Shulou(Shulou.com)06/01 Report--

The rollback process should be like this: the database records the latest SCN, the SCN of the incremental checkpoint, and the maximum SCN of the redo log. Through the SCN of the incremental checkpoint, it starts to apply the redo log to the maximum SCN of the redo log and even the latest SCN, so that the rollforward is completed. During the rollback, the initial scn and the front image of the session in the undo are read directly, and the data blocks are not undo one by one. Otherwise, if 10G data is written to a data file but there is no commit, how long will it take to roll back?

Redo and undo are used because redo log writes to the persistence layer before datafile, so you need to roll over the modified data in redo log when the instance starts, and roll back the data from undo if there is no commit.

At least every three seconds the checkpoint process records information in the control file about the checkpoint position in the online redo log.

-- SCN must increase by at least 1 every 3 seconds, even if nothing happens

Conditions for dbwn processes to write to data file

The system issues checkpoint instructions (when the ckpt process starts, it causes dbwn to write data file).

The number of dirty buffers reaches the specified threshold.

The service process cannot find the free buffer after searching for a certain number of data blocks.

The time for data writing is up.

Conditions for lgwr processes to write to redo log

Before the DWBR process writes the dirty buffer to the data file

The redo log record reaches 1/3 of the log buffer

The log buffer records more than 1m logs

Every 3 seconds

Commit transaction (execute Commit)

Conditions for ckpt processes to write SCN records

A log switch occurred.

Close the instance (except for SHUTDOWN ABORT).

Checkpoint operations are performed manually.

Is forcibly issued by the initialization parameters LOG_CHECKPOINT_INTERVAL and LOG_CHECKPOINT_TIMEOUT.

The situation of shutdown abort

The modified buffer does not write to the data file

Do not roll back uncommitted changes

Do not execute ckpt

Undo record SCN

When a transaction modifies data, the database writes a new SCN to the undo data segment assigned to the transaction

Redo record SCN

LGWR puts a commit record in the redo log buffer and writes it to disk immediately, along with the commit SCN and transaction's redo entries

Suppose, a session A, modifies a row with a column value of 5 to 3, but does not have time to commit, and executes shutdown abort

There may be three situations at this time.

1. This modification is written to data file (it doesn't matter whether you write data file or not, but also to redo log (if you write data file, it will trigger redo log first), the process of instance recovery during database startup.

2. This modification is not written to data file, but to redo log (in addition to commit, four other cases will be written to redo log), and the process of instance recovery during database startup.

3. This modification is not written to data file, nor to redo log (if the conditions for writing redo log do not occur during session A), the process of instance recovery during database startup

Rollforward rollback instance recovery is the following process

Since there is no ckpt, data file and control file do not record the SCN of session A.

If redo log has been written, redo log records the SCN of session A.

If no redo log,redo log is written, at least one SCN in the first 3 seconds of session A (the first 3 seconds of shutdown abort) is recorded.

Although there is no ckpt, control file still has the latest SCN,v$database.current_scn, which may also be larger than redo log's latest SCN

When the SMON process finds that the SCN of redo log, data file and control file is inconsistent, it uses redo log to roll forward, synchronizes the data file and controls the file to make them consistent with the latest SCN of the redo log file. After the current roll is completed, it finds that the ITL state of the data block of session An is uncommitted, and then rolls back using the front image of session An in undo, and once again synchronizes the data file, control file, and redo log file to ensure that the three are consistent SCN

After shutdown abort, executing start up automatically updates the SCN of data file and control file

SQL > select to_char (sysdate,'hh:mm:ss') from dual

TO_CHAR (

-

02:03:08

SQL > select CHECKPOINT_CHANGE#,CONTROLFILE_CHANGE#,CURRENT_SCN from v$database

CHECKPOINT_CHANGE# CONTROLFILE_CHANGE# CURRENT_SCN

1062765 1062824 1063536

SQL > select distinct CHECKPOINT_CHANGE# from v$datafile_header

CHECKPOINT_CHANGE#

-

1062765

SQL > select distinct CHECKPOINT_CHANGE#,last_change# from v$datafile

CHECKPOINT_CHANGE# LAST_CHANGE#

--

1062765

SQL > shutdown abort

ORACLE instance shut down.

SQL > startup

ORACLE instance started.

Total System Global Area 1272213504 bytes

Fixed Size 1344680 bytes

Variable Size 788532056 bytes

Database Buffers 469762048 bytes

Redo Buffers 12574720 bytes

Database mounted.

Database opened.

SQL > select to_char (sysdate,'hh:mm:ss') from dual

TO_CHAR (

-

02:03:50

SQL > select CHECKPOINT_CHANGE#,CONTROLFILE_CHANGE#,CURRENT_SCN from v$database

CHECKPOINT_CHANGE# CONTROLFILE_CHANGE# CURRENT_SCN

1083549 1083604 1083812

SQL > select distinct CHECKPOINT_CHANGE# from v$datafile_header

CHECKPOINT_CHANGE#

-

1083549

SQL > select distinct CHECKPOINT_CHANGE#,last_change# from v$datafile

CHECKPOINT_CHANGE# LAST_CHANGE#

--

1083549

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.