@title [筆記]事務處理

#1 Transaction Propagation Behavior
Required:Excute within a current tx, create a new one if none exists.
Supports: Excute within a current tx, execute without a tx if none exsits.
Mandatory: Excute within a current tx, throw an exception if none exists.
Requires New: Create a new tx and excute within the tx, suspend the current tx if one exists.
Not Supported: Excute without a tx, suspend the current tx if none exsits.
Never: Excute without a tx, throw an exception if a tx exsits.

#2 Transaction Isolation Level[1]
#2.1 Concurrent Problem
Dirty Reads: 臟讀（臟數據指已更新，還沒提交的數據）。事務T1讀取到事務T2中的臟數據。

Unrepeatable Reads: 不可重復讀。事務T1檢索到某行后，事務T2更新并提交了該行，若事務T2再次檢索該行，則會看到不一樣的結果。

Phantom Reads: 虛讀。事務T1檢索到符合某條件的行集后，事務T2插入并提交了滿足該條件的新行，若事務T2再次按該條件檢索，則會看到以前不存在的行“Phantom”。

#2.2 Isolation Level
+---------------+-------+------------------+-----------+
|Isolation Level|Phantom|Unrepeatable Reads|Dirty Reads|
+---------------+-------+------------------+-----------+
|Read Uncommited|   Y   |         Y        |     Y     |
+---------------+-------+------------------+-----------+
|Read Commited  |   Y   |         Y        |     N     |
+---------------+-------+------------------+-----------+
|Repeatable Read|   Y   |         N        |     N     |
+---------------+-------+------------------+-----------+
|Serializable   |   N   |         N        |     N     |
+---------------+-------+------------------+-----------+

#3 Timeout

#4 ReadOnly Transaction
只讀事務保證了多條查詢SQL的在事務級別的讀一致性。JDBC和數據庫會對只讀事務做一些優化。

[1] C.J.Date, An Introduction to Database Systems 7th.