetworkError, routes all recoverable errors through a single DisconnectedTemporarily state carrying the exact code/reason, and maps both timeout flavours to REQUEST_TIMEOUT in analytics (join-response via error code, transport via SocketTimeoutException cause). Also moves join-error analytics to the join flow only, adds a per-session SFU WS retry counter, and wires connectionTimeoutInMs from the builder to both the OkHttp client and the join-response deadline. Co-authored-by: Cursor <cursoragent@cursor.com> * Remove !! operator on session.value and send a proper Failure when session.value was cleared to null while connecting to sfu was still in progress * Update comment * fix(core): address PR review comments for SFU WS timeout handling - Rename SfuSocketStateEvent.WebSocketConnected to WebSocketEstablished to avoid the naming collision with the SfuSocketState.WebSocketConnected state. - Guard monitorSession() in _join recovery so it only runs when the original session is reused, preventing double-registration after rejoin/migrate. - Default connectionTimeoutInMs to 5s and fix stale KDoc. - Rename triggersReconnect() to canTriggersReconnect() for clarity. Co-authored-by: Cursor <cursoragent@cursor.com> * fix(core): classify OkHttp InterruptedIOException timeouts as REQUEST_TIMEOUT OkHttp connect/call timeouts surface as InterruptedIOException("timeout"), not SocketTimeoutException, so SFU join analytics were incorrectly reporting SFU_ERROR despite the failure reason being "timeout". Co-authored-by: Cursor <cursoragent@cursor.com> * fix(core): add safety timeout to connectInternal SFU socket wait Wrap the wait for a terminal SfuSocketState in withTimeoutOrNull so connectInternal always returns even if the socket state machine never reaches Connected or Disconnected. On timeout, return a recoverable Failure with REQUEST_TIMEOUT so higher-level reconnect logic can escalate. Co-authored-by: Cursor <cursoragent@cursor.com> --------- Co-authored-by: Cursor <cursoragent@cursor.com> Co-authored-by: Rahul Kumar Lohra <tgunix@gmail.com> * [skip ci] Update SDK sizes * CI: add least-privilege permissions to GitHub Actions workflows (#1742) * CI: add least-privilege permissions to GitHub Actions workflows Add explicit workflow-level permissions blocks to resolve CodeQL actions/missing-workflow-permissions alerts. Scopes are derived from workflow operations (git push, artifact upload, Danger, release lanes). Refs: APPSEC-164 * CI: add least-privilege permissions to GitHub Actions workflows Add explicit workflow-level permissions blocks to resolve CodeQL actions/missing-workflow-permissions alerts. Scopes are derived from workflow operations (git push, artifact upload, Danger, release lanes). Refs: APPSEC-164 * CI: add least-privilege permissions to GitHub Actions workflows Add explicit workflow-level permissions blocks to resolve CodeQL actions/missing-workflow-permissions alerts. Scopes are derived from workflow operations (git push, artifact upload, Danger, release lanes). Refs: APPSEC-164 * CI: add least-privilege permissions to GitHub Actions workflows Add explicit workflow-level permissions blocks to resolve CodeQL actions/missing-workflow-permissions alerts. Scopes are derived from workflow operations (git push, artifact upload, Danger, release lanes). Refs: APPSEC-164 * CI: add least-privilege permissions to GitHub Actions workflows Add explicit workflow-level permissions blocks to resolve CodeQL actions/missing-workflow-permissions alerts. Scopes are derived from workflow operations (git push, artifact upload, Danger, release lanes). Refs: APPSEC-164 * CI: add least-privilege permissions to GitHub Actions workflows Add explicit workflow-level permissions blocks to resolve CodeQL actions/missing-workflow-permissions alerts. Scopes are derived from workflow operations (git push, artifact upload, Danger, release lanes). Refs: APPSEC-164 * CI: add least-privilege permissions to GitHub Actions workflows Add explicit workflow-level permissions blocks to resolve CodeQL actions/missing-workflow-permissions alerts. Scopes are derived from workflow operations (git push, artifact upload, Danger, release lanes). Refs: APPSEC-164 * CI: add least-privilege permissions to GitHub Actions workflows Add explicit workflow-level permissions blocks to resolve CodeQL actions/missing-workflow-permissions alerts. Scopes are derived from workflow operations (git push, artifact upload, Danger, release lanes). Refs: APPSEC-164 * CI: add least-privilege permissions to GitHub Actions workflows Add explicit workflow-level permissions blocks to resolve CodeQL actions/missing-workflow-permissions alerts. Scopes are derived from workflow operations (git push, artifact upload, Danger, release lanes). Refs: APPSEC-164 * ci: move workflow permissions to job level (SonarCloud) * ci: move workflow permissions to job level (SonarCloud) * ci: move workflow permissions to job level (SonarCloud) * ci: move workflow permissions to job level (SonarCloud) * ci: move workflow permissions to job level (SonarCloud) * ci: move workflow permissions to job level (SonarCloud) * ci: move workflow permissions to job level (SonarCloud) * ci: move workflow permissions to job level (SonarCloud) * ci: move workflow permissions to job level (SonarCloud) * ci: move workflow permissions to job level (SonarCloud) * ci: move workflow permissions to job level (SonarCloud) * ci: move workflow permissions to job level (SonarCloud) * ci: move workflow permissions to job level (SonarCloud) * ci: move workflow permissions to job level (SonarCloud) * ci: move workflow permissions to job level (SonarCloud) * ci: move workflow permissions to job level (SonarCloud) * ci: move workflow permissions to job level (SonarCloud) * ci: move workflow permissions to job level (SonarCloud) * ci: repair workflow permissions blocks for actionlint/SonarCloud * ci: repair workflow permissions blocks for actionlint/SonarCloud * ci: repair workflow permissions blocks for actionlint/SonarCloud * ci: repair workflow permissions blocks for actionlint/SonarCloud * ci: repair workflow permissions blocks for actionlint/SonarCloud * ci: repair workflow permissions blocks for actionlint/SonarCloud * ci: repair workflow permissions blocks for actionlint/SonarCloud * ci: repair workflow permissions blocks for actionlint/SonarCloud * ci: repair workflow permissions blocks for actionlint/SonarCloud * ci: address CodeRabbit review feedback and fix workflow YAML * ci: repair workflow permissions blocks for actionlint/SonarCloud * ci: repair workflow permissions blocks for actionlint/SonarCloud * ci: repair workflow permissions blocks for actionlint/SonarCloud * ci: repair workflow permissions blocks for actionlint/SonarCloud * ci: repair workflow permissions blocks for actionlint/SonarCloud * ci: repair workflow permissions blocks for actionlint/SonarCloud * ci: repair workflow permissions blocks for actionlint/SonarCloud * ci: repair workflow permissions blocks for actionlint/SonarCloud * ci: repair workflow permissions blocks for actionlint/SonarCloud * ci: repair workflow permissions blocks for actionlint/SonarCloud * ci: repair workflow permissions blocks for actionlint/SonarCloud * ci: repair workflow permissions blocks for actionlint/SonarCloud * ci: repair workflow permissions blocks for actionlint/SonarCloud * ci: repair workflow permissions blocks for actionlint/SonarCloud * ci: repair workflow permissions blocks for actionlint/SonarCloud * ci: repair workflow permissions blocks for actionlint/SonarCloud * ci: repair workflow permissions blocks for actionlint/SonarCloud * ci: repair workflow permissions blocks for actionlint/SonarCloud * e2e: Port the chat test harness improvements: retry visibility, video on retries only (#1746) - RetryRule: each failed attempt that is retried is written as its own Allure result sharing the real test's historyId, with that attempt's steps, error and artifacts, so TestOps groups attempts as retries and can flag flaky tests - RetryRule: screen recording runs only on retry attempts; previously every test recorded an 8 Mbit/s video that was discarded on pass - RetryRule: the recording file uses methodName instead of displayName (the display name's parentheses break when the recording commands go through the device shell), and a recording stop failure no longer replaces the test result - RetryRule: move to io.getstream.video.android.rules (the file declared a chat package), remove the unused Retry annotation, simplify DatabaseOperations - Wait: waitForText and waitForCount poll at 50ms instead of spinning the CPU until timeout; remove the unused waitForTextToChange - Allurefile: name launches 'Cron checks' only for real scheduled events, so manual dispatches stay distinguishable in TestOps - Fastfile: batch_tests splits round-robin, always yielding exactly batch_count groups; the previous rounded-up slicing could produce fewer groups and crash on nil for the last batch Ported from GetStream/stream-chat-android#6568. * Recover initial join when connect safety-timeout leaves no reconnect (#1741) * fix(core): recover initial join when connect safety-timeout leaves no reconnect When RtcSession.connectInternal hits its own safety-timeout, the socket is left in a non-terminal state that stateJob ignores, so no Call.reconnect is ever launched and _join's didReconnectSucceed() would block forever. - Add SfuConnectionResult.Failure.reconnectTriggered so the join flow can tell whether a recovery loop has already been started by stateJob. - On the safety-timeout path, tear down the abandoned (still-in-flight) socket so a late Connected/JoinResponse can't resurface and resurrect the session. - In Call._join, trigger a REJOIN ourselves when a recoverable failure reports no reconnect was triggered; otherwise await the existing loop. - Rename canTriggersReconnect() to triggersStateJobReconnect() for clarity. - demo-app: align connectionTimeoutInMs with the SDK default (5s). Co-authored-by: Cursor <cursoragent@cursor.com> * refactor(core): type SFU connect failures instead of a reconnect flag Address PR review: reconnect-orchestration state should not live on the SfuConnectionResult.Failure DTO. Replace the boolean flag with a typed error so callers of connectInternal branch on the failure kind. - Add sealed SfuConnectException with Timeout (connect safety-timeout tore down a stuck socket; no recovery started) and Disconnected (terminal SFU state). - connectInternal now returns these typed errors instead of Exception("msg"). - Drop SfuConnectionResult.Failure.hasReconnectStarted; _join self-triggers a REJOIN only when the error is SfuConnectException.Timeout, else awaits the loop stateJob already started. - Update tests to construct/assert the typed errors. Co-authored-by: Cursor <cursoragent@cursor.com> * Replace SfuConnectException with SfuConnectFailureCause code (#1744) * fix: Refactor with SfuConnectFailureCause code * fix: Add kdoc * fix: Replace Singletone shared Call.testInstanceProvider.rtcSessionCreator with Call..unitTestRtcSessionFactory --------- Co-authored-by: Cursor <cursoragent@cursor.com> Co-authored-by: Aleksandar Apostolov <apostolov.alexandar@gmail.com> Co-authored-by: Rahul Kumar Lohra <tgunix@gmail.com> * AUTOMATION: Version Bump * test(e2e): harden recording and outgoing-call waits against backend latency (#1751) * test(e2e): harden recording and outgoing-call waits against backend latency The composite recorder can take 20-30s to start and emit call.recording_started, and the outgoing ringing screen only renders after the call create/ring network round-trip. The 5s default (recording icon) and 10s (outgoing decline button) waits time out before the UI appears, causing deterministic findObject NPEs in testParticipantRecordsCall and testUserRejectsTheOutgoingAudioCall. Co-authored-by: Cursor <cursoragent@cursor.com> * test(e2e): extend recording-consent and stop waits to 30s The recording-consent dialog (acceptCallRecording/declineCallRecording) and the recording-icon disappear assert are all gated on backend recorder events (call.recording_started / call.recording_stopped), which can take 20-30s. The 10s/5s waits time out first, so testParticipantRecordsCall still failed at acceptCallRecording after the initial icon-wait bump. Extend these waits to 30s. Co-authored-by: Cursor <cursoragent@cursor.com> * test(e2e): record for 60s so the recording icon is observable Composite recording spins up ~10-15s after the request, so recording for only 15s left a ~5s client-visible window that closed before the 3-view assertion loop could observe the icon. Record for 60s and keep the buddy in the call for 120s (matching the ReconnectionTests pattern) so the icon is reliably visible, then extend the "recording disappeared" wait to 70s to cover the longer run. Co-authored-by: Cursor <cursoragent@cursor.com> --------- Co-authored-by: Cursor <cursoragent@cursor.com> * chore: pass OpenAPI generator config via --opt flags (#1743) * chore: pass OpenAPI generator config via --opt flags The chat-side Kotlin generator now takes its configuration through the generic `--opt key=value` flag (via the Configurable interface) instead of bespoke named flags. Update the generation script accordingly: - Convert the generate-client invocation to `--opt key=value` form. - Drop `--model-dir` (the generator hardcodes the models directory; the flag was a no-op) and its now-unused MODEL_DIR variable / argument. - classes-to-skip is space-separated (the --opt slice flag splits comma lists). - Point REFERENCE_VALUE at `master`, since the generator changes land there. Generated output is unchanged (byte-identical vs the previous generator). * fix: use the renamed android-sdk generator option The backend generator's SDK option was renamed androidSdk -> android-sdk during review before landing on master. Update the invocation to match, otherwise generate-client fails with `kotlin: unknown option "androidSdk"`. * chore: drop retired --model-dir from the openapi generator gradle task generate_openapi_v2.sh no longer accepts --model-dir (the backend generator hardcodes the models directory and never honored the flag), so the Gradle caller passing it would fail with "Unknown argument". Remove the argument and the now-unused modelsDir property. * fix: default the openapi generator to clone master The Gradle task's default refValue still pointed at the merged (now-deleted) feature branch, so `generateOpenApiClient` would fail at `git clone --branch`. Default to master, matching the script's own default and the intended source. * Chore(deps): Bump actions/checkout from 7.0.0 to 7.0.1 (#1748) Bumps [actions/checkout](https://github.com/actions/checkout) from 7.0.0 to 7.0.1. - [Release notes](https://github.com/actions/checkout/releases) - [Changelog](https://github.com/actions/checkout/blob/main/CHANGELOG.md) - [Commits](actions/checkout@9c091bb...3d3c42e) --- updated-dependencies: - dependency-name: actions/checkout dependency-version: 7.0.1 dependency-type: direct:production update-type: version-update:semver-patch ... Signed-off-by: dependabot[bot] <support@github.com> Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com> Co-authored-by: Gianmarco <47775302+gpunto@users.noreply.github.com> * fix(core): stop orphaned publisher transceivers to prevent SFU rejoin loop (#1757) Three transceiver-lifecycle bugs in Publisher could strand live sendonly m-lines that SetPublisher.tracks never announces, causing the SFU to force-rejoin the publisher (VID-1376): - syncPublishOptions: the cleanup loop compared the cache against itself (always true), so every transceiver was torn down on each ChangePublishOptions event. Now keep transceivers whose option is still requested by the SFU. - publishStreamInternal fallback: replaced the old transceiver without stopping it, leaving an orphaned m-line. Now stop() the old one first. - addTransceiver: silently overwrote the cache entry for the same [track_type, publish_option_id], stranding the old transceiver. Now stop() any pre-existing transceiver before adding. All paths use stop() only (never dispose()) on a live PeerConnection; the PC owns the native transceiver/sender and frees it safely at teardown. Disposing mid-session is a use-after-free on network_thread (SIGSEGV). Adds unit tests covering all three cases. Co-authored-by: Cursor <cursoragent@cursor.com> * fix(core): resolve develop merge compile/runtime conflicts for v2 Keep UserRepository + StreamConnectionState + StreamClient as v2 SSOT while adopting develop analytics/degradation-preference changes under io.getstream.webrtc. Co-authored-by: Cursor <cursoragent@cursor.com> * chore(ui-compose): refresh API dump after develop merge Co-authored-by: Cursor <cursoragent@cursor.com> * Decompose Call into focused internal components (#1747) * refactor(core): decompose Call into focused internal components Break the ~2,260-line Call class into 11 internal collaborators under call/components (CallApiClient, CallStatsReporter, CallRenderer, CallEventManager, CallMediaManager, CallSessionManager, CallIceConnectionMonitor, CallConnectivityMonitor, CallJoinCoordinator, CallReconnector, CallLifecycleManager). Call remains a thin, binary-compatible public facade that delegates to them; public API is unchanged (apiCheck passes). Update white-box reflection tests to target the new component owners after internals moved out of Call. Co-authored-by: Cursor <cursoragent@cursor.com> * fix(core): remove duplicated join preflight from Call facade Call.join() ran join analytics, permission checks, and the guest-token wait before delegating to CallJoinCoordinator.join(), which performed the exact same preflight — so every join() executed it twice. Make the facade a pure delegation so the preflight runs only once in the coordinator. Co-authored-by: Cursor <cursoragent@cursor.com> * test(core): add unit tests for Call decomposition components Add JVM unit tests for the extracted Call collaborators (CallApiClient, CallEventManager, CallSessionManager, CallRenderer, CallMediaManager) to raise coverage on the refactor's new code toward the SonarCloud gate. Co-authored-by: Cursor <cursoragent@cursor.com> * test(core): add unit tests for join/reconnect/connectivity/ice components Broaden new-code coverage for the extracted Call components: exercise the CallJoinCoordinator retry loop and join-and-ring flow (via the RtcSession test factory), the CallConnectivityMonitor reconnect/leave listener, the reachable CallReconnector state-machine branches, the CallIceConnectionMonitor restart paths, plus additional CallMediaManager (monitorHeadset, not-selected devices) and CallApiClient (ring request, ringing create) cases. Co-authored-by: Cursor <cursoragent@cursor.com> * test(core): cover reconnector rejoin/migrate and renderer audio paths Add a unitTestRtcSessionFactory seam to CallReconnector's rejoin/migrate so the session-swap, monitor and finalize paths are unit-testable, and add tests for them (success + retry-until-exhausted). Also cover CallRenderer's incoming-audio track walking for all/selected participants. Co-authored-by: Cursor <cursoragent@cursor.com> * refactor(core): decouple Call collaborators from the Call facade Give the Call decomposition components explicit dependencies so they no longer reach into the Call facade: - CallApiClient, CallConnectivityMonitor, CallEventManager, CallIceConnectionMonitor, CallRenderer, CallStatsReporter, CallSessionManager and CallMediaManager now take the granular collaborators they need (type/id/scope/state/session/clientImpl/eglBase) instead of a Call. - Isolate the unavoidable identity hand-offs behind small seams/providers: RingingCallRegistrar for CallApiClient's ring/accept client-state writes, and a lazy () -> Call provider for CallMediaManager's MediaManagerImpl (a public type that requires a Call). - Behaviour is unchanged; component unit tests now construct each collaborator directly without a Call mock. The three orchestrators (CallReconnector, CallJoinCoordinator, CallLifecycleManager) still hold Call and are left for a follow-up. Co-authored-by: Cursor <cursoragent@cursor.com> * refactor(core): finish decoupling Call collaborators and repair the suite Removes the last Call references from the extracted components so no class under call/components holds the facade any more. CallLifecycleManager now takes its collaborators directly (lazy providers, since it is constructed before CallState and the monitors exist), which made Call.stopConnectionMonitors/stopStatsReporting/cancelSfuObservers/ shutDownJobsGracefully dead; they are removed. CallMediaManager gains disableLocalCapture() so the lifecycle no longer reaches through to the device handles. The three callback interfaces into Call (CallHost, CallTeardownHost, RingingCallRegistrar) were named after who implements them rather than what they do, and six of their eight methods did the same thing: register or deregister this call in the client's ringing/active registries. Two were byte-identical. They collapse into one ClientCallRegistry; the genuine outliers (hasRequiredPermissions, shutDownJobs) become plain lambdas. Two production fixes surfaced while repairing the tests: - CallMediaManager evaluated eglBase().eglBaseContext to build an argument for MediaManagerFactory.create, forcing a real EGL context before the factory ran. Call owns both the context and the factory, so the parameter is dropped and the factory resolves it itself. - The reconnect loop reads connectivity straight off the connection module (it must not go through CallConnectivityMonitor, which would close a dependency cycle), but injectMockNetwork was repointed at the monitor. The loop therefore polled the real provider and stalled without consuming an attempt. Injecting at the module fixes three reconnect tests that had been failing since the decomposition commit. JoinRecoverableFailureTest is rebuilt on the coordinator harness: it relied on spying Call and reflectively repointing CallJoinCoordinator.call, a field that no longer exists. Core suite: 978 tests, 0 failures. Co-authored-by: Cursor <cursoragent@cursor.com> * refactor(core): read session state from its owner instead of the Call facade CallSessionManager owns the session identity and reconnect bookkeeping, but Call still re-exposed it through internal accessors that mostly existed for a single caller. Remove them: location and nonFastReconnectAttempts were only reachable from tests, connectStartTime/reconnectStartTime had dead setters, and unifiedSessionId was read by RtcSession alone. RtcSession now takes CallSessionManager directly and reads session identity and reconnect timings from it. The elapsed-time arithmetic moves onto the manager as connectionTimeSeconds()/reconnectionTimeSeconds(), next to the timestamps it derives from. The two test-only reads move into CallTestSeams.kt so they stay out of the production API. Co-authored-by: Cursor <cursoragent@cursor.com> * refactor(core): break the reconnector/join-coordinator dependency cycle CallReconnector and CallJoinCoordinator each depended on the other, so one had to be injected as a lazy provider. Move the shared joinRequest into CallApiClient, which already owns the coordinator REST calls, and relocate the failed-SFU set to CallSessionManager so the request no longer has to ask the reconnector for it. Both orchestrators now depend on the api client and neither depends on the other. Also drops the provider lambdas around state, analytics, stats and media by declaring those components before their consumers. FailedSfuIdsTest no longer needs reflection into private reconnector members; the behaviour is covered directly in CallSessionManagerTest. Co-authored-by: Cursor <cursoragent@cursor.com> * chore: remove unused code (#1762) --------- Co-authored-by: Cursor <cursoragent@cursor.com> Co-authored-by: Rahul Kumar Lohra <tgunix@gmail.com> * [skip ci] Update SDK sizes * Make the Android Video SDK compatible with Android 17. (#1752) * Chore: make incoming call android 17 compatible * Chore: revert * Chore: refactor name * refactor: remove maxSize named argument for AGP 9 compatibility * refactor: remove unnecessary inline arg * update: Update streamLog to 1.3.4 which fixes namespace issue which is returned as error in AGP 9 * test: cover incoming-ring service type on Android 15/16/17 Move VERSION_CODES polyfills to shared AndroidVersionCodes.kt (fix BALAKLAVA typo). Force SDK_INT for API 35/36/37 tests and restore it via @after teardown. * AUTOMATION: Version Bump * test(e2e): widen call-start wait to 30s for reconnection/recording flow --------- Signed-off-by: dependabot[bot] <support@github.com> Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Co-authored-by: Cursor <cursoragent@cursor.com> Co-authored-by: Aleksandar Apostolov <apostolov.alexandar@gmail.com> Co-authored-by: Rahul Kumar Lohra <tgunix@gmail.com> Co-authored-by: coderabbitai[bot] <136622811+coderabbitai[bot]@users.noreply.github.com> Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com> Co-authored-by: Peter Matkovski <peter.matkovski@getstream.io> Co-authored-by: André Mion <andremion@gmail.com> Co-authored-by: Gianmarco <47775302+gpunto@users.noreply.github.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Goal
Closes #AND-1226 Add Video Analytics in call-join flow
Analytics Flow animation
You may open this and play this html animation to understand the flow
analytics-flow-animation.html
Give us visibility into the call-join funnel directly from the client. Today, when a user fails to get into a call (or joins but never sees/hears media), we have no client-side signal telling us which stage failed, why, and how long each stage took.
This PR adds a new analytics subsystem under
io.getstream.video.android.core.analyticsthat instruments every stage of the call lifecycle and reports pairedINITIATED/COMPLETEDevents to the newreportClientCallEventbackend endpoint:CoordinatorWSJoinInitiatedcall.join()SDK method invokedCoordinatorJoinWSJoinPeerConnectionConnectFirstAudioFrame/FirstVideoFrameMediaDevicePermissionEvery event carries correlation IDs (
join_attempt_id,stage_id,call_session_id,coordinator_connect_id,sfu_id), an outcome (SUCCESS/FAILURE),elapsed_time, retry counts and failure reason/code — so the backend can reconstruct the full join funnel per attempt.Delivery is best-effort but resilient: events are sent fire-and-forget with in-process exponential-backoff retry,
Implementation
Follow below
Architecture
The LLC analytics is split into five layers with a strict one-way dependency flow — observers never talk to the network, the dispatcher never knows about call semantics:
flowchart TB subgraph SDK["SDK core (instrumentation hooks)"] Call["Call.kt<br/>(join / leave / first frame)"] Rtc["RtcSession /<br/>SfuSocketConnection"] Coord["CoordinatorSocketStateService<br/>(VideoSocketState flow)"] end subgraph OBS["Observation layer — analytics.call / analytics.coordinator"] CA["CallAnalytics<br/>(per-call façade, owns all observers)"] JA["JoinAnalytics"] SA["SfuAnalytics"] PCA["PeerConnectionAnalytics<br/>(collects publisher/subscriber<br/>PeerConnectionState flows)"] AA["AudioAnalytics<br/>(AudioTrackSink, first remote frame)"] VA["VideoAnalytics<br/>(first video frame)"] MPO["MediaPermissionObserver"] CoA["CoordinatorAnalytics<br/>(client-scoped)"] end subgraph STATE["Shared correlation state"] JSH["JoinAnalyticsStateHolder<br/>joinStageAttemptId · joinReason ·<br/>callSessionId · stage guard"] SSH["SfuAnalyticsStateHolder<br/>sfuId · wsStage"] PSH["PeerConnectionAnalyticsStateHolder<br/>publisher/subscriber stage · jobs"] end subgraph REP["Reporting layer — analytics.reporting"] CER["ClientEventReporter<br/>(in-flight session registry,<br/>elapsed time, ICE state machine,<br/>abort flush)"] CEF["ClientEventFactory<br/>(builds ClientEvent payloads)"] end subgraph DISP["Dispatch layer — analytics.reporting.dispatcher"] ED["EventDispatcher (interface)"] IED["ImmediateEventDispatcher<br/>(coroutine per batch,<br/>exp. backoff retry)"] end API["ProductvideoApi.reportClientCallEvent<br/>(generated)"] Call --> CA Rtc --> SA Rtc --> PCA Coord --> CoA CA --> JA & SA & PCA & AA & VA & MPO JA & SA & PCA & AA & VA & MPO -. read/write .-> JSH & SSH PCA -.-> PSH JA & SA & PCA & AA & VA & MPO --> CER CoA --> CER CER --> CEF CER --> ED ED --- IED IED -->|"send (async)"| API IED -->|"on failure: save"| SDS SDS --> FDS SDS -.-> IMD IED -->|"retryPending: loadAndClear"| SDSLayer responsibilities
Call,RtcSession,SfuSocketConnection,CoordinatorSocketStateService,StreamVideoClientCallAnalytics(per-call façade),JoinAnalytics,SfuAnalytics,PeerConnectionAnalytics,AudioAnalytics,VideoAnalytics,MediaPermissionObserver,CoordinatorAnalytics(client-scoped)Stagestate machine (NOT_STARTED → IN_PROGRESS → NOT_STARTED) so a stage is reported exactly once per attempt, even when callbacks fire repeatedly.JoinAnalyticsStateHolder,SfuAnalyticsStateHolder,PeerConnectionAnalyticsStateHolderStateFlow-backed holders that share correlation context (joinStageAttemptId,joinReason,callSessionId,sfuId) across observers so events from different subsystems stitch together into one join attempt.ClientEventReporter,ClientEventFactory,InFlightSessionmodelsINITIATED/COMPLETEDpairing. On initiated, stores anInFlightSessionkeyed by a generatedstageId; on completed, removes it and computeselapsed_time. Also runs the peer-connection ICE state machine and the abort flush (below).EventDispatcher,ImmediateEventDispatcherComposition / scoping
StreamVideoinstance):ClientEventReporteris created inClientStateviaClientEventReporter.getDefault(context, api)and wired withImmediateEventDispatcherCoordinatorAnalyticslives onStreamVideoClientand observes the coordinator socket state flow.Call):CallAnalyticsinstantiates all per-call observers and their state holders, sharing the client-scoped reporter. This keeps correlation state isolated per callCommunication between components (happy path)
sequenceDiagram participant Call as Call / RtcSession participant Obs as Observers<br/>(Join/Sfu/PC/Media) participant SH as StateHolders participant Rep as ClientEventReporter participant Disp as ImmediateEventDispatcher participant DS as PendingEventDataSource participant BE as Backend API Note over Call,BE: Pre-call (client scope) Call->>Rep: CoordinatorAnalytics: WS Connecting (INITIAL_CONNECTION) Rep->>Disp: CoordinatorWS INITIATED Disp->>BE: reportClientCallEvent Call->>Rep: WS Connected Rep->>Disp: CoordinatorWS COMPLETED (success, retryCount) Rep->>Disp: retryPending() Disp->>DS: loadAndClear() — replay any persisted events Note over Call,BE: call.join() Call->>Obs: onJoinFunctionStart() Obs->>SH: new joinStageAttemptId Obs->>Rep: JoinInitiated INITIATED Call->>Obs: onJoinRequestStart(joinReason) Obs->>Rep: CoordinatorJoin INITIATED → stageId Rep->>Rep: store InFlightSession(stageId, startedAtMs) Call->>Obs: onJoinRequestSuccess(sessionId) Obs->>SH: callSessionId Obs->>Rep: CoordinatorJoin COMPLETED (elapsed = now − startedAtMs) Call->>Obs: SFU WS connect Obs->>Rep: WSJoin INITIATED / COMPLETED Call->>Obs: PeerConnectionState changes (publisher & subscriber flows) Obs->>Rep: onPeerConnectionStateChanged(role, iceState) Note over Rep: ICE state machine:<br/>CHECKING → open session (close superseded as FAILURE)<br/>CONNECTED → COMPLETED success<br/>FAILED → COMPLETED failure Call->>Obs: first video frame / first remote audio frame Obs->>Rep: FirstVideoFrame / FirstAudioFrame INITIATED Rep->>Disp: send(event) Disp->>BE: reportClientCallEventAbort path: when the user leaves a call while any stage is still
IN_PROGRESS(CallAnalytics.onCallLeave),ClientEventReporter.abortAllPostCallInFlight(reason)snapshots and clears every in-flight session and emits aCOMPLETED/FAILUREevent for each, with anAnalyticsCallAbortReason(CLIENT_ABORTEDorBACKEND_LEAVE) — so abandoned joins are never silently dropped from the funnel.Retry mechanism
Delivery is an in-process backoff loop for transient blips
flowchart TB A["dispatcher.send(events)"] --> B["launch coroutine<br/>(UserScope)"] B --> C{"POST reportClientCallEvent"} C -->|success| D([done]) C -->|failure| E{"retryable?<br/>HTTP 5xx · SocketTimeout ·<br/>Connect · UnknownHost · IOException"} E -->|"yes, attempts < 5"| F["delay 500ms × 2^attempt<br/>(500 → 1000 → 2000 → 4000 → 8000)"] F --> C E -->|"no / attempts exhausted"| G["dataSource.save(events)<br/>append to pending_events.ndjson"] H["Coordinator WS connected<br/>(initial connect or reconnect)"] --> I["retryPending()"] I --> J["loadAndClear()<br/>drain batch of 10"] J --> AImmediateEventDispatcher.retryInternal): up to 5 attempts with exponential backoff (base 500 ms, doubling to 8 s). Only transient failures are retried — HTTP 5xx and connectivity exceptions (SocketTimeoutException,ConnectException,UnknownHostException,IOException). Client errors (4xx) fail fast without burning retries.Implementation details
io.getstream.video.android.core.analytics(~20 classes): observation (call.observer,coordinator), reporting (reporting,reporting.model), dispatch (reporting.dispatcher).Everything is
internal— no public API surface added (.apichanges are from generated models only).ClientEvent,ReportClientEventRequest/ResponseandProductvideoApi.reportClientCallEventendpoint.ClientEventFactorycentralizes payload construction: SDK version, user agent (capped at 512 chars), user id, timestamps, permission status mapping, and the client-generatedcoordinator_connect_id.flatMapLatestoverRtcSession.publisher/subscriberStateFlows, so observers automatically re-attach across rejoins/ICE restarts (stopAndObservePeerConnections).AudioTrackSinkto remote tracks with strict real-time-thread discipline (no logging/blocking on the WebRTC audio thread; CAS guard + coroutine hand-off; sink removal deferred to the coroutine to avoid deadlocking WebRTC's sink-list lock). Currently behind a disabled flag (isEnabled = false).INITIATED/COMPLETEDpair per attempt.CallLeaveReasonwork from this branch: leave reasons map toAnalyticsCallAbortReasonfor the abort flush.🎨 UI Changes
None
Testing
Summary by CodeRabbit
Release Notes