1) redo和undo的“藕斷絲連”
前滾redo
回滾undo
undo受到redo的保護(hù)，以便歷經(jīng)前滾后在buffer cache中得到undo，并將這些undo運(yùn)用到buffer cache里的數(shù)據(jù)塊和索引塊上，使那些數(shù)據(jù)文件能“趕上”數(shù)據(jù)庫(kù)的其余部分，從而保持?jǐn)?shù)據(jù)庫(kù)的一致性。
二者是DBA和開(kāi)發(fā)人員的橋梁
回滾過(guò)程絕不涉及重做日志，只有恢復(fù)和歸檔時(shí)會(huì)讀取重做日志。因?yàn)椋刈鋈罩臼怯脕?lái)寫的，而不是用來(lái)讀！

2) 刪除聯(lián)機(jī)重做日志文件組的注意事項(xiàng)

①日志組為active和current狀態(tài)時(shí)不可以刪除
②日志組在數(shù)據(jù)庫(kù)級(jí)別刪除后操作系統(tǒng)上的文件不會(huì)被級(jí)鏈刪除
③對(duì)于一個(gè)Oracle數(shù)據(jù)庫(kù)實(shí)例，至少要包含兩個(gè)聯(lián)機(jī)重做日志組

3) redo buffer由誰(shuí)來(lái)寫入？

> redo buffer的寫入 也是 lgwr嗎？

No. Server (shadow) processes write to log buffer. LGWR writes redo records from the log buffer to the log file.

> 而undo buffer因?yàn)榫褪窃赿b cache buffer,所以也是DBWR讀入的嗎？

Not quite sure what you mean. "Undo buffer" is just a buffer in buffer cache that corresponds to a block in an undo tablespace.
Treat it like any other data block. It's written into buffer cache by a server process and written out to datafiles by DBWn.

> 涉及IO（特別是異步），就需要一個(gè)專門的進(jìn)程來(lái)完成

A server process can always read datafiles. But normally only DBWn can write to them. Also remember a server process can read and write a tempfile

理解redo（8）redo和undo的一點(diǎn)關(guān)系及刪除聯(lián)機(jī)日志文件組的注意事項(xiàng)