Merge remote-tracking branch 'origin/3.12.4' into bug-fix/print-on-vi…

…olated-block-ordering
arangodb · neunhoef · May 2, 2025 · Apr 30, 2025 · Apr 30, 2025 · Apr 30, 2025
commit 325d1f27cf01f5b1f56748823fd6c173b2aaab26
diff --git a/CHANGELOG b/CHANGELOG
@@ -1,3 +1,14 @@
+3.12.4.4 (XXXX-XX-XX)
+---------------------
+
+* Add detection of two threads working on an ExecutionBlock concurrently.
+
+* Add detection of two threads waiting for a PrefetchTask concurrently.
+
+* Add visibility into the state if a thread waits for more than a second
+  on a PrefetchTask.
+
+
 3.12.4.3 (2025-04-25)
 ---------------------
 

diff --git a/arangod/Aql/ExecutionBlock.h b/arangod/Aql/ExecutionBlock.h
@@ -140,7 +140,7 @@ class ExecutionBlock {
   [[nodiscard]] auto isDependencyInList(
       std::unordered_set<ExecutionBlock*> const& seenBlocks) const noexcept
       -> ExecutionBlock*;
-    
+
   [[nodiscard]] auto hasStoppedAsyncTasks() const noexcept -> bool;
 
  protected:
@@ -185,7 +185,8 @@ class ExecutionBlock {
   /// @brief if this is set, we are done, this is reset to false by execute()
   bool _done;
 
-  /// @brief if this is set, we have stopped async tasks, this is set to true by stopAsyncTasks()
+  /// @brief if this is set, we have stopped async tasks, this is set to true by
+  /// stopAsyncTasks()
   bool _stoppedAsyncTasks{false};
 
 #ifdef ARANGODB_ENABLE_MAINTAINER_MODE

diff --git a/arangod/Aql/ExecutionBlockImpl.tpp b/arangod/Aql/ExecutionBlockImpl.tpp
@@ -354,13 +354,30 @@ ExecutionBlockImpl<Executor>::~ExecutionBlockImpl() {
 
 template<class Executor>
 void ExecutionBlockImpl<Executor>::stopAsyncTasks() {
-  TRI_ASSERT(!_stoppedAsyncTasks) << "Someone already stopped async tasks for " << printBlockInfo();
+  TRI_ASSERT(!_stoppedAsyncTasks)
+      << "Someone already stopped async tasks for " << printBlockInfo();
   _stoppedAsyncTasks = true;
-  if (_prefetchTask && !_prefetchTask->isConsumed() &&
-      !_prefetchTask->tryClaim()) {
-    // some thread is still working on our prefetch task
-    // -> we need to wait for that task to finish first!
-    _prefetchTask->waitFor();
+  if (_prefetchTask) {
+    // Double use diagnostics:
+    uint64_t userCount = _numberOfUsers.fetch_add(1);
+    if (userCount > 0) {
+      _logStacktrace.store(true, std::memory_order_relaxed);
+      LOG_TOPIC("52637", WARN, Logger::AQL)
+          << "ALERT: Double use of ExecutionBlock detected, stacktrace:";
+      CrashHandler::logBacktrace();
+    }
+    auto guard = scopeGuard([&]() noexcept {
+      _numberOfUsers.fetch_sub(1);
+      if (_logStacktrace.load(std::memory_order_relaxed)) {
+        LOG_TOPIC("52638", WARN, Logger::AQL) << "ALERT: Found _logStacktrace:";
+        CrashHandler::logBacktrace();
+      }
+    });
+    if (!_prefetchTask->isConsumed() && !_prefetchTask->tryClaim()) {
+      // some thread is still working on our prefetch task
+      // -> we need to wait for that task to finish first!
+      _prefetchTask->waitFor();
+    }
   }
 }
 
@@ -1076,17 +1093,20 @@ auto ExecutionBlockImpl<Executor>::executeFetcher(ExecutionContext& ctx,
                 LOG_TOPIC("14d20", WARN, Logger::AQL)
                     << "[query#" << block->getQuery().id() << "] ALERT"
                     << block->printBlockInfo()
-                    << " was asked to stop async task. We still start one. This is an allowed rare race.";
+                    << " was asked to stop async task. We still start one. "
+                       "This is an allowed rare race.";
               }
 
-              auto stopGuard = ScopeGuard([block, hasStoppedAsyncTasks]() noexcept {
-                if (hasStoppedAsyncTasks) {
-                  LOG_TOPIC("14d21", WARN, Logger::AQL)
-                      << "[query#" << block->getQuery().id() << "] CLEAR ALERT"
-                      << block->printBlockInfo()
-                      << " We completed the task of the aforementioned race. All is fine.";
-                }
-              });
+              auto stopGuard =
+                  ScopeGuard([block, hasStoppedAsyncTasks]() noexcept {
+                    if (hasStoppedAsyncTasks) {
+                      LOG_TOPIC("14d21", WARN, Logger::AQL)
+                          << "[query#" << block->getQuery().id()
+                          << "] CLEAR ALERT" << block->printBlockInfo()
+                          << " We completed the task of the aforementioned "
+                             "race. All is fine.";
+                    }
+                  });
 
               TRI_IF_FAILURE("AsyncPrefetch::blocksDestroyedOutOfOrder") {
                 using namespace std::chrono_literals;