Update upstream source from tag 'upstream/6.0.0'master

Update to upstream version '6.0.0'
with Debian dir 5604a80ec1c96ca76f25f40d78e6ef855abec322

1 files changed, 37 insertions, 41 deletions

diff --git a/src/mongo/db/pipeline/window_function/partition_iterator_test.cpp b/src/mongo/db/pipeline/window_function/partition_iterator_test.cpp
index 8b7a6beab76..cb6b5bcf1bf 100644
--- a/src/mongo/db/pipeline/window_function/partition_iterator_test.cpp
+++ b/src/mongo/db/pipeline/window_function/partition_iterator_test.cpp
@@ -499,20 +499,22 @@ TEST_F(PartitionIteratorTest, MemoryUsageAccountsForDocumentIteratorCache) {
     const auto mock = DocumentSourceMock::createForTest(docs, getExpCtx());
 
     [[maybe_unused]] auto accessor = makeDefaultAccessor(mock, boost::none);
-    size_t initialDocSize = docs[0].getDocument().getCurrentApproximateSize();
+    size_t initialDocSize = docs[0].getDocument().getApproximateSize();
 
-    // Pull in the first document, and verify the reported size of the iterator is roughly the size
-    // of the document.
+    // Pull in the first document, and verify the reported size of the iterator is roughly double
+    // the size of the document. The size of the iterator is double the size of the document because
+    // we greedily fill the cache, so each internal document in memory stores two copies of
+    // largeStr.
     ASSERT_DOCUMENT_EQ(*_iter->current(), docs[0].getDocument());
-    ASSERT_GT(_iter->getApproximateSize(), initialDocSize);
-    ASSERT_LT(_iter->getApproximateSize(), initialDocSize + 500);
+    ASSERT_GT(_iter->getApproximateSize(), initialDocSize * 2);
+    ASSERT_LT(_iter->getApproximateSize(), initialDocSize * 2 + 500);
 
     // Pull in the second document. Both docs remain in the cache so the reported memory should
     // include both.
     advance();
     ASSERT_DOCUMENT_EQ(*_iter->current(), docs[1].getDocument());
-    ASSERT_GT(_iter->getApproximateSize(), initialDocSize);
-    ASSERT_LT(_iter->getApproximateSize(), initialDocSize * 2 + 500);
+    ASSERT_GT(_iter->getApproximateSize(), initialDocSize * 2 * 2);
+    ASSERT_LT(_iter->getApproximateSize(), initialDocSize * 2 * 2 + 500);
 }
 
 TEST_F(PartitionIteratorTest, MemoryUsageAccountsForArraysInDocumentIteratorCache) {
@@ -523,21 +525,21 @@ TEST_F(PartitionIteratorTest, MemoryUsageAccountsForArraysInDocumentIteratorCach
     const auto mock = DocumentSourceMock::createForTest(docs, getExpCtx());
 
     [[maybe_unused]] auto accessor = makeDefaultAccessor(mock, boost::none);
-    size_t initialDocSize = docs[0].getDocument().getCurrentApproximateSize();
+    size_t initialDocSize = docs[0].getDocument().getApproximateSize();
 
-    // Pull in the first document, and verify the reported size of the iterator is roughly the size
-    // of the document. The reason we can't use EQ is that for memory tracking we call shred() so
-    // that the document cache will not increase when fields are accessed
+    // Pull in the first document, and verify the reported size of the iterator is roughly
+    // triple the size of the document. The reason for this is that 'largeStr' is cached twice; once
+    // for the 'arr' element and once for the nested 'subObj' element.
     ASSERT_DOCUMENT_EQ(*_iter->current(), docs[0].getDocument());
-    ASSERT_GT(_iter->getApproximateSize(), initialDocSize);
-    ASSERT_LT(_iter->getApproximateSize(), initialDocSize + 1024);
+    ASSERT_GT(_iter->getApproximateSize(), initialDocSize * 3);
+    ASSERT_LT(_iter->getApproximateSize(), initialDocSize * 3 + 1024);
 
     // Pull in the second document. Both docs remain in the cache so the reported memory should
     // include both.
     advance();
     ASSERT_DOCUMENT_EQ(*_iter->current(), docs[1].getDocument());
-    ASSERT_GT(_iter->getApproximateSize(), initialDocSize * 2);
-    ASSERT_LT(_iter->getApproximateSize(), initialDocSize * 2 + 1024);
+    ASSERT_GT(_iter->getApproximateSize(), (initialDocSize * 3) * 2);
+    ASSERT_LT(_iter->getApproximateSize(), (initialDocSize * 3) * 2 + 1024);
 }
 
 TEST_F(PartitionIteratorTest, MemoryUsageAccountsForNestedArraysInDocumentIteratorCache) {
@@ -548,21 +550,21 @@ TEST_F(PartitionIteratorTest, MemoryUsageAccountsForNestedArraysInDocumentIterat
     const auto mock = DocumentSourceMock::createForTest(docs, getExpCtx());
 
     [[maybe_unused]] auto accessor = makeDefaultAccessor(mock, boost::none);
-    size_t initialDocSize = docs[0].getDocument().getCurrentApproximateSize();
+    size_t initialDocSize = docs[0].getDocument().getApproximateSize();
 
-    // Pull in the first document, and verify the reported size of the iterator is roughly the size
-    // of the document. The reason we can't use EQ is that for memory tracking we call shred() so
-    // that the document cache will not increase when fields are accessed
+    // Pull in the first document, and verify the reported size of the iterator is roughly
+    // triple the size of the document. The reason for this is that 'largeStr' is cached twice; once
+    // for the 'arr' element and once for the nested 'subObj' element.
     ASSERT_DOCUMENT_EQ(*_iter->current(), docs[0].getDocument());
-    ASSERT_GT(_iter->getApproximateSize(), initialDocSize);
-    ASSERT_LT(_iter->getApproximateSize(), initialDocSize + 1024);
+    ASSERT_GT(_iter->getApproximateSize(), initialDocSize * 3);
+    ASSERT_LT(_iter->getApproximateSize(), initialDocSize * 3 + 1024);
 
     // Pull in the second document. Both docs remain in the cache so the reported memory should
     // include both.
     advance();
     ASSERT_DOCUMENT_EQ(*_iter->current(), docs[1].getDocument());
-    ASSERT_GT(_iter->getApproximateSize(), initialDocSize * 2);
-    ASSERT_LT(_iter->getApproximateSize(), initialDocSize * 2 + 1024);
+    ASSERT_GT(_iter->getApproximateSize(), (initialDocSize * 3) * 2);
+    ASSERT_LT(_iter->getApproximateSize(), (initialDocSize * 3) * 2 + 1024);
 }
 
 TEST_F(PartitionIteratorTest, MemoryUsageAccountsForNestedObjInDocumentIteratorCache) {
@@ -573,12 +575,13 @@ TEST_F(PartitionIteratorTest, MemoryUsageAccountsForNestedObjInDocumentIteratorC
     const auto mock = DocumentSourceMock::createForTest(docs, getExpCtx());
 
     [[maybe_unused]] auto accessor = makeDefaultAccessor(mock, boost::none);
-    size_t initialDocSize = docs[0].getDocument().getCurrentApproximateSize();
+    size_t initialDocSize = docs[0].getDocument().getApproximateSize();
 
-    // Pull in the first document, and verify the reported size.
+    // Pull in the first document, and verify the reported size. TODO SERVER-57011: The approximate
+    // size should not double count the nested strings.
     ASSERT_DOCUMENT_EQ(*_iter->current(), docs[0].getDocument());
-    ASSERT_GT(_iter->getApproximateSize(), initialDocSize);
-    ASSERT_LT(_iter->getApproximateSize(), initialDocSize * 2);
+    ASSERT_GT(_iter->getApproximateSize(), initialDocSize * 3);
+    ASSERT_LT(_iter->getApproximateSize(), initialDocSize * 4);
 }
 
 TEST_F(PartitionIteratorTest, MemoryUsageAccountsForReleasedDocuments) {
@@ -589,27 +592,20 @@ TEST_F(PartitionIteratorTest, MemoryUsageAccountsForReleasedDocuments) {
     const auto mock = DocumentSourceMock::createForTest(docs, getExpCtx());
 
     auto accessor = makeDefaultAccessor(mock, boost::none);
-    size_t initialDocSize = docs[0].getDocument().getCurrentApproximateSize();
+    size_t initialDocSize = docs[0].getDocument().getApproximateSize();
 
-    // Pull in the first document, and verify the reported size of the iterator is roughly the size
-    // of the document.
+    // Pull in the first document, and verify the reported size of the iterator is roughly double
+    // the size of the document.
     ASSERT_DOCUMENT_EQ(*accessor[0], docs[0].getDocument());
-    ASSERT_GT(_iter->getApproximateSize(), initialDocSize);
-    ASSERT_LT(_iter->getApproximateSize(), initialDocSize + 1024);
-
-    // Read the field so that it is coppied into the cache. This will make the document bigger but
-    // shouldn't affect memory tracking.
-    auto iterSizeBeforeAccess = _iter->getApproximateSize();
-    docs[0].getDocument()["a"];
-    ASSERT_GT(docs[0].getDocument().getCurrentApproximateSize(), initialDocSize);
-    ASSERT_EQ(_iter->getApproximateSize(), iterSizeBeforeAccess);
+    ASSERT_GT(_iter->getApproximateSize(), initialDocSize * 2);
+    ASSERT_LT(_iter->getApproximateSize(), initialDocSize * 2 + 1024);
 
     // The accessor will have marked the first document as expired, and thus freed on the next call
     // to advance().
     advance();
     ASSERT_DOCUMENT_EQ(*_iter->current(), docs[1].getDocument());
-    ASSERT_GT(_iter->getApproximateSize(), initialDocSize);
-    ASSERT_LT(_iter->getApproximateSize(), initialDocSize + 1024);
+    ASSERT_GT(_iter->getApproximateSize(), initialDocSize * 2);
+    ASSERT_LT(_iter->getApproximateSize(), initialDocSize * 2 + 1024);
 }
 
 TEST_F(PartitionIteratorTest, ManualPolicy) {