[FLINK]Support processing time tumble window for nexmark q12

velox/exec/StreamingAggregation.cpp

@@ -210,10 +210,9 @@ void StreamingAggregation::assignGroups() {
   auto numInput = input_->size();
 
   inputGroups_.resize(numInput);
-
   // Look for the end of the last group.
   vector_size_t index = 0;
-  if (prevInput_) {
+  if (prevInput_ && numGroups_ > 0) {
     auto prevIndex = prevInput_->size() - 1;
     auto* prevGroup = groups_[numGroups_ - 1];
     for (; index < numInput; ++index) {
@@ -334,10 +333,9 @@ RowVectorPtr StreamingAggregation::getOutput() {
   evaluateAggregates();
 
   RowVectorPtr output;
-
-  if (numGroups_ > minOutputBatchSize_) {
+  if (numGroups_ >= minOutputBatchSize_) {
     output = createOutput(
-        std::min(numGroups_ - 1, static_cast<size_t>(maxOutputBatchSize_)));
+        std::min(numGroups_, static_cast<size_t>(maxOutputBatchSize_)));
   }
 
   prevInput_ = input_;

velox/experimental/stateful/GroupWindowAggregator.cpp

@@ -83,13 +83,12 @@ void GroupWindowAggregator::advance() {
   }
 
   // 1. Partition input by key
-  std::map<uint32_t, RowVectorPtr> keyToData = keySelector_->partition(input_);
+  std::map<int64_t, RowVectorPtr> keyToData = keySelector_->partition(input_);
   for (const auto& [key, keyedData] : keyToData) {
     // 2. Set the current key in the context
     windowContext_->setCurrentKey(key);
     // 3. Partition the keyed data by rowtime or processing time
-    std::map<uint32_t, RowVectorPtr> timestampToData =
-        sliceAssigner_->assignSliceEnd(keyedData);
+    std::map<int64_t, RowVectorPtr> timestampToData = sliceAssigner_->assignSliceEnd(keyedData);
     for (const auto& [timestamp, data] : timestampToData) {
       // 4. Assign data to window
       std::vector<TimeWindow> windows =
@@ -124,7 +123,7 @@ void GroupWindowAggregator::advance() {
 }
 
 void GroupWindowAggregator::onEventTime(
-    std::shared_ptr<TimerHeapInternalTimer<uint32_t, TimeWindow>> timer) {
+    std::shared_ptr<TimerHeapInternalTimer<int64_t, TimeWindow>> timer) {
   windowContext_->setCurrentKey(timer->key());
   if (triggerContext_->onEventTime(timer->ns(), timer->timestamp())) {
     // Fire the window and emit result.
@@ -187,7 +186,7 @@ void GroupWindowAggregator::emitWindowResult(uint32_t key, TimeWindow window) {
 
 GroupWindowAggregator::WindowTriggerContext::WindowTriggerContext(
     std::shared_ptr<WindowTrigger> trigger,
-    std::shared_ptr<InternalTimerService<uint32_t, TimeWindow>>
+    std::shared_ptr<InternalTimerService<int64_t, TimeWindow>>
         internalTimerService,
     int shiftTimeZone)
     : trigger_(std::move(trigger)),
@@ -236,7 +235,6 @@ int64_t GroupWindowAggregator::WindowTriggerContext::getCurrentWatermark() {
 void GroupWindowAggregator::WindowTriggerContext::registerProcessingTimeTimer(
     uint32_t key,
     TimeWindow window,
-
     int64_t time) {
   internalTimerService_->registerProcessingTimeTimer(key, window, time);
 }
@@ -287,7 +285,7 @@ void GroupWindowAggregator::WindowTriggerContext::mergePartitionedState(
 WindowContext::WindowContext(
     GroupWindowAggsHandler* windowAggregator,
     std::shared_ptr<ValueState<uint32_t, TimeWindow, RowVectorPtr>> windowState,
-    std::shared_ptr<InternalTimerService<uint32_t, TimeWindow>> timerService,
+    std::shared_ptr<InternalTimerService<int64_t, TimeWindow>> timerService,
     std::shared_ptr<TriggerContext> triggerContext,
     std::shared_ptr<StreamOperatorStateHandler> stateHandler,
     int shiftTimeZone,

velox/experimental/stateful/GroupWindowAggregator.h

@@ -22,7 +22,6 @@
 #include "velox/experimental/stateful/TimerHeapInternalTimer.h"
 #include "velox/experimental/stateful/Triggerable.h"
 #include "velox/experimental/stateful/window/SliceAssigner.h"
-#include "velox/experimental/stateful/window/WindowBuffer.h"
 #include "velox/experimental/stateful/window/WindowProcessFunction.h"
 #include "velox/experimental/stateful/window/WindowTrigger.h"
 
@@ -33,7 +32,7 @@ class WindowContext;
 /// This class is related to AggregateWindowOperator in Flink.
 /// It's for group window aggregator. Rename it to GroupWindowAggregator.
 class GroupWindowAggregator : public StatefulOperator,
-                              public Triggerable<uint32_t, TimeWindow> {
+                              public Triggerable<int64_t, TimeWindow> {
  public:
   GroupWindowAggregator(
       std::unique_ptr<GroupWindowAggsHandler> windowAggerator,
@@ -61,7 +60,7 @@ class GroupWindowAggregator : public StatefulOperator,
     return "GroupWindowAggregator";
   }
 
-  void onEventTime(std::shared_ptr<TimerHeapInternalTimer<uint32_t, TimeWindow>>
+  void onEventTime(std::shared_ptr<TimerHeapInternalTimer<int64_t, TimeWindow>>
                        timer) override;
 
  private:
@@ -71,7 +70,7 @@ class GroupWindowAggregator : public StatefulOperator,
    public:
     WindowTriggerContext(
         std::shared_ptr<WindowTrigger> trigger,
-        std::shared_ptr<InternalTimerService<uint32_t, TimeWindow>>
+        std::shared_ptr<InternalTimerService<int64_t, TimeWindow>>
             internalTimerService,
         int shiftTimeZone);
 
@@ -122,7 +121,7 @@ class GroupWindowAggregator : public StatefulOperator,
 
    private:
     std::shared_ptr<WindowTrigger> trigger_;
-    std::shared_ptr<InternalTimerService<uint32_t, TimeWindow>>
+    std::shared_ptr<InternalTimerService<int64_t, TimeWindow>>
         internalTimerService_;
     int shiftTimeZone_;
     TimeWindow window;
@@ -145,7 +144,7 @@ class GroupWindowAggregator : public StatefulOperator,
 
   RowVectorPtr input_;
   std::shared_ptr<ValueState<uint32_t, TimeWindow, RowVectorPtr>> windowState_;
-  std::shared_ptr<InternalTimerService<uint32_t, TimeWindow>>
+  std::shared_ptr<InternalTimerService<int64_t, TimeWindow>>
       windowTimerService_;
   std::shared_ptr<TriggerContext> triggerContext_;
   std::shared_ptr<WindowContext> windowContext_;
@@ -158,7 +157,7 @@ class WindowContext : public FunctionContext<TimeWindow> {
       GroupWindowAggsHandler* windowAggregator,
       std::shared_ptr<ValueState<uint32_t, TimeWindow, RowVectorPtr>>
           windowState,
-      std::shared_ptr<InternalTimerService<uint32_t, TimeWindow>> timerService,
+      std::shared_ptr<InternalTimerService<int64_t, TimeWindow>> timerService,
       std::shared_ptr<TriggerContext> triggerContext,
       std::shared_ptr<StreamOperatorStateHandler> stateHandler,
       int shiftTimeZone,
@@ -201,7 +200,7 @@ class WindowContext : public FunctionContext<TimeWindow> {
   std::shared_ptr<ValueState<uint32_t, TimeWindow, RowVectorPtr>> windowState_;
   std::shared_ptr<ValueState<uint32_t, TimeWindow, RowVectorPtr>>
       previousWindowState_;
-  std::shared_ptr<InternalTimerService<uint32_t, TimeWindow>> timerService_;
+  std::shared_ptr<InternalTimerService<int64_t, TimeWindow>> timerService_;
   std::shared_ptr<TriggerContext> triggerContext_;
   std::shared_ptr<StreamOperatorStateHandler> stateHandler_;
   int shiftTimeZone_;