Make exceededMaxSpillLevelLimit_ in HashBuild and HashProbe tsan_atom…

…ic (#11165)

Summary:
Pull Request resolved: #11165

setupSpiller and canReclaim may be called concurrently in HashBuild and HashProbe
which leads to TSAN errors in our tests as the write to and read from
exceededMaxSpillLevelLimit_ respectively.

In practice reclaim checks for the case where exceededMaxSpillLevelLimit_ has been
updated since the last time canReclaim has been called and handles that case, so this
will not cause issues.

Marking exceededMaxSpillLevelLimit_ tsan_atomic to get the tests green.

Reviewed By: xiaoxmeng, tanjialiang

Differential Revision: D63853455

fbshipit-source-id: a633f92fe521004b7b99a26b8fea29647b7e29c3

velox/exec/HashBuild.h

@@ -217,7 +217,7 @@ class HashBuild final : public Operator {
 
   std::shared_ptr<HashJoinBridge> joinBridge_;
 
-  bool exceededMaxSpillLevelLimit_{false};
+  tsan_atomic<bool> exceededMaxSpillLevelLimit_{false};
 
   State state_{State::kRunning};
 

velox/exec/HashProbe.h

@@ -620,7 +620,7 @@ class HashProbe : public Operator {
   // Indicates if this hash probe has exceeded max spill limit which is not
   // allowed to spill. This is reset when hash probe operator starts to probe
   // the next previously spilled hash table partition.
-  bool exceededMaxSpillLevelLimit_{false};
+  tsan_atomic<bool> exceededMaxSpillLevelLimit_{false};
 
   // The partition bits used to spill the hash table.
   HashBitRange tableSpillHashBits_;