feat: keep sessions and replays missed by sampling until the session errors

features/dd-sdk-android-rum/api/apiSurface

@@ -68,6 +68,8 @@ data class com.datadog.android.rum.RumConfiguration
     fun setSessionSampleRate(Float): Builder
     fun setRemoteConfigurationEnabled(Boolean): Builder
     fun setBeforeSampling(BeforeSamplingCallback): Builder
+    fun setSessionOnError(Boolean): Builder
+    fun setSessionReplayOnError(Boolean): Builder
     fun collectAccessibility(Boolean): Builder
     fun setTelemetrySampleRate(Float): Builder
     fun trackUserInteractions(Array<com.datadog.android.rum.tracking.ViewAttributesProvider> = emptyArray(), com.datadog.android.rum.tracking.InteractionPredicate = NoOpInteractionPredicate()): Builder
@@ -1659,7 +1661,7 @@ data class com.datadog.android.rum.model.ViewEvent
       fun fromJson(kotlin.String): Application
       fun fromJsonObject(com.google.gson.JsonObject): Application
   data class ViewEventSession
-    constructor(kotlin.String, ViewEventSessionType, kotlin.Boolean? = null, kotlin.Boolean? = true, kotlin.Boolean? = null)
+    constructor(kotlin.String, ViewEventSessionType, kotlin.Boolean? = null, kotlin.Boolean? = true, kotlin.Boolean? = null, kotlin.Boolean? = null, kotlin.Boolean? = null)
     fun toJson(): com.google.gson.JsonElement
     companion object 
       fun fromJson(kotlin.String): ViewEventSession

features/dd-sdk-android-rum/api/dd-sdk-android-rum.api

@@ -128,6 +128,8 @@ public final class com/datadog/android/rum/RumConfiguration$Builder {
 	public final fun setRemoteConfigurationEnabled (Z)Lcom/datadog/android/rum/RumConfiguration$Builder;
 	public final fun setResourceEventMapper (Lcom/datadog/android/event/EventMapper;)Lcom/datadog/android/rum/RumConfiguration$Builder;
 	public final fun setSessionListener (Lcom/datadog/android/rum/RumSessionListener;)Lcom/datadog/android/rum/RumConfiguration$Builder;
+	public final fun setSessionOnError (Z)Lcom/datadog/android/rum/RumConfiguration$Builder;
+	public final fun setSessionReplayOnError (Z)Lcom/datadog/android/rum/RumConfiguration$Builder;
 	public final fun setSessionSampleRate (F)Lcom/datadog/android/rum/RumConfiguration$Builder;
 	public final fun setSlowFramesConfiguration (Lcom/datadog/android/rum/configuration/SlowFramesConfiguration;)Lcom/datadog/android/rum/RumConfiguration$Builder;
 	public final fun setTelemetrySampleRate (F)Lcom/datadog/android/rum/RumConfiguration$Builder;
@@ -6037,20 +6039,24 @@ public final class com/datadog/android/rum/model/ViewEvent$Usr$Companion {
 
 public final class com/datadog/android/rum/model/ViewEvent$ViewEventSession {
 	public static final field Companion Lcom/datadog/android/rum/model/ViewEvent$ViewEventSession$Companion;
-	public fun <init> (Ljava/lang/String;Lcom/datadog/android/rum/model/ViewEvent$ViewEventSessionType;Ljava/lang/Boolean;Ljava/lang/Boolean;Ljava/lang/Boolean;)V
-	public synthetic fun <init> (Ljava/lang/String;Lcom/datadog/android/rum/model/ViewEvent$ViewEventSessionType;Ljava/lang/Boolean;Ljava/lang/Boolean;Ljava/lang/Boolean;ILkotlin/jvm/internal/DefaultConstructorMarker;)V
+	public fun <init> (Ljava/lang/String;Lcom/datadog/android/rum/model/ViewEvent$ViewEventSessionType;Ljava/lang/Boolean;Ljava/lang/Boolean;Ljava/lang/Boolean;Ljava/lang/Boolean;Ljava/lang/Boolean;)V
+	public synthetic fun <init> (Ljava/lang/String;Lcom/datadog/android/rum/model/ViewEvent$ViewEventSessionType;Ljava/lang/Boolean;Ljava/lang/Boolean;Ljava/lang/Boolean;Ljava/lang/Boolean;Ljava/lang/Boolean;ILkotlin/jvm/internal/DefaultConstructorMarker;)V
 	public final fun component1 ()Ljava/lang/String;
 	public final fun component2 ()Lcom/datadog/android/rum/model/ViewEvent$ViewEventSessionType;
 	public final fun component3 ()Ljava/lang/Boolean;
 	public final fun component4 ()Ljava/lang/Boolean;
 	public final fun component5 ()Ljava/lang/Boolean;
-	public final fun copy (Ljava/lang/String;Lcom/datadog/android/rum/model/ViewEvent$ViewEventSessionType;Ljava/lang/Boolean;Ljava/lang/Boolean;Ljava/lang/Boolean;)Lcom/datadog/android/rum/model/ViewEvent$ViewEventSession;
-	public static synthetic fun copy$default (Lcom/datadog/android/rum/model/ViewEvent$ViewEventSession;Ljava/lang/String;Lcom/datadog/android/rum/model/ViewEvent$ViewEventSessionType;Ljava/lang/Boolean;Ljava/lang/Boolean;Ljava/lang/Boolean;ILjava/lang/Object;)Lcom/datadog/android/rum/model/ViewEvent$ViewEventSession;
+	public final fun component6 ()Ljava/lang/Boolean;
+	public final fun component7 ()Ljava/lang/Boolean;
+	public final fun copy (Ljava/lang/String;Lcom/datadog/android/rum/model/ViewEvent$ViewEventSessionType;Ljava/lang/Boolean;Ljava/lang/Boolean;Ljava/lang/Boolean;Ljava/lang/Boolean;Ljava/lang/Boolean;)Lcom/datadog/android/rum/model/ViewEvent$ViewEventSession;
+	public static synthetic fun copy$default (Lcom/datadog/android/rum/model/ViewEvent$ViewEventSession;Ljava/lang/String;Lcom/datadog/android/rum/model/ViewEvent$ViewEventSessionType;Ljava/lang/Boolean;Ljava/lang/Boolean;Ljava/lang/Boolean;Ljava/lang/Boolean;Ljava/lang/Boolean;ILjava/lang/Object;)Lcom/datadog/android/rum/model/ViewEvent$ViewEventSession;
 	public fun equals (Ljava/lang/Object;)Z
 	public static final fun fromJson (Ljava/lang/String;)Lcom/datadog/android/rum/model/ViewEvent$ViewEventSession;
 	public static final fun fromJsonObject (Lcom/google/gson/JsonObject;)Lcom/datadog/android/rum/model/ViewEvent$ViewEventSession;
 	public final fun getHasReplay ()Ljava/lang/Boolean;
 	public final fun getId ()Ljava/lang/String;
+	public final fun getSampledForError ()Ljava/lang/Boolean;
+	public final fun getSampledForErrorReplay ()Ljava/lang/Boolean;
 	public final fun getSampledForReplay ()Ljava/lang/Boolean;
 	public final fun getType ()Lcom/datadog/android/rum/model/ViewEvent$ViewEventSessionType;
 	public fun hashCode ()I

features/dd-sdk-android-rum/src/main/json/rum/view-schema.json

@@ -420,6 +420,16 @@
               "type": "boolean",
               "description": "Whether this session has been sampled for replay",
               "readOnly": true
+            },
+            "sampled_for_error": {
+              "type": "boolean",
+              "description": "FlashCat fork - whether this session was kept only because it reported an error; its detail starts where the withheld buffer reached",
+              "readOnly": true
+            },
+            "sampled_for_error_replay": {
+              "type": "boolean",
+              "description": "FlashCat fork - whether the replay of this session was kept only because the session reported an error",
+              "readOnly": true
             }
           },
           "readOnly": true

features/dd-sdk-android-rum/src/main/kotlin/com/datadog/android/rum/Rum.kt

@@ -127,6 +127,10 @@ object Rum {
 
         val rumAppStartupTelemetryReporter = RumAppStartupTelemetryReporter.create(sdkCore = sdkCore)
 
+        val handler = Handler(Looper.getMainLooper())
+
+        val withheldEvents = rumFeature.withheldEvents
+
         return DatadogRumMonitor(
             applicationId = rumFeature.applicationId,
             sdkCore = sdkCore,
@@ -137,8 +141,11 @@ object Rum {
             // simply asks again, and there is nothing to ask with when the app did not opt in.
             onSessionDrawn = { rumFeature.remoteConfigController?.onSessionStarted() },
             beforeSampling = rumFeature.configuration.beforeSampling,
-            writer = rumFeature.dataWriter,
-            handler = Handler(Looper.getMainLooper()),
+            writer = withheldEvents ?: rumFeature.dataWriter,
+            withheldEvents = withheldEvents,
+            sessionOnError = rumFeature.configuration.sessionOnError,
+            sessionReplayOnError = rumFeature.configuration.sessionReplayOnError,
+            handler = handler,
             telemetryEventHandler = TelemetryEventHandler(
                 sdkCore = sdkCore,
                 eventSampler = RateBasedSampler(rumFeature.telemetrySampleRate),

features/dd-sdk-android-rum/src/main/kotlin/com/datadog/android/rum/RumConfiguration.kt

@@ -104,6 +104,51 @@ data class RumConfiguration internal constructor(
             return this
         }
 
+        /**
+         * Also keep the sessions the session sample rate does not keep, but only those that
+         * report an error.
+         *
+         * Such a session is collected in memory without uploading anything, keeping only its last
+         * minute. If it reports an error, that minute is uploaded and the session carries on like
+         * any collected one; if it ends without one, what it held is thrown away and nothing of it
+         * is ever sent. Its events report a session sample rate of 0 - it stands for itself rather
+         * than for the population a rate would imply - and its view events carry
+         * `session.sampled_for_error`. An error dropped by an event mapper does not count, and
+         * neither does a session whose rate [setBeforeSampling] set to 0.
+         *
+         * A native crash is reported at the next launch with the last view, like for any collected
+         * session, but without the minute leading up to it.
+         *
+         * Disabled by default. With [setRemoteConfigurationEnabled], the console's value wins over
+         * this one.
+         *
+         * @param enabled whether to keep the sessions that report an error.
+         */
+        fun setSessionOnError(enabled: Boolean): Builder {
+            rumConfig = rumConfig.copy(sessionOnError = enabled)
+            return this
+        }
+
+        /**
+         * Also keep the replays the Session Replay sample rate does not keep, but only those of
+         * sessions that report an error.
+         *
+         * Such a replay is recorded in memory without uploading anything, keeping only its last
+         * minute; it is uploaded if the session reports an error, and thrown away if it ends
+         * without one. A session kept only through [setSessionOnError] holds its replay until its
+         * events are released, whether or not the replay rate kept it. View events of a session
+         * whose replay is kept this way carry `session.sampled_for_error_replay`.
+         *
+         * Needs the Session Replay feature. Disabled by default. With
+         * [setRemoteConfigurationEnabled], the console's value wins over this one.
+         *
+         * @param enabled whether to keep the replays of sessions that report an error.
+         */
+        fun setSessionReplayOnError(enabled: Boolean): Builder {
+            rumConfig = rumConfig.copy(sessionReplayOnError = enabled)
+            return this
+        }
+
         /**
          * Whether to collect accessibility attributes - this is disabled by default.
          *

features/dd-sdk-android-rum/src/main/kotlin/com/datadog/android/rum/internal/DatadogLateCrashReporter.kt

@@ -95,7 +95,7 @@ internal class DatadogLateCrashReporter(
             )
             writeScope {
                 rumWriter.write(it, toSendErrorEvent, EventType.CRASH)
-                if (lastViewEvent.isWithinSessionAvailability) {
+                if (lastViewEvent.isWithinSessionAvailability || lastViewEvent.isOnlyKnownHere) {
                     val updatedViewEvent = updateViewEvent(lastViewEvent)
                     rumWriter.write(it, updatedViewEvent, EventType.CRASH)
                 }
@@ -156,7 +156,7 @@ internal class DatadogLateCrashReporter(
                     // RUM events are written
                     sdkCore.writeLastFatalAnrSent(anrExitInfo.timestamp)
                     rumWriter.write(it, toSendErrorEvent, EventType.CRASH)
-                    if (lastViewEvent.isWithinSessionAvailability) {
+                    if (lastViewEvent.isWithinSessionAvailability || lastViewEvent.isOnlyKnownHere) {
                         val updatedViewEvent = updateViewEvent(lastViewEvent)
                         rumWriter.write(it, updatedViewEvent, EventType.CRASH)
                     }
@@ -326,6 +326,13 @@ internal class DatadogLateCrashReporter(
     private val ViewEvent.sampleRate: Float
         get() = dd.configuration?.sessionSampleRate?.toFloat() ?: 0f
 
+    /**
+     * FLASHCAT FORK - a session kept only on error uploaded nothing before it crashed: the view
+     * written locally is the only one the intake will ever get, however old it is by now.
+     */
+    private val ViewEvent.isOnlyKnownHere: Boolean
+        get() = session.sampledForError == true
+
     private val ViewEvent.isWithinSessionAvailability: Boolean
         get() {
             val now = sdkCore.timeProvider.getDeviceTimestampMillis()

features/dd-sdk-android-rum/src/main/kotlin/com/datadog/android/rum/internal/FeaturesContextResolver.kt

@@ -27,8 +27,37 @@ internal class FeaturesContextResolver {
         return (sessionReplayMetadata?.get(VIEW_RECORDS_COUNT_KEY) as? Long) ?: 0L
     }
 
+    /**
+     * FLASHCAT FORK - whether the replay of this session is kept only because it reports an error,
+     * or null when Session Replay is not enabled.
+     */
+    fun resolveSampledForErrorReplay(datadogContext: DatadogContext, sessionId: String): Boolean? {
+        val sessionReplayContext =
+            datadogContext.featuresContext[Feature.SESSION_REPLAY_FEATURE_NAME] ?: return null
+        return sessionReplayContext[REPLAY_ON_ERROR_SESSION_KEY] == sessionId
+    }
+
+    /**
+     * FLASHCAT FORK - whether this session is sampled for replay, reported only where an on-error
+     * mode is in play for it (null otherwise, so nothing changes for anyone who did not opt in). A
+     * replay still withheld counts when the session's events are withheld too: it is released along
+     * with them, so if these events ever reach the intake, so does the replay.
+     */
+    fun resolveSampledForReplay(datadogContext: DatadogContext, sessionId: String, sampledForError: Boolean): Boolean? {
+        val sessionReplayContext =
+            datadogContext.featuresContext[Feature.SESSION_REPLAY_FEATURE_NAME] ?: return null
+        if (!sampledForError && sessionReplayContext[REPLAY_ON_ERROR_SESSION_KEY] != sessionId) return null
+        val isRecording = sessionReplayContext[REPLAY_ENABLED_KEY] as? Boolean ?: false
+        val isWithheld = sessionReplayContext[REPLAY_WITHHELD_KEY] as? Boolean ?: false
+        return isRecording && (!isWithheld || sampledForError)
+    }
+
     companion object {
         internal const val HAS_REPLAY_KEY = "has_replay"
         internal const val VIEW_RECORDS_COUNT_KEY = "records_count"
+        internal const val VIEW_WITHHELD_RECORDS_COUNT_KEY = "withheld_records_count"
+        internal const val REPLAY_ENABLED_KEY = "session_replay_is_enabled"
+        internal const val REPLAY_ON_ERROR_SESSION_KEY = "session_replay_on_error_session_id"
+        internal const val REPLAY_WITHHELD_KEY = "session_replay_withheld"
     }
 }