From 9c054e285d8980daae01b45672c46899f3af237d Mon Sep 17 00:00:00 2001
From: Trey Moen <50057480+greatgitsby@users.noreply.github.com>
Date: Sun, 13 Sep 2026 14:19:59 -0700
Subject: [PATCH] replay: retry failed segment loads (#38897)

* replay: retry failed segment loads

* replay: rename test helper for spellcheck

* replay: include portable temporary directory declaration

* replay: name attempt limit and shorten test delays

* replay: drop timing-dependent retry tests
---
 openpilot/tools/replay/seg_mgr.cc | 29 ++++++++++++++++++++++++++---
 openpilot/tools/replay/seg_mgr.h  |  9 +++++++++
 2 files changed, 35 insertions(+), 3 deletions(-)

diff --git a/openpilot/tools/replay/seg_mgr.cc b/openpilot/tools/replay/seg_mgr.cc
index 074e7e53db..2c616bd004 100644
--- a/openpilot/tools/replay/seg_mgr.cc
+++ b/openpilot/tools/replay/seg_mgr.cc
@@ -51,10 +51,11 @@ void SegmentManager::setCurrentSegment(int seg_num) {
 void SegmentManager::manageSegmentCache() {
   while (true) {
     std::unique_lock lock(mutex_);
-    cv_.wait(lock, [this]() { return exit_ || needs_update_; });
+    cv_.wait_until(lock, next_retry_, [this]() { return exit_ || needs_update_; });
     if (exit_) break;
 
     needs_update_ = false;
+    next_retry_ = std::chrono::steady_clock::time_point::max();
     auto cur = segments_.lower_bound(cur_seg_num_);
     if (cur == segments_.end()) continue;
 
@@ -69,8 +70,12 @@ void SegmentManager::manageSegmentCache() {
     bool merged = mergeSegments(begin, end);
 
     // Free segments outside the current range
-    std::for_each(segments_.begin(), begin, [](auto &segment) { segment.second.reset(); });
-    std::for_each(end, segments_.end(), [](auto &segment) { segment.second.reset(); });
+    auto evict = [this](auto &segment) {
+      segment.second.reset();
+      load_attempts_.erase(segment.first);
+    };
+    std::for_each(segments_.begin(), begin, evict);
+    std::for_each(end, segments_.end(), evict);
 
     if (merged && onSegmentMergedCallback_) {
       onSegmentMergedCallback_();  // Notify listener that segments have been merged
@@ -121,7 +126,25 @@ void SegmentManager::loadSegmentsInRange(SegmentMap::iterator begin, SegmentMap:
   auto tryLoadSegment = [this](auto first, auto last) {
     for (auto it = first; it != last; ++it) {
       auto &segment_ptr = it->second;
+      auto &attempt = load_attempts_[it->first];
+      if (segment_ptr && segment_ptr->getState() == Segment::LoadState::Failed) {
+        // A failed object must not permanently occupy its cache slot. Back off
+        // between retries while allowing other segments to load meanwhile.
+        if (attempt.count >= MAX_SEGMENT_LOAD_ATTEMPTS) continue;
+        const auto now = std::chrono::steady_clock::now();
+        if (attempt.retry_at == std::chrono::steady_clock::time_point::max()) {
+          attempt.retry_at = now + std::chrono::seconds(attempt.count);
+        }
+        if (now < attempt.retry_at) {
+          next_retry_ = std::min(next_retry_, attempt.retry_at);
+          continue;
+        }
+        segment_ptr.reset();
+        rWarning("retrying segment %d (attempt %d/%d)", it->first, attempt.count + 1, MAX_SEGMENT_LOAD_ATTEMPTS);
+      }
       if (!segment_ptr) {
+        ++attempt.count;
+        attempt.retry_at = std::chrono::steady_clock::time_point::max();
         if (onBenchmarkEvent_) {
           onBenchmarkEvent_(it->first, "loading");
         }
diff --git a/openpilot/tools/replay/seg_mgr.h b/openpilot/tools/replay/seg_mgr.h
index 55b412b25a..6542335caf 100644
--- a/openpilot/tools/replay/seg_mgr.h
+++ b/openpilot/tools/replay/seg_mgr.h
@@ -1,5 +1,6 @@
 #pragma once
 
+#include <chrono>
 #include <condition_variable>
 #include <map>
 #include <mutex>
@@ -9,6 +10,7 @@
 #include "tools/replay/route.h"
 
 constexpr int MIN_SEGMENTS_CACHE = 5;
+constexpr int MAX_SEGMENT_LOAD_ATTEMPTS = 3;  // Includes the initial load.
 
 using SegmentMap = std::map<int, std::shared_ptr<Segment>>;
 
@@ -56,4 +58,11 @@ private:
   std::function<void()> onSegmentMergedCallback_ = nullptr;
   std::function<void(int, const std::string&)> onBenchmarkEvent_ = nullptr;
   std::set<int> merged_segments_;
+  struct LoadAttempt {
+    int count = 0;
+    std::chrono::steady_clock::time_point retry_at = std::chrono::steady_clock::time_point::max();
+  };
+  // Accessed only by the cache management thread.
+  std::map<int, LoadAttempt> load_attempts_;
+  std::chrono::steady_clock::time_point next_retry_ = std::chrono::steady_clock::time_point::max();
 };