PostgreSQL源码学习删除数据#0

  • Z时代
  • 2024-01-10
  • 分类：综合

database

以一条delete from test where a = 123;的SQL语句为例，跟踪删除数据的代码逻辑。（PG版本为12.2）

删除数据主要的函数是heap_delete。查看调用栈：

#0  heap_delete (relation=0x7f67ac24ca28, tid=0x7fff92b2c8ba, cid=0, 
    crosscheck=0x0, wait=true, tmfd=0x7fff92b2c7d0, changingPart=false)
    at heapam.c:2447
#1  0x00000000004d3bb7 in heapam_tuple_delete (relation=0x7f67ac24ca28, 
    tid=0x7fff92b2c8ba, cid=0, snapshot=0x2e37060, crosscheck=0x0, wait=true, 
    tmfd=0x7fff92b2c7d0, changingPart=false) at heapam_handler.c:314
#2  0x00000000006dac91 in table_tuple_delete (rel=0x7f67ac24ca28, 
    tid=0x7fff92b2c8ba, cid=0, snapshot=0x2e37060, crosscheck=0x0, wait=true, 
    tmfd=0x7fff92b2c7d0, changingPart=false)
    at ../../../src/include/access/tableam.h:1230
#3  0x00000000006dbfcb in ExecDelete (mtstate=0x2dcd1f0, 
    tupleid=0x7fff92b2c8ba, oldtuple=0x0, planSlot=0x2dce5c8, 
    epqstate=0x2dcd2e8, estate=0x2dcce70, processReturning=true, 
    canSetTag=true, changingPart=false, tupleDeleted=0x0, epqreturnslot=0x0)
    at nodeModifyTable.c:768
#4  0x00000000006de016 in ExecModifyTable (pstate=0x2dcd1f0)
    at nodeModifyTable.c:2226

heap_delete函数

//src/include/access/heapam.h
extern TM_Result heap_delete(Relation relation, ItemPointer tid,
					 CommandId cid, Snapshot crosscheck, bool wait,
					 struct TM_FailureData *tmfd, bool changingPart);

TransactionId xid = GetCurrentTransactionId();
Assert(ItemPointerIsValid(tid));
/* 禁止并行操作时进行删除 */
if (IsInParallelMode())
	ereport(ERROR,
			(errcode(ERRCODE_INVALID_TRANSACTION_STATE),
			 errmsg("cannot delete tuples during a parallel operation")));
/* 根据元组的tid获取其所在的块，并取得块对应的page */
block = ItemPointerGetBlockNumber(tid);
buffer = ReadBuffer(relation, block);
page = BufferGetPage(buffer);
/* 在锁定buffer前，若page全可见需要将其visibility map给pin到共享内存，之后要修改 */
if (PageIsAllVisible(page))
	visibilitymap_pin(relation, block, &vmbuffer);
LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
/* 如果刚刚在锁定buffer期间正好page被其它进程修改变得全可见了（很稀有的情况），
 * 那么需要解锁buffer，pin了vm page后重新锁定 */
if (vmbuffer == InvalidBuffer && PageIsAllVisible(page))
{
	LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
	visibilitymap_pin(relation, block, &vmbuffer);
	LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
}
/* 从page中获取元组的ItemId */
lp = PageGetItemId(page, ItemPointerGetOffsetNumber(tid));
	Assert(ItemIdIsNormal(lp));
/* 获取元组的各项信息 */
tp.t_tableOid = RelationGetRelid(relation);
tp.t_data = (HeapTupleHeader) PageGetItem(page, lp);
tp.t_len = ItemIdGetLength(lp);
tp.t_self = *tid;
/* goto跳跃点1 */
l1:
/* 检查此元组是否允许被更新 */
result = HeapTupleSatisfiesUpdate(&tp, cid, buffer);
/* 元组当前已不可见，不允许被更新 */
if (result == TM_Invisible)
{
	UnlockReleaseBuffer(buffer);
	ereport(ERROR,
			(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
			 errmsg("attempted to delete invisible tuple")));
}
/* 元组正在被其它事务更新 */
else if (result == TM_BeingModified && wait)
{
	/* 之后要进行解锁，所以先把一些信息拷贝出来 */
	xwait = HeapTupleHeaderGetRawXmax(tp.t_data);
	infomask = tp.t_data->t_infomask;
	/* 在并发的更新或删除事务结束之前进行休眠，在睡眠前获取元组锁用来保持等待的顺序 */
	if (infomask & HEAP_XMAX_IS_MULTI)
	{
		/* 给定的MultiXactId是否与当前持有给定锁的事务冲突 */
		if (DoesMultiXactIdConflict((MultiXactId) xwait, infomask,
						LockTupleExclusive, &current_is_member))
		{
			/* 解锁buffer */
			LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
			/* 有需要的话获取元组的锁 */
			if (!current_is_member)
				heap_acquire_tuplock(relation, &(tp.t_self), LockTupleExclusive,
							 LockWaitBlock, &have_tuple_lock);
			/* 睡眠等待multixact，唤醒后重新锁定buffer */
			MultiXactIdWait((MultiXactId) xwait, MultiXactStatusUpdate, infomask,
							relation, &(tp.t_self), XLTW_Delete,
							NULL);
			LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
			/* 如果睡眠的这段时间xmax相关信息发生了变化，返回跳跃点l1重新开始 */
			if (xmax_infomask_changed(tp.t_data->t_infomask, infomask) ||
				!TransactionIdEquals(HeapTupleHeaderGetRawXmax(tp.t_data),
							 xwait))
				goto l1;
		}
	}
	/* 不是并发的更新或删除事务，但是需要等待前面的事务结束的情况 */
	else if (!TransactionIdIsCurrentTransactionId(xwait))
	{
		/* 等待常规事务结束，但首先要获取元组锁 */
		LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
		heap_acquire_tuplock(relation, &(tp.t_self), LockTupleExclusive,
						 LockWaitBlock, &have_tuple_lock);
		XactLockTableWait(xwait, relation, &(tp.t_self), XLTW_Delete);
		LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
		/* 若等待期间xmax发生了变化，返回跳跃点l1 */
		if (xmax_infomask_changed(tp.t_data->t_infomask, infomask) ||
			!TransactionIdEquals(HeapTupleHeaderGetRawXmax(tp.t_data),
							 xwait))
			goto l1;
		/* 否则，更新元组关于xmax的提交或中止的提示位 */
		UpdateXmaxHintBits(tp.t_data, buffer, xwait);
	}
	/* 若前面xmax中止，或提交后只锁定但没有更新，那么下面可以继续正常做delete操作 */
	if ((tp.t_data->t_infomask & HEAP_XMAX_INVALID) ||
		HEAP_XMAX_IS_LOCKED_ONLY(tp.t_data->t_infomask) ||
		HeapTupleHeaderIsOnlyLocked(tp.t_data))
		result = TM_Ok;
	else if (!ItemPointerEquals(&tp.t_self, &tp.t_data->t_ctid) ||
			 HeapTupleHeaderIndicatesMovedPartitions(tp.t_data))
		result = TM_Updated;
	else
		result = TM_Deleted;
}
/* 对事务快照模式RI更新执行附加检查 */
if (crosscheck != InvalidSnapshot && result == TM_Ok)
{
	if (!HeapTupleSatisfiesVisibility(&tp, crosscheck, buffer))
		result = TM_Updated;
}
/* 无法执行delete，填充TM_FailureData然后return */
if (result != TM_Ok)
{
	Assert(result == TM_SelfModified ||
		   result == TM_Updated ||
		   result == TM_Deleted ||
		   result == TM_BeingModified);
	Assert(!(tp.t_data->t_infomask & HEAP_XMAX_INVALID));
	Assert(result != TM_Updated ||
		   !ItemPointerEquals(&tp.t_self, &tp.t_data->t_ctid));
	tmfd->ctid = tp.t_data->t_ctid;
	tmfd->xmax = HeapTupleHeaderGetUpdateXid(tp.t_data);
	if (result == TM_SelfModified)
		tmfd->cmax = HeapTupleHeaderGetCmax(tp.t_data);
	else
		tmfd->cmax = InvalidCommandId;
	UnlockReleaseBuffer(buffer);
	if (have_tuple_lock)
		UnlockTupleTuplock(relation, &(tp.t_self), LockTupleExclusive);
	if (vmbuffer != InvalidBuffer)
		ReleaseBuffer(vmbuffer);
	return result;
}
/* 即将执行实际的删除，首先检查序列化冲突 */
CheckForSerializableConflictIn(relation, &tp, buffer);
/* 有必要时需要将cid换为combo cid（包含xmin和xmax），
 * 当元组被一个还未提交的事务插入，但是当前事务已不是该元组的xmin对应的事务 */
HeapTupleHeaderAdjustCmax(tp.t_data, &cid, &iscombo);
/* 获取可以标识元组身份的副本 */
old_key_tuple = ExtractReplicaIdentity(relation, &tp, true, &old_key_copied);
/* 若当前事务可能是第一个multixact操作的话，执行此函数可以保证当前事务
 * 不会成为其它更老的multiact操作的成员 */
MultiXactIdSetOldestMember();
/* 计算新的xmax和infomask */
compute_new_xmax_infomask(HeapTupleHeaderGetRawXmax(tp.t_data),
					  tp.t_data->t_infomask, tp.t_data->t_infomask2,
					  xid, LockTupleExclusive, true,
					  &new_xmax, &new_infomask, &new_infomask2);
/* 这里开始禁止ereport(ERROR) */
START_CRIT_SECTION();
/* 如果这个事务提交，元组很快就会变为dead状态。在这里给page设置
 * 一个标志，当xid低于OldestXmin时，此page会成为被修剪的候选页 */
PageSetPrunable(page, xid);
/* 清除page的全可见标志 */
if (PageIsAllVisible(page))
{
	all_visible_cleared = true;
	PageClearAllVisible(page);
	visibilitymap_clear(relation, BufferGetBlockNumber(buffer),
					vmbuffer, VISIBILITYMAP_VALID_BITS);
}
/* 存储xact删除元组的事务信息（将元组置为不可见=删除） */
tp.t_data->t_infomask &= ~(HEAP_XMAX_BITS | HEAP_MOVED);
tp.t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
tp.t_data->t_infomask |= new_infomask;
tp.t_data->t_infomask2 |= new_infomask2;
HeapTupleHeaderClearHotUpdated(tp.t_data);
HeapTupleHeaderSetXmax(tp.t_data, new_xmax);
HeapTupleHeaderSetCmax(tp.t_data, cid, iscombo);
/* 确保t_ctid中没有指向前面的链接了 */
tp.t_data->t_ctid = tp.t_self;
/* 表明这是将元组移动到了另一个分区 */
if (changingPart)
	HeapTupleHeaderSetMovedPartitions(tp.t_data);
/* 标记脏块 */
MarkBufferDirty(buffer);
/* xlog相关流程 */
if (RelationNeedsWAL(relation))
{
	/* 若是系统表元组，需要发送combocids来正确解码，
	 * 因此记录一条XLOG_HEAP2_NEW_CID记录 */
	if (RelationIsAccessibleInLogicalDecoding(relation))
		log_heap_new_cid(relation, &tp);
	xlrec.flags = 0;
	if (all_visible_cleared)
		xlrec.flags |= XLH_DELETE_ALL_VISIBLE_CLEARED;
	if (changingPart)
		xlrec.flags |= XLH_DELETE_IS_PARTITION_MOVE;
	xlrec.infobits_set = compute_infobits(tp.t_data->t_infomask,
										  tp.t_data->t_infomask2);
	xlrec.offnum = ItemPointerGetOffsetNumber(&tp.t_self);
	xlrec.xmax = new_xmax;
	if (old_key_tuple != NULL)
	{
		if (relation->rd_rel->relreplident == REPLICA_IDENTITY_FULL)
			xlrec.flags |= XLH_DELETE_CONTAINS_OLD_TUPLE;
		else
			xlrec.flags |= XLH_DELETE_CONTAINS_OLD_KEY;
	}
	/* 开始构造WAL记录，将数据添加到正在构造的WAL记录中 */
	XLogBeginInsert();
	XLogRegisterData((char *) &xlrec, SizeOfHeapDelete);
	XLogRegisterBuffer(0, buffer, REGBUF_STANDARD);
	/* 如果有的话，记录被删除的元组的身份副本 */
	if (old_key_tuple != NULL)
	{
		xlhdr.t_infomask2 = old_key_tuple->t_data->t_infomask2;
		xlhdr.t_infomask = old_key_tuple->t_data->t_infomask;
		xlhdr.t_hoff = old_key_tuple->t_data->t_hoff;
		XLogRegisterData((char *) &xlhdr, SizeOfHeapHeader);
		XLogRegisterData((char *) old_key_tuple->t_data
						 + SizeofHeapTupleHeader,
						 old_key_tuple->t_len
						 - SizeofHeapTupleHeader);
	}
	/* 复制源包含在记录中 */
	XLogSetRecordFlags(XLOG_INCLUDE_ORIGIN);
	/* 插入xlog记录 */
	recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_DELETE);
	/* 设置page的LSN */
	PageSetLSN(page, recptr);
}
END_CRIT_SECTION();
/* 解锁 */
LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
if (vmbuffer != InvalidBuffer)
		ReleaseBuffer(vmbuffer);
/* 如果元组有toast项，删除它 */
if (relation->rd_rel->relkind != RELKIND_RELATION &&
	relation->rd_rel->relkind != RELKIND_MATVIEW)
{
	/* toast表里的数据不应该递归的toast */
	Assert(!HeapTupleHasExternal(&tp));
}
else if (HeapTupleHasExternal(&tp))
	toast_delete(relation, &tp, false);
/* 对cache中的无效信息（tuple）进行注册（Cache同步机制） */
CacheInvalidateHeapTuple(relation, &tp, NULL);
/* 释放buffer */
ReleaseBuffer(buffer);
/* 释放元组锁 */
if (have_tuple_lock)
	UnlockTupleTuplock(relation, &(tp.t_self), LockTupleExclusive);
/* 更新统计信息 */
pgstat_count_heap_delete(relation);
/* 如果有的话释放old_key_tuple */
if (old_key_tuple != NULL && old_key_copied)
	heap_freetuple(old_key_tuple);
return TM_Ok;

以上是 PostgreSQL源码学习删除数据#0 的全部内容， 来源链接： utcz.com/z/534051.html

回到顶部