4. Log File Operations
Redo is written to disk when
User commits
Log Buffer 1/3 full (_log_io_size)
Log Buffer fills 1M
Every 3 seconds
DBWR asks LGWR to flush redo
Sessions Commiting wait for LGWR
Copyright 2006 Kyle Hailey #.4
5. log buffer space
Wait for space in the redo log buffer in SGA
Solution
1. Increase log_buffer parameter in init.ora
Above 3M log_buffer little affect, if still a problem then
backup is at disk level
1. Improve disk IO for redo
Faster disk
Raw file
Direct IO
Dedicated disk
p1, p2, p3 – no values
Copyright 2006 Kyle Hailey #.5
6. Log Buffer Space
SGA
Log Library Buffer
Buffer Cache Cache
Buffer Cache
Log Buffer
LGWR
User1
1. Log Buffer too small
User2
2. LGWR too slow
User3 Slow disk
Insert
Update
REDO Log Files delete
#.6
Copyright 2006 Kyle Hailey
7. log file sync
Wait for redo flush upon:
Commit
Rollback
Arguments
P1 = buffer# in log buffer that needs to be flushed
P2 = not used
P3 = not used
select parameter1, parameter2, parameter3
from v$event_name
where name=‘log file sync';
PARAMETER1 PARAMETER2 PARAMETER3
buffer#
Copyright 2006 Kyle Hailey
#.7
9. Log File Sync: Solutions
Commit less
Often possible in loops that commit every loop
Commit every 50 or 100 instead
Possibly 10gR2
ALTER SYSTEM SET COMMIT_WRITE = BATCH, NOWAIT
Commit could be lost if machine crash or IO error
Improve IO
Use Raw Device or Direct IO
Consider Ram Disks
Can stripe if redo writes are comparable to stripe size
Striping shouldn’t hurt
Striping can help
Ex: imp – can have large redo writes – can improve by 10-
30%
Alternate disks for redo and archiving of redo
(_high_priority_processes)
Copyright 2006 Kyle Hailey #.9
10. Log File Sync depends on:
log file parallel write
Time it takes for LGWR to write out changes
If log file sync =~ log file parallel write
And the time is slow ( > 3ms) look into IO issues
If log file sync >> log file parallel write
Look at CPU starvation issues
Copyright 2006 Kyle Hailey #.10
14. log file switch (checkpoint incomplete)
No p1,p2,p3 args
Wait for checkpoint to complete because all log
files are full
Solutions
Increase size of log files
Copyright 2006 Kyle Hailey #.14
18. switch logfile command
Same as log file switch completion but the
command is executed by the dba
Alter system switch logfile;
Copyright 2006 Kyle Hailey #.18
19. Concerns – Recovery Time
What happens to recovery time if I change my redo
log file sizes
Larger Redo Log size can increase recovery time
but
There are init.ora parameters to limit this
Copyright 2006 Kyle Hailey #.19
20. Incremental Checkpoints (9iR2+)
FAST_START_MTTR_TARGET
Seconds to Recovery
Easy and accuracy
Is overridden by FAST_START_IO_TARGET
Is overridden by LOG_CHECKPOINT_INTERVAL
alter system set fast_start_mttr_target=17 scope=both;
SQL> select ESTIMATED_MTTR from V$INSTANCE_RECOVERY;
SQL> select ESTIMATED_MTTR from V$INSTANCE_RECOVERY;
ESTIMATED_MTTR
ESTIMATED_MTTR
--------------
--------------
21
21
Copyright 2006 Kyle Hailey #.20
21. Recovery and Checkpoints
SGA
Log Library Buffer DBWR
Buffer Cache Cache
LGWR
Data Files
Current Position
Needed for Recovery
1 2 3
Incremental Checkpoint
REDO Log Files Copyright 2006 Kyle Hailey #.21
22. DBWR dirty List and LGWR
Buffers
DBWR usually
just writes out
LGWR Current dirty blocks at
Position end of LRU
until checkpoint
Incremental Checkpoint
DBWR
Checkpoint a
Now, DBWR Block xxxx
keeps a Block xxxx
checkpoint list that Block xxxx
Block xxxx
it writes out
Copyright 2006 Kyle Hailey #.22
23. DBWR dirty List
MRU - Hot Buffer Headers LRU - Cold
DBWR also has to Dirty List
track dirty blocks at the Block xxxx
cold end of the LRU Block xxxx
Block xxxx DBWR
Block xxxx
Copyright 2006 Kyle Hailey #.23
24. DBWR merges Dirty and Checkpoint
MRU - Hot Buffer Headers LRU - Cold
Checkpoint a Dirty List
Block xxxx
Block xxxx
Block xxxx
Block xxxx
DBWR
Block xxxx Block xxxx
Block xxxx Block xxxx
Write List
Block xxxx
Block xxxx
Block xxxx
Block xxxx Data Files
Copyright 2006 Kyle Hailey #.24
25. log file switch (private strand flush incomplete)
New wait 10g
Like a “log file switch Completion”
Copyright 2006 Kyle Hailey #.25
26. Redo Wait Solutions
log file sync
Commit less, put redo logs on faster disks
log buffer space
Increase log buffer no more than 32M, then tune LGWR
log file switch completion
Increase log file sizes
log file switch (checkpoint incomplete)
Add log files (or increase log file size)
switch logfile command
Avoid switching log files
log file switch (private strand flush incomplete)
increase log file sizes
log file switch (archiving needed) ***
Archive log running out of space
Copyright 2006 Kyle Hailey #.26
Notes de l'éditeur
Users who are creating redo information typically from Insert update delete Statements have to write the information into the redo log buffer. If the buffer fills up before the LGWR can write out the information then users have to wait.
Redo log 1 fills up. LGWR switches to log Redo log 2 which requires Get next log file from control file Get Redo Copy and Redo Allocation latch Flush redo Close File Update Controlfile Set new file to Current Set old file to Active If in Archivelog mode add file to archive list Open all members of new logfile group Write the SCN to the headers Enable redo log generation At the same time DBWR makes a list of all blocks in the buffer cache that are dirty And have redo in log 1. This list of blocks has to be written out To disk before LGWR can reuse log 1
In this case none of the DBWR checkpoints 1, 2 or 3 finish before LGWR filled up all 3 redo logs. IN this case users must wait until DBWR finishes the checkpoints of writing all the dirty blocks out. In older versions of Oracle the checkpoints were merged together, so all 3 checkpoints had to finish before Redo log 1 could be reusued. In later versions ( I think starting in 9) the checkpoints were kept separate thus once checkpoint 1 had finished, then log 1 could be reused.
IN this case the archiver for some reason, hasn’t been able to archive log 1 and now LGWR needs to reause it. IN this case all transactional activity in the database comes to a halt. To any user with tranactions, the database has effectively hung. This is almost always caused by the archive destination filling. Make room on the destination disk. You can manually stop and start the archiver to make sure it restarts after room is made archive log stop; -- make room in log_archive_dest archive log start;
select ESTIMATED_MTTR from v$instance_recovery; From Chris Foot 10G Automatic Checkpoint Tuning If you do not set FAST_START_MTTR_TARGET, or set it to a very large value, Oracle10g will provide automatic checkpoint tuning. The database will write out dirty blocks from the cache as fast as possible without negatively impacting database performance. The DBA is no longer required to set any of the aforementioned checkpoint parameters. SELECT TARGET_MTTR, ESTIMATED_MTTR, CKPT_BLOCK_WRITES FROM V$INSTANCE_RECOVERY CKPT_BLOCK_WRITES = represents overhead from fast_start_mttr_target
From Metalink: The message means that we haven't completed writing all the redo information to the log when we are trying to switch. It is similar in nature to a "checkpoint not complete" except that is only involves the redo being written to the log. The log switch can not occur until all of the redo has been written. A "strand" is new terminology for 10g and it deals with latches for redo . Strands are a mechanism to allow multiple allocation latches for processes to write redo more efficiently in the redo buffer and is related to the log_parallelism parameter present in 9i. The concept of a strand is to ensure that the redo generation rate for an instance is optimal and that when there is some kind of redo contention then the number of strands is dynamically adjusted to compensate. The initial allocation for the number of strands depends on the number of CPU's and is started with 2 strands with one strand for active redo generation. For large scale enterprise systems the amount of redo generation is large and hence these strands are *made active* as and when the foregrounds encounter this redo contention (allocated latch related contention) when this concept of dynamic strands comes into play. There is always shared strands and a number of private strands . Oracle 10g has some major changes in the mechanisms for redo (and undo), which seem to be aimed at reducing contention. Instead of redo being recorded in real time, it can be recorded 'privately' and pumped into the redo log buffer on commit. Similary the undo can be generated as 'in memory undo' and applied in bulk. This affect the memory used for redo management and the possibility to flush it in pieces. The message you get is related to internal Cache Redo File management. You can disregard these messages as normal messages. When you switch logs all private strands have to be flushed to the current log before the switch is allowed to proceed.