本節簡單介紹了PostgreSQL在執行插入過程中與儲存相關的函式RecordAndGetPageWithFreeSpace,該函式由RelationGetBufferForTuple函式呼叫,返回滿足條件的block。

### 一、資料結構
**FSMAddress**
內部的FSM處理過程以邏輯地址scheme的方式工作,樹的每一個層次都可以認為是一個獨立的地址檔案.
```
/*
 * The internal FSM routines work on a logical addressing scheme. Each
 * level of the tree can be thought of as a separately addressable file.
 * 內部的FSM處理過程工作在一個邏輯地址scheme上.
 * 樹的每一個層次都可以認為是一個獨立的地址檔案.
 */
typedef struct
{
    //層次
    int         level;          /* level */
    //該層次內的頁編號
    int         logpageno;      /* page number within the level */
} FSMAddress;

/* Address of the root page. */
//根頁地址
static const FSMAddress FSM_ROOT_ADDRESS = {FSM_ROOT_LEVEL, 0};

```

**FSMPage**
 FSM page資料結構.詳細可參看src/backend/storage/freespace/README.
```
/*
 * Structure of a FSM page. See src/backend/storage/freespace/README for
 * details.
 * FSM page資料結構.詳細可參看src/backend/storage/freespace/README.
 */
typedef struct
{
    /*
     * fsm_search_avail() tries to spread the load of multiple backends by
     * returning different pages to different backends in a round-robin
     * fashion. fp_next_slot points to the next slot to be returned (assuming
     * there's enough space on it for the request). It's defined as an int,
     * because it's updated without an exclusive lock. uint16 would be more
     * appropriate, but int is more likely to be atomically
     * fetchable/storable.
     * fsm_search_avail()函式嘗試透過在一輪迴圈中返回不同的頁面到不同的後臺程式,
     *   從而分散在後臺程式上分散負載.
     * 該欄位因為無需獨佔鎖,因此定義為整型.
     * unit16可能會更合適,但整型看起來更適合於原子提取和儲存.
     */
    int         fp_next_slot;

/*
     * fp_nodes contains the binary tree, stored in array. The first
     * NonLeafNodesPerPage elements are upper nodes, and the following
     * LeafNodesPerPage elements are leaf nodes. Unused nodes are zero.
     * fp_nodes以陣列的形式儲存二叉樹.
     * 第一個NonLeafNodesPerPage元素是上一層的節點,接下來的LeafNodesPerPage元素是葉子節點.
     * 未使用的節點為0.
     */
    uint8       fp_nodes[FLEXIBLE_ARRAY_MEMBER];
} FSMPageData;

typedef FSMPageData *FSMPage;

```

**FSMLocalMap**
對於小表,不需要建立FSM來儲存空間資訊,使用本地的記憶體對映資訊.
```

/* Either already tried, or beyond the end of the relation */
//已嘗試或者已在表的末尾之後
#define FSM_LOCAL_NOT_AVAIL 0x00

/* Available to try */
//可用於嘗試
#define FSM_LOCAL_AVAIL     0x01

/*
 * For small relations, we don't create FSM to save space, instead we use
 * local in-memory map of pages to try.  To locate free space, we simply try
 * pages directly without knowing ahead of time how much free space they have.
 * 對於小表,不需要建立FSM來儲存空間資訊,使用本地的記憶體對映資訊.
 * 為了定位空閒空間,我們不需要知道他們有多少空閒空間而是直接簡單的對page進行嘗試.
 *
 * Note that this map is used to the find the block with required free space
 * for any given relation.  We clear this map when we have found a block with
 * enough free space, when we extend the relation, or on transaction abort.
 * See src/backend/storage/freespace/README for further details.
 * 注意這個map用於搜尋給定表的請求空閒空間.
 * 在找到有足夠空閒空間的block/擴充套件了relation/在事務回滾時,則清除這個map的資訊.
 * 詳細可檢視src/backend/storage/freespace/README.
 */
typedef struct
{
    BlockNumber nblocks;//塊數
    uint8       map[HEAP_FSM_CREATION_THRESHOLD];//陣列
}           FSMLocalMap;

static FSMLocalMap fsm_local_map =
{
    0,
    {
        FSM_LOCAL_NOT_AVAIL
    }
};

#define FSM_LOCAL_MAP_EXISTS (fsm_local_map.nblocks > 0)

```

### 二、原始碼解讀
RecordAndGetPageWithFreeSpace返回滿足條件的block,其主要邏輯如下:
1.初始化相關變數
2.如存在本地map,則首先使用該檔案,呼叫fsm_local_search
3.如果沒有本地map也沒有FSM,建立本地map,然後呼叫fsm_local_search
4.使用FSM搜尋
4.1獲取FSM中原page可用空間對應的catalog
4.2根據所需空間大小,獲取FSM中相應的catalog
4.3根據原頁面,獲取heap block所在的位置(FSMAddress)
4.4檢索獲取目標slot
4.5如目標slot合法,則獲取相應的block,否則使用fsm_search搜尋合適的block

```
/*
 * RecordAndGetPageWithFreeSpace - update info about a page and try again.
 * RecordAndGetPageWithFreeSpace - 更新page info並再次嘗試.
 *
 * We provide this combo form to save some locking overhead, compared to
 * separate RecordPageWithFreeSpace + GetPageWithFreeSpace calls. There's
 * also some effort to return a page close to the old page; if there's a
 * page with enough free space on the same FSM page where the old one page
 * is located, it is preferred.
 * 相對於單獨的RecordPageWithFreeSpace + GetPageWithFreeSpace呼叫,
 *   我們提供這個組合形式用於節省一些鎖的負載.
 * 這裡同樣儲存一些努力用於返回接近舊page的page.
 * 如果與舊的page在同一個FSM page上有足夠空閒空間的page存在,那這個page會被選中.
 *
 * For very small heap relations that don't have a FSM, we update the local
 * map to indicate we have tried a page, and return the next page to try.
 * 對於非常小的堆表,是不需要FSM的,直接更新本地map來提示程式需要嘗試獲得一個page,並返回下一個page.
 */
BlockNumber
RecordAndGetPageWithFreeSpace(Relation rel, BlockNumber oldPage,
                              Size oldSpaceAvail, Size spaceNeeded)
{
    int         old_cat;
    int         search_cat;
    FSMAddress  addr;//FSM地址
    uint16      slot;//槽號
    int         search_slot;
    BlockNumber nblocks = InvalidBlockNumber;

/* First try the local map, if it exists. */
    //如存在本地map,則首先使用該檔案.
    //#define FSM_LOCAL_MAP_EXISTS (fsm_local_map.nblocks > 0)
    if (FSM_LOCAL_MAP_EXISTS)
    {
        Assert((rel->rd_rel->relkind == RELKIND_RELATION ||
                rel->rd_rel->relkind == RELKIND_TOASTVALUE) &&
               fsm_local_map.map[oldPage] == FSM_LOCAL_AVAIL);
        //設定oldPage為不可用
        fsm_local_map.map[oldPage] = FSM_LOCAL_NOT_AVAIL;
        //搜尋並返回結果
        return fsm_local_search();
    }

if (!fsm_allow_writes(rel, oldPage, InvalidBlockNumber, &nblocks))
    {
        //---- 如果FSM不允許寫
        /*
         * If we have neither a local map nor a FSM, we probably just tried
         * the target block in the smgr relation entry and failed, so we'll
         * need to create the local map.
         * 如果沒有本地map也沒有FSM,
         *   那麼我們只是嘗試了smgr relation中的目標block而且失敗了,那麼需要建立本地map.
         */
        //設定本地map
        fsm_local_set(rel, nblocks);
        //搜尋本地map
        return fsm_local_search();
    }

/* Normal FSM logic follows */
    //------ 使用FSM的邏輯
    //oldSpaceAvail/32,最大255/254
    old_cat = fsm_space_avail_to_cat(oldSpaceAvail);
    //(needed + FSM_CAT_STEP - 1) / FSM_CAT_STEP
    //#define FSM_CAT_STEP   (BLCKSZ / FSM_CATEGORIES)
    //#define FSM_CATEGORIES   256
    search_cat = fsm_space_needed_to_cat(spaceNeeded);

/* Get the location of the FSM byte representing the heap block */
    //獲得對應heap block的位置
    addr = fsm_get_location(oldPage, &slot);
    //在給定的FSM page和slot中設定值,並返回slot
    search_slot = fsm_set_and_search(rel, addr, slot, old_cat, search_cat);

/*
     * If fsm_set_and_search found a suitable new block, return that.
     * Otherwise, search as usual.
     * 如fsm_set_and_search成功找到合適的block,則返回;否則,執行常規的檢索.
     */
    if (search_slot != -1)
        return fsm_get_heap_blk(addr, search_slot);
    else
        return fsm_search(rel, search_cat);
}

/*
 * Search the local map for an available block to try, in descending order.
 * As such, there is no heuristic available to decide which order will be
 * better to try, but the probability of having space in the last block in the
 * map is higher because that is the most recent block added to the heap.
 * 以倒序的方式檢索本地map找可用的block.
 * 在這種情況下,沒有特別好的辦法用於確定那種排序方法更好,
 *   但在map中最後一個block中存在空閒空間的可能性更高,因為這是最近新增到堆中的block.
 *
 * This function is used when there is no FSM.
 * 如無FSM則使用該函式.
 */
static BlockNumber
fsm_local_search(void)
{
    BlockNumber target_block;

/* Local map must be set by now. */
    //現在本地map必須已設定
    Assert(FSM_LOCAL_MAP_EXISTS);
    //目標block
    target_block = fsm_local_map.nblocks;
    do
    {
        //迴圈
        target_block--;//從最後一個block開始
        if (fsm_local_map.map[target_block] == FSM_LOCAL_AVAIL)
            return target_block;//最後一個block可用,則返回
    } while (target_block > 0);

//target_block == 0
    /*
     * If we didn't find any available block to try in the local map, then
     * clear it.  This prevents us from using the map again without setting it
     * first, which would otherwise lead to the same conclusion again and
     * again.
     * 在本地map中沒有發現可用的block,則清除相關資訊.
     * 這可以防止我們在沒有正確設定map的情況下使用該map,
     * 這會導致重複的相同結論(沒有可用的block).
     */
    FSMClearLocalMap();
    //返回InvalidBlockNumber
    return InvalidBlockNumber;
}

/*
 * Initialize or update the local map of blocks to try, for when there is
 * no FSM.
 * 如無FSM,則初始化並更新本地map
 *
 * When we initialize the map, the whole heap is potentially available to
 * try.  Testing revealed that trying every block can cause a small
 * performance dip compared to when we use a FSM, so we try every other
 * block instead.
 * 在我們初始化map的時候,整個堆可能已可用.
 * 測試表名,與使用FSM相比,嘗試每個塊會導致小幅的效能下降,因此嘗試每一個塊.
 */
static void
fsm_local_set(Relation rel, BlockNumber cur_nblocks)
{
    BlockNumber blkno,
                cached_target_block;

/* The local map must not be set already. */
    //驗證
    Assert(!FSM_LOCAL_MAP_EXISTS);

/*
     * Starting at the current last block in the relation and working
     * backwards, mark alternating blocks as available.
     * 在關係的當前最後一個塊開始往後減少,標記可更新的塊可用.
     */
    blkno = cur_nblocks - 1;//最後一個塊
    while (true)
    {
        //更新為可用
        fsm_local_map.map[blkno] = FSM_LOCAL_AVAIL;
        if (blkno >= 2)
            blkno -= 2;
        else
            break;
    }

/* Cache the number of blocks. */
    //快取塊數
    fsm_local_map.nblocks = cur_nblocks;

/* Set the status of the cached target block to 'unavailable'. */
    //設定快取的目標塊狀態為未可用
    cached_target_block = RelationGetTargetBlock(rel);
    if (cached_target_block != InvalidBlockNumber &&
        cached_target_block < cur_nblocks)
        fsm_local_map.map[cached_target_block] = FSM_LOCAL_NOT_AVAIL;
}

/*
 * Return category corresponding x bytes of free space
 * 返回相應有x位元組空間空間的目錄
 */
static uint8
fsm_space_avail_to_cat(Size avail)
{
    int         cat;
    //確保請求的小於塊大小
    Assert(avail < BLCKSZ);
    //如大於最大請求大小,返回255
    //#define MaxFSMRequestSize   MaxHeapTupleSize
    //#define MaxHeapTupleSize   (BLCKSZ - MAXALIGN(SizeOfPageHeaderData + sizeof(ItemIdData)))
    if (avail >= MaxFSMRequestSize)
        return 255;
    //#define FSM_CAT_STEP   (BLCKSZ / FSM_CATEGORIES)
    //#define FSM_CATEGORIES   256
    //塊大小為8K則FSM_CAT_STEP = 32
    cat = avail / FSM_CAT_STEP;

/*
     * The highest category, 255, is reserved for MaxFSMRequestSize bytes or
     * more.
     * 最高層的目錄,255,保留用於MaxFSMRequestSize或者更大的大小.
     */
    if (cat > 254)
        cat = 254;//返回254

return (uint8) cat;
}

/*
 * Which category does a page need to have, to accommodate x bytes of data?
 * While fsm_size_to_avail_cat() rounds down, this needs to round up.
 * 哪一個目錄有需要的page,可滿足x bytes大小的資料.
 * 因為fsm_size_to_avail_cat()往下取整,因此這裡需要往上取整.
 */
static uint8
fsm_space_needed_to_cat(Size needed)
{
    int         cat;

/* Can't ask for more space than the highest category represents */
    //不能要求最大目錄可能表示的空間大小
    if (needed > MaxFSMRequestSize)
        elog(ERROR, "invalid FSM request size %zu", needed);

if (needed == 0)
        return 1;

cat = (needed + FSM_CAT_STEP - 1) / FSM_CAT_STEP;

if (cat > 255)
        cat = 255;

return (uint8) cat;
}

/*
 * Return the FSM location corresponding to given heap block.
 * 返回給定堆block的FSM位置.
 */
//addr = fsm_get_location(oldPage, &slot);
static FSMAddress
fsm_get_location(BlockNumber heapblk, uint16 *slot)
{
    FSMAddress  addr;

addr.level = FSM_BOTTOM_LEVEL;
    //#define SlotsPerFSMPage   LeafNodesPerPage
    //#define LeafNodesPerPage   (NodesPerPage - NonLeafNodesPerPage)
    //#define NodesPerPage (BLCKSZ - MAXALIGN(SizeOfPageHeaderData) - \
                       offsetof(FSMPageData, fp_nodes))
    //#define NonLeafNodesPerPage (BLCKSZ / 2 - 1)
    addr.logpageno = heapblk / SlotsPerFSMPage;
    *slot = heapblk % SlotsPerFSMPage;

return addr;
}

```

### 三、跟蹤分析
測試指令碼
```
15:54:13 (xdb@[local]:5432)testdb=# insert into t1 values (1,'1','1');

```
啟動gdb,設定斷點
```
(gdb) b RecordAndGetPageWithFreeSpace
Breakpoint 1 at 0x8879e4: file freespace.c, line 152.
(gdb) c
Continuing.

Breakpoint 1, RecordAndGetPageWithFreeSpace (rel=0x7fad0df13788, oldPage=1, oldSpaceAvail=16, spaceNeeded=32)
    at freespace.c:152
152     int         old_cat = fsm_space_avail_to_cat(oldSpaceAvail);
(gdb) 
```
輸入引數
```
(gdb) p *rel
$5 = {rd_node = {spcNode = 1663, dbNode = 16402, relNode = 50820}, rd_smgr = 0x2084b00, rd_refcnt = 1, rd_backend = -1, 
  rd_islocaltemp = false, rd_isnailed = false, rd_isvalid = true, rd_indexvalid = 1 '\001', rd_statvalid = false, 
  rd_createSubid = 0, rd_newRelfilenodeSubid = 0, rd_rel = 0x7fad0df139a0, rd_att = 0x7fad0df13ab8, rd_id = 50820, 
  rd_lockInfo = {lockRelId = {relId = 50820, dbId = 16402}}, rd_rules = 0x0, rd_rulescxt = 0x0, trigdesc = 0x0, 
  rd_rsdesc = 0x0, rd_fkeylist = 0x0, rd_fkeyvalid = false, rd_partkeycxt = 0x0, rd_partkey = 0x0, rd_pdcxt = 0x0, 
  rd_partdesc = 0x0, rd_partcheck = 0x0, rd_indexlist = 0x7fad0df12820, rd_oidindex = 0, rd_pkindex = 0, 
  rd_replidindex = 0, rd_statlist = 0x0, rd_indexattr = 0x0, rd_projindexattr = 0x0, rd_keyattr = 0x0, rd_pkattr = 0x0, 
  rd_idattr = 0x0, rd_projidx = 0x0, rd_pubactions = 0x0, rd_options = 0x0, rd_index = 0x0, rd_indextuple = 0x0, 
  rd_amhandler = 0, rd_indexcxt = 0x0, rd_amroutine = 0x0, rd_opfamily = 0x0, rd_opcintype = 0x0, rd_support = 0x0, 
  rd_supportinfo = 0x0, rd_indoption = 0x0, rd_indexprs = 0x0, rd_indpred = 0x0, rd_exclops = 0x0, rd_exclprocs = 0x0, 
  rd_exclstrats = 0x0, rd_amcache = 0x0, rd_indcollation = 0x0, rd_fdwroutine = 0x0, rd_toastoid = 0, 
  pgstat_info = 0x20785f0}
(gdb) 
```
1.初始化相關變數
2.如存在本地map,則首先使用該檔案,呼叫fsm_local_search
3.如果沒有本地map也沒有FSM,建立本地map,然後呼叫fsm_local_search
4.使用FSM搜尋
4.1獲取FSM中原page可用空間對應的catalog --> 0
4.2根據所需空間大小,獲取FSM中相應的catalog --> 1
```
(gdb) n
153     int         search_cat = fsm_space_needed_to_cat(spaceNeeded);
(gdb) 
159     addr = fsm_get_location(oldPage, &slot);
(gdb) p old_cat
$1 = 0
(gdb) p search_cat
$2 = 1
(gdb) 
```
4.3根據原頁面,獲取heap block所在的位置(FSMAddress)
```
(gdb) n
161     search_slot = fsm_set_and_search(rel, addr, slot, old_cat, search_cat);
(gdb) p addr
$3 = {level = 0, logpageno = 0}
(gdb) 
```
4.4檢索獲取目標slot
```
(gdb) n
167     if (search_slot != -1)
(gdb) p search_slot
$4 = 4
(gdb) 
```
4.5如目標slot合法,則獲取相應的block,否則使用fsm_search搜尋合適的block
```
(gdb) n
168         return fsm_get_heap_blk(addr, search_slot);
(gdb) 
171 }
(gdb) 
RelationGetBufferForTuple (relation=0x7fad0df13788, len=32, otherBuffer=0, options=0, bistate=0x0, vmbuffer=0x7ffe1b797dcc, 
    vmbuffer_other=0x0) at hio.c:397
397     while (targetBlock != InvalidBlockNumber)
(gdb) p targetBlock
$6 = 4
(gdb) 
```
DONE!

### 四、參考資料
[PG Source Code](https://doxygen.postgresql.org)
[Database Cluster, Databases, and Tables](http://www.interdb.jp/pg/pgsql01.html)

PostgreSQL 原始碼解讀（145）- Storage Manager#1（RecordAndGetPageWithFreeSpace）

相關文章