From 653914f47e1fd53823e2a5e17018a84428831b70 Mon Sep 17 00:00:00 2001
From: Clark <clark@signal.org>
Date: Tue, 23 Jan 2024 11:34:22 -0500
Subject: [PATCH] Allow multiple message latency percentiles.

---
 .../SlowNotificationHeuristics.kt             | 28 +++++++++++--------
 1 file changed, 16 insertions(+), 12 deletions(-)

diff --git a/app/src/main/java/org/thoughtcrime/securesms/notifications/SlowNotificationHeuristics.kt b/app/src/main/java/org/thoughtcrime/securesms/notifications/SlowNotificationHeuristics.kt
index 28182db88c..87dfc946b2 100644
--- a/app/src/main/java/org/thoughtcrime/securesms/notifications/SlowNotificationHeuristics.kt
+++ b/app/src/main/java/org/thoughtcrime/securesms/notifications/SlowNotificationHeuristics.kt
@@ -21,6 +21,7 @@ import org.thoughtcrime.securesms.util.SignalLocalMetrics
 import java.util.concurrent.TimeUnit
 import kotlin.time.Duration.Companion.days
 import kotlin.time.Duration.Companion.hours
+import kotlin.time.Duration.Companion.minutes
 
 /**
  * Heuristic for estimating if a user has been experiencing issues with delayed notifications.
@@ -52,10 +53,12 @@ object SlowNotificationHeuristics {
       minimumEventAgeMs = 3.days.inWholeMilliseconds,
       minimumServiceEventCount = 10,
       serviceStartFailurePercentage = 0.5f,
-      messageLatencyPercentage = 75,
-      messageLatencyThreshold = 6.hours.inWholeMilliseconds,
       minimumMessageLatencyEvents = 50,
-      weeklyFailedQueueDrains = 5
+      weeklyFailedQueueDrains = 5,
+      messageLatencyPercentiles = mapOf(
+        90 to 2.hours.inWholeMilliseconds,
+        50 to 30.minutes.inWholeMilliseconds
+      )
     )
   }
 
@@ -102,7 +105,7 @@ object SlowNotificationHeuristics {
 
     val failedServiceStarts = hasRepeatedFailedServiceStarts(metrics, configuration.minimumEventAgeMs, configuration.minimumServiceEventCount, configuration.serviceStartFailurePercentage)
     val failedQueueDrains = isFailingToDrainQueue(metrics, configuration.minimumEventAgeMs, configuration.weeklyFailedQueueDrains)
-    val longMessageLatency = hasLongMessageLatency(metrics, configuration.minimumEventAgeMs, configuration.messageLatencyPercentage, configuration.minimumMessageLatencyEvents, configuration.messageLatencyThreshold)
+    val longMessageLatency = hasLongMessageLatency(metrics, configuration.minimumEventAgeMs, configuration.minimumMessageLatencyEvents, configuration.messageLatencyPercentiles)
 
     if (failedServiceStarts || failedQueueDrains || longMessageLatency) {
       Log.w(TAG, "User seems to be having delayed notifications: failed-service-starts=$failedServiceStarts failedQueueDrains=$failedQueueDrains longMessageLatency=$longMessageLatency")
@@ -171,7 +174,7 @@ object SlowNotificationHeuristics {
     return true
   }
 
-  private fun hasLongMessageLatency(metrics: List<LocalMetricsDatabase.EventMetrics>, minimumEventAgeMs: Long, percentage: Int, messageThreshold: Int, durationThreshold: Long): Boolean {
+  private fun hasLongMessageLatency(metrics: List<LocalMetricsDatabase.EventMetrics>, minimumEventAgeMs: Long, messageThreshold: Int, percentiles: Map<Int, Long>): Boolean {
     if (!haveEnoughData(SignalLocalMetrics.MessageLatency.NAME_HIGH, minimumEventAgeMs)) {
       Log.d(TAG, "insufficient data for message latency")
       return false
@@ -182,13 +185,15 @@ object SlowNotificationHeuristics {
       return false
     }
     val db = LocalMetricsDatabase.getInstance(ApplicationDependencies.getApplication())
-    val averageLatency = db.eventPercent(SignalLocalMetrics.MessageLatency.NAME_HIGH, percentage.coerceAtMost(100).coerceAtLeast(0))
+    for ((percentage, threshold) in percentiles.entries) {
+      val averageLatency = db.eventPercent(SignalLocalMetrics.MessageLatency.NAME_HIGH, percentage.coerceAtMost(100).coerceAtLeast(0))
 
-    val longMessageLatency = averageLatency > durationThreshold
-    if (longMessageLatency) {
-      Log.w(TAG, "User has high average message latency of $averageLatency ms over $eventCount events")
+      if (averageLatency > threshold) {
+        Log.w(TAG, "User has high average message latency of $averageLatency ms over $eventCount events over threshold of $threshold ms")
+        return true
+      }
     }
-    return longMessageLatency
+    return false
   }
 
   private fun haveEnoughData(eventName: String, minimumEventAgeMs: Long): Boolean {
@@ -206,6 +211,5 @@ data class Configuration(
   val serviceStartFailurePercentage: Float,
   val weeklyFailedQueueDrains: Int,
   val minimumMessageLatencyEvents: Int,
-  val messageLatencyThreshold: Long,
-  val messageLatencyPercentage: Int
+  val messageLatencyPercentiles: Map<Int, Long>
 )