让我们建立一个简单的数据库8

第 8 部分 - B-树叶节点的格式

　　我们将把表格的存储结构从未排序的行数组改为 B-树。这是一项相当重大的改动，需要通过多篇文章来逐步实现。到本文结束时，我们将定义叶节点的布局，并支持将键值对插入到单节点树中。但在开始之前，让我们先回顾一下为何要改用树结构。

其他表格格式

　　在当前的格式下，每页仅存储行（不包含元数据），因此空间利用率相当高。插入操作也很快，因为我们只需在末尾追加数据。然而，查找特定行只能通过扫描整个表来实现。如果要删除一行，我们必须将后续的所有行向后移动以填补空缺。

　　如果我们将表存储为数组，但保持按 ID 排序的行，就可以使用二分查找来查找特定的 ID。然而，插入操作会很慢，因为我们必须移动大量行来腾出空间。

　　因此，我们选择采用树结构。树中的每个节点可能包含不同数量的行，所以我们需要在每个节点中存储一些信息来记录其包含的行数。此外，所有不存储任何行的内部节点也会带来存储开销。虽然数据库文件会变大，但我们换来了快速的插入、删除和查找操作。

节点头格式

　　叶节点和内部节点的布局不同。让我们定义一个枚举类型来记录节点类型：

+typedef enum { NODE_INTERNAL, NODE_LEAF } NodeType;

　　每个节点对应一个页面。内部节点通过存储其子节点的所在页码来指向这些子节点。B树向分页器查询特定页码，并返回指向页面缓存的指针。页面按页码顺序依次存储在数据库文件中。

　　节点需要在页面开头的头部存储一些元数据。每个节点都会存储其节点类型、是否为根节点，以及指向其父节点的指针（以便查找节点的兄弟节点）。我为每个头部字段的大小和偏移量定义了常量：

+/*
+ * Common Node Header Layout
+ */
+const uint32_t NODE_TYPE_SIZE = sizeof(uint8_t);
+const uint32_t NODE_TYPE_OFFSET = 0;
+const uint32_t IS_ROOT_SIZE = sizeof(uint8_t);
+const uint32_t IS_ROOT_OFFSET = NODE_TYPE_SIZE;
+const uint32_t PARENT_POINTER_SIZE = sizeof(uint32_t);
+const uint32_t PARENT_POINTER_OFFSET = IS_ROOT_OFFSET + IS_ROOT_SIZE;
+const uint8_t COMMON_NODE_HEADER_SIZE =
+    NODE_TYPE_SIZE + IS_ROOT_SIZE + PARENT_POINTER_SIZE;

叶节点的格式

　　除了这些常见的头部字段外，叶节点还需要存储其包含的“单元”数量。一个单元即为一个键值对。

+/*
+ * Leaf Node Header Layout
+ */
+const uint32_t LEAF_NODE_NUM_CELLS_SIZE = sizeof(uint32_t);
+const uint32_t LEAF_NODE_NUM_CELLS_OFFSET = COMMON_NODE_HEADER_SIZE;
+const uint32_t LEAF_NODE_HEADER_SIZE =
+    COMMON_NODE_HEADER_SIZE + LEAF_NODE_NUM_CELLS_SIZE;

 　　叶节点的主体是一个单元数组。每个单元由一个键和一个值（一个序列化行）组成。

+/*
+ * Leaf Node Body Layout
+ */
+const uint32_t LEAF_NODE_KEY_SIZE = sizeof(uint32_t);
+const uint32_t LEAF_NODE_KEY_OFFSET = 0;
+const uint32_t LEAF_NODE_VALUE_SIZE = ROW_SIZE;
+const uint32_t LEAF_NODE_VALUE_OFFSET =
+    LEAF_NODE_KEY_OFFSET + LEAF_NODE_KEY_SIZE;
+const uint32_t LEAF_NODE_CELL_SIZE = LEAF_NODE_KEY_SIZE + LEAF_NODE_VALUE_SIZE;
+const uint32_t LEAF_NODE_SPACE_FOR_CELLS = PAGE_SIZE - LEAF_NODE_HEADER_SIZE;
+const uint32_t LEAF_NODE_MAX_CELLS =
+    LEAF_NODE_SPACE_FOR_CELLS / LEAF_NODE_CELL_SIZE;

　　根据这些常量，叶节点的布局目前如下所示：

叶子节点格式

　　在头部为每个布尔值占用整整一个字节，虽然在空间利用率上稍显低效，但这使得编写代码访问这些值变得更加容易。

　　另外请注意，末尾存在一些浪费的空间。我们在头部之后尽可能多地存储单元格，但剩余的空间无法容纳一个完整的单元格。我们将其留空，以避免单元格被分割到不同的节点中。

访问叶节点的字段

　　访问键、值和元数据的代码都涉及使用我们刚刚定义的常量进行的指针运算。

+uint32_t* leaf_node_num_cells(void* node) {
+  return node + LEAF_NODE_NUM_CELLS_OFFSET;
+}
+
+void* leaf_node_cell(void* node, uint32_t cell_num) {
+  return node + LEAF_NODE_HEADER_SIZE + cell_num * LEAF_NODE_CELL_SIZE;
+}
+
+uint32_t* leaf_node_key(void* node, uint32_t cell_num) {
+  return leaf_node_cell(node, cell_num);
+}
+
+void* leaf_node_value(void* node, uint32_t cell_num) {
+  return leaf_node_cell(node, cell_num) + LEAF_NODE_KEY_SIZE;
+}
+
+void initialize_leaf_node(void* node) { *leaf_node_num_cells(node) = 0; }
+

　　这些方法返回指向该值的指针，因此既可作为读取器（getter），也可作为设置器（setter）使用。

分页器和表格对象的更改

　　每个节点都将占用整整一页空间，即使该页并未填满。这意味着我们的分页器不再需要支持对部分页面的读写操作。

-void pager_flush(Pager* pager, uint32_t page_num, uint32_t size) {
+void pager_flush(Pager* pager, uint32_t page_num) {
   if (pager->pages[page_num] == NULL) {
     printf("Tried to flush null page\n");
     exit(EXIT_FAILURE);
@@ -242,7 +337,7 @@ void pager_flush(Pager* pager, uint32_t page_num, uint32_t size) {
   }
 
   ssize_t bytes_written =
-      write(pager->file_descriptor, pager->pages[page_num], size);
+      write(pager->file_descriptor, pager->pages[page_num], PAGE_SIZE);
 
   if (bytes_written == -1) {
     printf("Error writing: %d\n", errno);

 void db_close(Table* table) {
   Pager* pager = table->pager;
-  uint32_t num_full_pages = table->num_rows / ROWS_PER_PAGE;
 
-  for (uint32_t i = 0; i < num_full_pages; i++) {
+  for (uint32_t i = 0; i < pager->num_pages; i++) {
     if (pager->pages[i] == NULL) {
       continue;
     }
-    pager_flush(pager, i, PAGE_SIZE);
+    pager_flush(pager, i);
     free(pager->pages[i]);
     pager->pages[i] = NULL;
   }
 
-  // There may be a partial page to write to the end of the file
-  // This should not be needed after we switch to a B-tree
-  uint32_t num_additional_rows = table->num_rows % ROWS_PER_PAGE;
-  if (num_additional_rows > 0) {
-    uint32_t page_num = num_full_pages;
-    if (pager->pages[page_num] != NULL) {
-      pager_flush(pager, page_num, num_additional_rows * ROW_SIZE);
-      free(pager->pages[page_num]);
-      pager->pages[page_num] = NULL;
-    }
-  }
-
   int result = close(pager->file_descriptor);
   if (result == -1) {
     printf("Error closing db file.\n");

　　现在，将页数存储在数据库中比存储行数更有意义。页数应与分页器对象相关联，而非与表相关联，因为这是数据库使用的页数，而非特定表的页数。B树通过其根节点的页号进行标识，因此表对象需要记录该页号。

 const uint32_t PAGE_SIZE = 4096;
 const uint32_t TABLE_MAX_PAGES = 100;
-const uint32_t ROWS_PER_PAGE = PAGE_SIZE / ROW_SIZE;
-const uint32_t TABLE_MAX_ROWS = ROWS_PER_PAGE * TABLE_MAX_PAGES;
 
 typedef struct {
   int file_descriptor;
   uint32_t file_length;
+  uint32_t num_pages;
   void* pages[TABLE_MAX_PAGES];
 } Pager;
 
 typedef struct {
   Pager* pager;
-  uint32_t num_rows;
+  uint32_t root_page_num;
 } Table;

@@ -127,6 +200,10 @@ void* get_page(Pager* pager, uint32_t page_num) {
     }
 
     pager->pages[page_num] = page;
+
+    if (page_num >= pager->num_pages) {
+      pager->num_pages = page_num + 1;
+    }
   }
 
   return pager->pages[page_num];

@@ -184,6 +269,12 @@ Pager* pager_open(const char* filename) {
   Pager* pager = malloc(sizeof(Pager));
   pager->file_descriptor = fd;
   pager->file_length = file_length;
+  pager->num_pages = (file_length / PAGE_SIZE);
+
+  if (file_length % PAGE_SIZE != 0) {
+    printf("Db file is not a whole number of pages. Corrupt file.\n");
+    exit(EXIT_FAILURE);
+  }
 
   for (uint32_t i = 0; i < TABLE_MAX_PAGES; i++) {
     pager->pages[i] = NULL;

游标对象的更改

　　游标代表表中的一个位置。当我们的表还是一个简单的行数组时，我们只需知道行号就能访问该行。现在它变成了一棵树，我们通过节点的页号以及该节点内的单元格编号来确定位置。

 typedef struct {
   Table* table;
-  uint32_t row_num;
+  uint32_t page_num;
+  uint32_t cell_num;
   bool end_of_table;  // Indicates a position one past the last element
 } Cursor;

 Cursor* table_start(Table* table) {
   Cursor* cursor = malloc(sizeof(Cursor));
   cursor->table = table;
-  cursor->row_num = 0;
-  cursor->end_of_table = (table->num_rows == 0);
+  cursor->page_num = table->root_page_num;
+  cursor->cell_num = 0;
+
+  void* root_node = get_page(table->pager, table->root_page_num);
+  uint32_t num_cells = *leaf_node_num_cells(root_node);
+  cursor->end_of_table = (num_cells == 0);
 
   return cursor;
 }

 Cursor* table_end(Table* table) {
   Cursor* cursor = malloc(sizeof(Cursor));
   cursor->table = table;
-  cursor->row_num = table->num_rows;
+  cursor->page_num = table->root_page_num;
+
+  void* root_node = get_page(table->pager, table->root_page_num);
+  uint32_t num_cells = *leaf_node_num_cells(root_node);
+  cursor->cell_num = num_cells;
   cursor->end_of_table = true;
 
   return cursor;
 }

 void* cursor_value(Cursor* cursor) {
-  uint32_t row_num = cursor->row_num;
-  uint32_t page_num = row_num / ROWS_PER_PAGE;
+  uint32_t page_num = cursor->page_num;
   void* page = get_page(cursor->table->pager, page_num);
-  uint32_t row_offset = row_num % ROWS_PER_PAGE;
-  uint32_t byte_offset = row_offset * ROW_SIZE;
-  return page + byte_offset;
+  return leaf_node_value(page, cursor->cell_num);
 }

 void cursor_advance(Cursor* cursor) {
-  cursor->row_num += 1;
-  if (cursor->row_num >= cursor->table->num_rows) {
+  uint32_t page_num = cursor->page_num;
+  void* node = get_page(cursor->table->pager, page_num);
+
+  cursor->cell_num += 1;
+  if (cursor->cell_num >= (*leaf_node_num_cells(node))) {
     cursor->end_of_table = true;
   }
 }

插入到叶节点

　　在本文中，我们只实现足够的功能来构建一棵单节点树。回顾上一篇文章，树最初是一个空的叶节点：

空的 B-树

　　可以不断添加键值对，直到叶节点已满：

单节点 B 树

　　当我们首次打开数据库时，数据库文件是空的，因此我们将第 0 页初始化为一个空的叶节点（即根节点）：

 Table* db_open(const char* filename) {
   Pager* pager = pager_open(filename);
-  uint32_t num_rows = pager->file_length / ROW_SIZE;
 
   Table* table = malloc(sizeof(Table));
   table->pager = pager;
-  table->num_rows = num_rows;
+  table->root_page_num = 0;
+
+  if (pager->num_pages == 0) {
+    // New database file. Initialize page 0 as leaf node.
+    void* root_node = get_page(pager, 0);
+    initialize_leaf_node(root_node);
+  }
 
   return table;
 }

　　接下来，我们将编写一个函数，用于将键值对插入到叶节点中。该函数将接受一个游标作为输入，以表示应插入键值对的位置。

+void leaf_node_insert(Cursor* cursor, uint32_t key, Row* value) {
+  void* node = get_page(cursor->table->pager, cursor->page_num);
+
+  uint32_t num_cells = *leaf_node_num_cells(node);
+  if (num_cells >= LEAF_NODE_MAX_CELLS) {
+    // Node full
+    printf("Need to implement splitting a leaf node.\n");
+    exit(EXIT_FAILURE);
+  }
+
+  if (cursor->cell_num < num_cells) {
+    // Make room for new cell
+    for (uint32_t i = num_cells; i > cursor->cell_num; i--) {
+      memcpy(leaf_node_cell(node, i), leaf_node_cell(node, i - 1),
+             LEAF_NODE_CELL_SIZE);
+    }
+  }
+
+  *(leaf_node_num_cells(node)) += 1;
+  *(leaf_node_key(node, cursor->cell_num)) = key;
+  serialize_row(value, leaf_node_value(node, cursor->cell_num));
+}
+

　　我们尚未实现节点拆分，因此如果节点已满，系统会报错。接下来，我们将单元格向右移动一个位置，为新单元格腾出空间。然后，我们将新的键值对写入该空位。

　　由于我们假设树中只有一个节点，因此我们的 execute_insert() 函数只需调用这个辅助方法：

 ExecuteResult execute_insert(Statement* statement, Table* table) {
-  if (table->num_rows >= TABLE_MAX_ROWS) {
+  void* node = get_page(table->pager, table->root_page_num);
+  if ((*leaf_node_num_cells(node) >= LEAF_NODE_MAX_CELLS)) {
     return EXECUTE_TABLE_FULL;
   }
 
   Row* row_to_insert = &(statement->row_to_insert);
   Cursor* cursor = table_end(table);
 
-  serialize_row(row_to_insert, cursor_value(cursor));
-  table->num_rows += 1;
+  leaf_node_insert(cursor, row_to_insert->id, row_to_insert);
 
   free(cursor);

　　经过这些修改，我们的数据库应该能像以前一样正常工作了！不过现在它会更早地返回“表已满”错误，因为我们目前还无法拆分根节点。

　　叶节点最多能容纳多少行？

打印常量的命令

　　我正在添加一个新的元命令，用于打印出几个有用的常量。

+void print_constants() {
+  printf("ROW_SIZE: %d\n", ROW_SIZE);
+  printf("COMMON_NODE_HEADER_SIZE: %d\n", COMMON_NODE_HEADER_SIZE);
+  printf("LEAF_NODE_HEADER_SIZE: %d\n", LEAF_NODE_HEADER_SIZE);
+  printf("LEAF_NODE_CELL_SIZE: %d\n", LEAF_NODE_CELL_SIZE);
+  printf("LEAF_NODE_SPACE_FOR_CELLS: %d\n", LEAF_NODE_SPACE_FOR_CELLS);
+  printf("LEAF_NODE_MAX_CELLS: %d\n", LEAF_NODE_MAX_CELLS);
+}
+
@@ -294,6 +376,14 @@ MetaCommandResult do_meta_command(InputBuffer* input_buffer, Table* table) {
   if (strcmp(input_buffer->buffer, ".exit") == 0) {
     db_close(table);
     exit(EXIT_SUCCESS);
+  } else if (strcmp(input_buffer->buffer, ".constants") == 0) {
+    printf("Constants:\n");
+    print_constants();
+    return META_COMMAND_SUCCESS;
   } else {
     return META_COMMAND_UNRECOGNIZED_COMMAND;
   }

　　我还要添加一个测试，以便在这些常量发生变化时收到提醒：

+  it 'prints constants' do
+    script = [
+      ".constants",
+      ".exit",
+    ]
+    result = run_script(script)
+
+    expect(result).to match_array([
+      "db > Constants:",
+      "ROW_SIZE: 293",
+      "COMMON_NODE_HEADER_SIZE: 6",
+      "LEAF_NODE_HEADER_SIZE: 10",
+      "LEAF_NODE_CELL_SIZE: 297",
+      "LEAF_NODE_SPACE_FOR_CELLS: 4086",
+      "LEAF_NODE_MAX_CELLS: 13",
+      "db > ",
+    ])
+  end

　　所以，我们的表现在可以容纳 13 行了！

树结构可视化

　　为了辅助调试和可视化，我还添加了一个 meta 命令，用于输出 B 树的结构图。

+void print_leaf_node(void* node) {
+  uint32_t num_cells = *leaf_node_num_cells(node);
+  printf("leaf (size %d)\n", num_cells);
+  for (uint32_t i = 0; i < num_cells; i++) {
+    uint32_t key = *leaf_node_key(node, i);
+    printf("  - %d : %d\n", i, key);
+  }
+}
+

@@ -294,6 +376,14 @@ MetaCommandResult do_meta_command(InputBuffer* input_buffer, Table* table) {
   if (strcmp(input_buffer->buffer, ".exit") == 0) {
     db_close(table);
     exit(EXIT_SUCCESS);
+  } else if (strcmp(input_buffer->buffer, ".btree") == 0) {
+    printf("Tree:\n");
+    print_leaf_node(get_page(table->pager, 0));
+    return META_COMMAND_SUCCESS;
   } else if (strcmp(input_buffer->buffer, ".constants") == 0) {
     printf("Constants:\n");
     print_constants();
     return META_COMMAND_SUCCESS;
   } else {
     return META_COMMAND_UNRECOGNIZED_COMMAND;
   }

　　以及一个测试

+  it 'allows printing out the structure of a one-node btree' do
+    script = [3, 1, 2].map do |i|
+      "insert #{i} user#{i} person#{i}@example.com"
+    end
+    script << ".btree"
+    script << ".exit"
+    result = run_script(script)
+
+    expect(result).to match_array([
+      "db > Executed.",
+      "db > Executed.",
+      "db > Executed.",
+      "db > Tree:",
+      "leaf (size 3)",
+      "  - 0 : 3",
+      "  - 1 : 1",
+      "  - 2 : 2",
+      "db > "
+    ])
+  end

　　哎呀，我们仍然没有按排序顺序存储行。你会发现，execute_insert() 函数会将数据插入到 table_end() 返回位置处的叶节点中。因此，行仍按插入时的顺序存储，就像之前一样。

下一次

　　这一切看起来可能像是倒退了一步。现在我们的数据库存储的行数比以前少了，而且行仍然是以未排序的顺序存储的。但正如我开头所说，这是一项重大变革，将其分解为可管理的步骤非常重要。

　　下次，我们将实现通过主键查找记录的功能，并开始以排序的顺序存储行。

完整差异

@@ -62,29 +62,101 @@ const uint32_t ROW_SIZE = ID_SIZE + USERNAME_SIZE + EMAIL_SIZE;

 const uint32_t PAGE_SIZE = 4096;
 #define TABLE_MAX_PAGES 100
-const uint32_t ROWS_PER_PAGE = PAGE_SIZE / ROW_SIZE;
-const uint32_t TABLE_MAX_ROWS = ROWS_PER_PAGE * TABLE_MAX_PAGES;
 
 typedef struct {
   int file_descriptor;
   uint32_t file_length;
+  uint32_t num_pages;
   void* pages[TABLE_MAX_PAGES];
 } Pager;
 
 typedef struct {
   Pager* pager;
-  uint32_t num_rows;
+  uint32_t root_page_num;
 } Table;
 
 typedef struct {
   Table* table;
-  uint32_t row_num;
+  uint32_t page_num;
+  uint32_t cell_num;
   bool end_of_table;  // Indicates a position one past the last element
 } Cursor;

+typedef enum { NODE_INTERNAL, NODE_LEAF } NodeType;
+
+/*
+ * Common Node Header Layout
+ */
+const uint32_t NODE_TYPE_SIZE = sizeof(uint8_t);
+const uint32_t NODE_TYPE_OFFSET = 0;
+const uint32_t IS_ROOT_SIZE = sizeof(uint8_t);
+const uint32_t IS_ROOT_OFFSET = NODE_TYPE_SIZE;
+const uint32_t PARENT_POINTER_SIZE = sizeof(uint32_t);
+const uint32_t PARENT_POINTER_OFFSET = IS_ROOT_OFFSET + IS_ROOT_SIZE;
+const uint8_t COMMON_NODE_HEADER_SIZE =
+    NODE_TYPE_SIZE + IS_ROOT_SIZE + PARENT_POINTER_SIZE;
+
+/*
+ * Leaf Node Header Layout
+ */
+const uint32_t LEAF_NODE_NUM_CELLS_SIZE = sizeof(uint32_t);
+const uint32_t LEAF_NODE_NUM_CELLS_OFFSET = COMMON_NODE_HEADER_SIZE;
+const uint32_t LEAF_NODE_HEADER_SIZE =
+    COMMON_NODE_HEADER_SIZE + LEAF_NODE_NUM_CELLS_SIZE;
+
+/*
+ * Leaf Node Body Layout
+ */
+const uint32_t LEAF_NODE_KEY_SIZE = sizeof(uint32_t);
+const uint32_t LEAF_NODE_KEY_OFFSET = 0;
+const uint32_t LEAF_NODE_VALUE_SIZE = ROW_SIZE;
+const uint32_t LEAF_NODE_VALUE_OFFSET =
+    LEAF_NODE_KEY_OFFSET + LEAF_NODE_KEY_SIZE;
+const uint32_t LEAF_NODE_CELL_SIZE = LEAF_NODE_KEY_SIZE + LEAF_NODE_VALUE_SIZE;
+const uint32_t LEAF_NODE_SPACE_FOR_CELLS = PAGE_SIZE - LEAF_NODE_HEADER_SIZE;
+const uint32_t LEAF_NODE_MAX_CELLS =
+    LEAF_NODE_SPACE_FOR_CELLS / LEAF_NODE_CELL_SIZE;
+
+uint32_t* leaf_node_num_cells(void* node) {
+  return node + LEAF_NODE_NUM_CELLS_OFFSET;
+}
+
+void* leaf_node_cell(void* node, uint32_t cell_num) {
+  return node + LEAF_NODE_HEADER_SIZE + cell_num * LEAF_NODE_CELL_SIZE;
+}
+
+uint32_t* leaf_node_key(void* node, uint32_t cell_num) {
+  return leaf_node_cell(node, cell_num);
+}
+
+void* leaf_node_value(void* node, uint32_t cell_num) {
+  return leaf_node_cell(node, cell_num) + LEAF_NODE_KEY_SIZE;
+}
+
+void print_constants() {
+  printf("ROW_SIZE: %d\n", ROW_SIZE);
+  printf("COMMON_NODE_HEADER_SIZE: %d\n", COMMON_NODE_HEADER_SIZE);
+  printf("LEAF_NODE_HEADER_SIZE: %d\n", LEAF_NODE_HEADER_SIZE);
+  printf("LEAF_NODE_CELL_SIZE: %d\n", LEAF_NODE_CELL_SIZE);
+  printf("LEAF_NODE_SPACE_FOR_CELLS: %d\n", LEAF_NODE_SPACE_FOR_CELLS);
+  printf("LEAF_NODE_MAX_CELLS: %d\n", LEAF_NODE_MAX_CELLS);
+}
+
+void print_leaf_node(void* node) {
+  uint32_t num_cells = *leaf_node_num_cells(node);
+  printf("leaf (size %d)\n", num_cells);
+  for (uint32_t i = 0; i < num_cells; i++) {
+    uint32_t key = *leaf_node_key(node, i);
+    printf("  - %d : %d\n", i, key);
+  }
+}
+
 void print_row(Row* row) {
     printf("(%d, %s, %s)\n", row->id, row->username, row->email);
 }
@@ -101,6 +173,8 @@ void deserialize_row(void *source, Row* destination) {
     memcpy(&(destination->email), source + EMAIL_OFFSET, EMAIL_SIZE);
 }
 
+void initialize_leaf_node(void* node) { *leaf_node_num_cells(node) = 0; }
+
 void* get_page(Pager* pager, uint32_t page_num) {
   if (page_num > TABLE_MAX_PAGES) {
     printf("Tried to fetch page number out of bounds. %d > %d\n", page_num,
@@ -128,6 +202,10 @@ void* get_page(Pager* pager, uint32_t page_num) {
     }
 
     pager->pages[page_num] = page;
+
+    if (page_num >= pager->num_pages) {
+      pager->num_pages = page_num + 1;
+    }
   }
 
   return pager->pages[page_num];
@@ -136,8 +214,12 @@ void* get_page(Pager* pager, uint32_t page_num) {
 Cursor* table_start(Table* table) {
   Cursor* cursor = malloc(sizeof(Cursor));
   cursor->table = table;
-  cursor->row_num = 0;
-  cursor->end_of_table = (table->num_rows == 0);
+  cursor->page_num = table->root_page_num;
+  cursor->cell_num = 0;
+
+  void* root_node = get_page(table->pager, table->root_page_num);
+  uint32_t num_cells = *leaf_node_num_cells(root_node);
+  cursor->end_of_table = (num_cells == 0);
 
   return cursor;
 }
@@ -145,24 +227,28 @@ Cursor* table_start(Table* table) {
 Cursor* table_end(Table* table) {
   Cursor* cursor = malloc(sizeof(Cursor));
   cursor->table = table;
-  cursor->row_num = table->num_rows;
+  cursor->page_num = table->root_page_num;
+
+  void* root_node = get_page(table->pager, table->root_page_num);
+  uint32_t num_cells = *leaf_node_num_cells(root_node);
+  cursor->cell_num = num_cells;
   cursor->end_of_table = true;
 
   return cursor;
 }
 
 void* cursor_value(Cursor* cursor) {
-  uint32_t row_num = cursor->row_num;
-  uint32_t page_num = row_num / ROWS_PER_PAGE;
+  uint32_t page_num = cursor->page_num;
   void* page = get_page(cursor->table->pager, page_num);
-  uint32_t row_offset = row_num % ROWS_PER_PAGE;
-  uint32_t byte_offset = row_offset * ROW_SIZE;
-  return page + byte_offset;
+  return leaf_node_value(page, cursor->cell_num);
 }
 
 void cursor_advance(Cursor* cursor) {
-  cursor->row_num += 1;
-  if (cursor->row_num >= cursor->table->num_rows) {
+  uint32_t page_num = cursor->page_num;
+  void* node = get_page(cursor->table->pager, page_num);
+
+  cursor->cell_num += 1;
+  if (cursor->cell_num >= (*leaf_node_num_cells(node))) {
     cursor->end_of_table = true;
   }
 }
@@ -185,6 +271,12 @@ Pager* pager_open(const char* filename) {
   Pager* pager = malloc(sizeof(Pager));
   pager->file_descriptor = fd;
   pager->file_length = file_length;
+  pager->num_pages = (file_length / PAGE_SIZE);
+
+  if (file_length % PAGE_SIZE != 0) {
+    printf("Db file is not a whole number of pages. Corrupt file.\n");
+    exit(EXIT_FAILURE);
+  }
 
   for (uint32_t i = 0; i < TABLE_MAX_PAGES; i++) {
     pager->pages[i] = NULL;
@@ -194,11 +285,15 @@ Pager* pager_open(const char* filename) {
@@ -195,11 +287,16 @@ Pager* pager_open(const char* filename) {
 
 Table* db_open(const char* filename) {
   Pager* pager = pager_open(filename);
-  uint32_t num_rows = pager->file_length / ROW_SIZE;
 
   Table* table = malloc(sizeof(Table));
   table->pager = pager;
-  table->num_rows = num_rows;
+  table->root_page_num = 0;
+
+  if (pager->num_pages == 0) {
+    // New database file. Initialize page 0 as leaf node.
+    void* root_node = get_page(pager, 0);
+    initialize_leaf_node(root_node);
+  }
 
   return table;
 }
@@ -234,7 +331,7 @@ void close_input_buffer(InputBuffer* input_buffer) {
     free(input_buffer);
 }
 
-void pager_flush(Pager* pager, uint32_t page_num, uint32_t size) {
+void pager_flush(Pager* pager, uint32_t page_num) {
   if (pager->pages[page_num] == NULL) {
     printf("Tried to flush null page\n");
     exit(EXIT_FAILURE);
@@ -242,7 +337,7 @@ void pager_flush(Pager* pager, uint32_t page_num, uint32_t size) {
@@ -249,7 +346,7 @@ void pager_flush(Pager* pager, uint32_t page_num, uint32_t size) {
   }
 
   ssize_t bytes_written =
-      write(pager->file_descriptor, pager->pages[page_num], size);
+      write(pager->file_descriptor, pager->pages[page_num], PAGE_SIZE);
 
   if (bytes_written == -1) {
     printf("Error writing: %d\n", errno);
@@ -252,29 +347,16 @@ void pager_flush(Pager* pager, uint32_t page_num, uint32_t size) {
@@ -260,29 +357,16 @@ void pager_flush(Pager* pager, uint32_t page_num, uint32_t size) {
 
 void db_close(Table* table) {
   Pager* pager = table->pager;
-  uint32_t num_full_pages = table->num_rows / ROWS_PER_PAGE;
 
-  for (uint32_t i = 0; i < num_full_pages; i++) {
+  for (uint32_t i = 0; i < pager->num_pages; i++) {
     if (pager->pages[i] == NULL) {
       continue;
     }
-    pager_flush(pager, i, PAGE_SIZE);
+    pager_flush(pager, i);
     free(pager->pages[i]);
     pager->pages[i] = NULL;
   }
 
-  // There may be a partial page to write to the end of the file
-  // This should not be needed after we switch to a B-tree
-  uint32_t num_additional_rows = table->num_rows % ROWS_PER_PAGE;
-  if (num_additional_rows > 0) {
-    uint32_t page_num = num_full_pages;
-    if (pager->pages[page_num] != NULL) {
-      pager_flush(pager, page_num, num_additional_rows * ROW_SIZE);
-      free(pager->pages[page_num]);
-      pager->pages[page_num] = NULL;
-    }
-  }
-
   int result = close(pager->file_descriptor);
   if (result == -1) {
     printf("Error closing db file.\n");
@@ -305,6 +389,14 @@ MetaCommandResult do_meta_command(InputBuffer* input_buffer, Table *table) {
   if (strcmp(input_buffer->buffer, ".exit") == 0) {
     db_close(table);
     exit(EXIT_SUCCESS);
+  } else if (strcmp(input_buffer->buffer, ".btree") == 0) {
+    printf("Tree:\n");
+    print_leaf_node(get_page(table->pager, 0));
+    return META_COMMAND_SUCCESS;
+  } else if (strcmp(input_buffer->buffer, ".constants") == 0) {
+    printf("Constants:\n");
+    print_constants();
+    return META_COMMAND_SUCCESS;
   } else {
     return META_COMMAND_UNRECOGNIZED_COMMAND;
   }
@@ -354,16 +446,39 @@ PrepareResult prepare_statement(InputBuffer* input_buffer,
   return PREPARE_UNRECOGNIZED_STATEMENT;
 }
 
+void leaf_node_insert(Cursor* cursor, uint32_t key, Row* value) {
+  void* node = get_page(cursor->table->pager, cursor->page_num);
+
+  uint32_t num_cells = *leaf_node_num_cells(node);
+  if (num_cells >= LEAF_NODE_MAX_CELLS) {
+    // Node full
+    printf("Need to implement splitting a leaf node.\n");
+    exit(EXIT_FAILURE);
+  }
+
+  if (cursor->cell_num < num_cells) {
+    // Make room for new cell
+    for (uint32_t i = num_cells; i > cursor->cell_num; i--) {
+      memcpy(leaf_node_cell(node, i), leaf_node_cell(node, i - 1),
+             LEAF_NODE_CELL_SIZE);
+    }
+  }
+
+  *(leaf_node_num_cells(node)) += 1;
+  *(leaf_node_key(node, cursor->cell_num)) = key;
+  serialize_row(value, leaf_node_value(node, cursor->cell_num));
+}
+
 ExecuteResult execute_insert(Statement* statement, Table* table) {
-  if (table->num_rows >= TABLE_MAX_ROWS) {
+  void* node = get_page(table->pager, table->root_page_num);
+  if ((*leaf_node_num_cells(node) >= LEAF_NODE_MAX_CELLS)) {
     return EXECUTE_TABLE_FULL;
   }
 
   Row* row_to_insert = &(statement->row_to_insert);
   Cursor* cursor = table_end(table);
 
-  serialize_row(row_to_insert, cursor_value(cursor));
-  table->num_rows += 1;
+  leaf_node_insert(cursor, row_to_insert->id, row_to_insert);
 
   free(cursor);

　　以下是规格：
​​​​​​​

+  it 'allows printing out the structure of a one-node btree' do
+    script = [3, 1, 2].map do |i|
+      "insert #{i} user#{i} person#{i}@example.com"
+    end
+    script << ".btree"
+    script << ".exit"
+    result = run_script(script)
+
+    expect(result).to match_array([
+      "db > Executed.",
+      "db > Executed.",
+      "db > Executed.",
+      "db > Tree:",
+      "leaf (size 3)",
+      "  - 0 : 3",
+      "  - 1 : 1",
+      "  - 2 : 2",
+      "db > "
+    ])
+  end
+
+  it 'prints constants' do
+    script = [
+      ".constants",
+      ".exit",
+    ]
+    result = run_script(script)
+
+    expect(result).to match_array([
+      "db > Constants:",
+      "ROW_SIZE: 293",
+      "COMMON_NODE_HEADER_SIZE: 6",
+      "LEAF_NODE_HEADER_SIZE: 10",
+      "LEAF_NODE_CELL_SIZE: 297",
+      "LEAF_NODE_SPACE_FOR_CELLS: 4086",
+      "LEAF_NODE_MAX_CELLS: 13",
+      "db > ",
+    ])
+  end
 end