From decdfdae1082e93506242b308b8de2e81052f9ab Mon Sep 17 00:00:00 2001
From: Chun Han <116052805+MrPresent-Han@users.noreply.github.com>
Date: Tue, 17 Dec 2024 08:05:12 -0500
Subject: [PATCH] fix: growing-groupby-crush(#38533) (#38538)

related: #38533

Signed-off-by: MrPresent-Han <chun.han@gmail.com>
Co-authored-by: MrPresent-Han <chun.han@gmail.com>
---
 .../src/exec/operator/groupby/SearchGroupByOperator.h  |  8 +++++---
 internal/core/src/mmap/ChunkVector.h                   | 10 +++++++++-
 internal/core/src/segcore/ConcurrentVector.h           |  5 +++++
 3 files changed, 19 insertions(+), 4 deletions(-)

diff --git a/internal/core/src/exec/operator/groupby/SearchGroupByOperator.h b/internal/core/src/exec/operator/groupby/SearchGroupByOperator.h
index 80116de12be3a..481080902a4fa 100644
--- a/internal/core/src/exec/operator/groupby/SearchGroupByOperator.h
+++ b/internal/core/src/exec/operator/groupby/SearchGroupByOperator.h
@@ -51,10 +51,12 @@ class GrowingDataGetter : public DataGetter<T> {
     T
     Get(int64_t idx) const {
         if constexpr (std::is_same_v<std::string, T>) {
-            return T(growing_raw_data_->view_element(idx));
-        } else {
-            return growing_raw_data_->operator[](idx);
+            if (growing_raw_data_->is_mmap()) {
+                // when scalar data is mapped, it's needed to get the scalar data view and reconstruct string from the view
+                return T(growing_raw_data_->view_element(idx));
+            }
         }
+        return growing_raw_data_->operator[](idx);
     }
 };
 
diff --git a/internal/core/src/mmap/ChunkVector.h b/internal/core/src/mmap/ChunkVector.h
index cc485b238fddb..74341195c8edc 100644
--- a/internal/core/src/mmap/ChunkVector.h
+++ b/internal/core/src/mmap/ChunkVector.h
@@ -47,6 +47,9 @@ class ChunkVectorBase {
     virtual SpanBase
     get_span(int64_t chunk_id) = 0;
 
+    virtual bool
+    is_mmap() const = 0;
+
  protected:
     std::atomic<int64_t> counter_ = 0;
 };
@@ -107,7 +110,7 @@ class ThreadSafeChunkVector : public ChunkVectorBase<Type> {
     ChunkViewType<Type>
     view_element(int64_t chunk_id, int64_t chunk_offset) override {
         std::shared_lock<std::shared_mutex> lck(mutex_);
-        auto chunk = vec_[chunk_id];
+        auto& chunk = vec_[chunk_id];
         if constexpr (IsMmap) {
             return chunk.view(chunk_offset);
         } else if constexpr (std::is_same_v<std::string, Type>) {
@@ -184,6 +187,11 @@ class ThreadSafeChunkVector : public ChunkVectorBase<Type> {
         }
     }
 
+    bool
+    is_mmap() const override {
+        return mmap_descriptor_ != nullptr;
+    }
+
  private:
     mutable std::shared_mutex mutex_;
     storage::MmapChunkDescriptorPtr mmap_descriptor_ = nullptr;
diff --git a/internal/core/src/segcore/ConcurrentVector.h b/internal/core/src/segcore/ConcurrentVector.h
index 484ff7d293c04..67675e4515276 100644
--- a/internal/core/src/segcore/ConcurrentVector.h
+++ b/internal/core/src/segcore/ConcurrentVector.h
@@ -318,6 +318,11 @@ class ConcurrentVectorImpl : public VectorBase {
         chunks_ptr_->clear();
     }
 
+    bool
+    is_mmap() const {
+        return chunks_ptr_->is_mmap();
+    }
+
  private:
     void
     set_data(ssize_t element_offset,