failure of storage containing another copy of oracle redo and archive logs

[Jawed Akhtar]

Hi,
1. Could we keep multiple redo and archive logs on different storages attaced to hosts via SAN.
2. What would happen if any one location (storage) containing second archive nad redo logs is unavailaible. The main file system containing database and first copy of remain avialable. Would the database continue to work or will hang.
Thanks in advance.

Regd
Jawed Akhtar

[Eric Antunes · Accepted solution]

Hi Jawed,

1- Yes you can and you must. Duplexing is recomended for redologs and archivelogs because a disk failure won't crash your database. Create at least 2 redolog groups with 2 logfile members, each in a different storage. Use log_archive_dest and log_archive_duplex_dest to define both archivelog destinations. If you need more help defining redolog groups and members don't hesitate.

2- The database will continue if you have log_archive_min_succeed_dest=1.

Best Regards,

Eric Antunes

[Eric Antunes]

2- The database will continue if you have log_archive_min_succeed_dest=1 and, of course, the datafiles are not located in the bad storage...

[Jawed Akhtar]

Thanks Eric,
You have told that database will continue to work if any one of archive destiantion is avaialble by enabling parameter LOG_ARCHIVE_MIN_SUCCEED_DEST = 1 .

1. What about redo logs. Is there any parameter which ensures that databse will continue to work if any one of destinations for redo logs fails in case of redolog multiplexing.


Thanks

[Yogeeraj_1]

hi,

Please note that if Oracle cannot write to a redo log file at any one time, that log file can no longer be trusted, and Oracle marks it as "STALE". This indicates that the log file cannot be relied upon to provide all the data written to the log.

In general, the stale status of a redo log member should not be a cause for great concern, unless you observe that this happens too frequently.

Note that, a stale log is not necessarily an invalid log, but more of an "in-doubt" one. Once the corresponding redo group becomes the current one again, the stale status will go away by itself.

Mirroring your redo logs reduces the risks associated with a stale log file.

hope this helps too!

kind regards
yogeeraj

[Jawed Akhtar]

Thanks yoogeraj,

1. We have alocated some space from two separate storages devices on SAN to one server runing oracle. One strage has main file system for oracle database and its redo and archive files. Another secondary storage has a copy of redo logs and archive logs. Few days ago the secondary storage was unavaialble on SAN due to some fibre cable problem.
2. We observed that oracle hanged due to missing path of secondary storage having another copy of redo log files.

I would like to know

a. Was it due to unavailibilty of secondary redo logs or something else.
b. Is there any way or parameter to ensure that oracle will contiune to work even secondary stroage is abruplty removed or unavailable.

Thanks
Jawed Akhtar

[Eric Antunes]

Hi Jawed,

You must also have the control files duplexed...

Can you post the following query results (?):

select name, value from v$parameter where name like '%archive%'
union
select name, value from v$parameter where name like '%control_files%';

And

select * from v$logfile;

Best Regards,

Eric Antunes

[Jawed Akhtar]

Hi Eric,
Yes we have control files duplexed as well,
But I am unix administrator, So I don't access ORacle database.

My questions remains same. Could you give me a genral idea about my aforesaid questions. Does the duplexing of control files have any effect in availibilty of database if one location containing copies of redo and control files fails.

Thanks and Regards
Jawed Akhtar

[Eric Antunes]

"Does the duplexing of control files have any effect in availibilty of database if one location containing copies of redo and control files fails."

Yes it have: duplexing ensure your database will continue up!

What you get is invalid redolog files at that failed location (but database continues to write to the good ones):

GROUP# STATUS MEMBER
2 /disc1/oradata/TST/log2a.dbf
1 /disc1/oradata/TST/log1a.dbf
2 INVALID /disc2/app/oracle/TST/log2b.dbf
1 INVALID /disc2/app/oracle/TST/log1b.dbf
3 /disc1/oradata/TST/log3a.dbf
3 INVALID /disc2/app/oracle/TST/log3b.dbf

If you don't have duplexing and the location of redologs and/or control files fails, your database crashes...

Best Regards,

Eric Antunes

[Jawed Akhtar]

Thanks Eric,
Any good link to conecpts of Redo and control file multiplexing in Oracel 9i Database.

Regd
Jawed Akhtar

[Eric Antunes]

Yes,

There are 2 good Oracle tech sites I know:

Metalink (http://metalink.oracle.com)

Oracle Technology Network (http://www.oracle.com/technology/index.html)

In Metalink you must be registered.

There is this doc. in the OTN:

http://www.oracle.com/technology/deploy/availability/pdf/MAA_WP.pdf (search with multiplex in it)

Best Regards,

Eric Antunes

[Eric Antunes]

Hi again,

To create new redolog members are in the Generic Documentation CD that comes with Oracle Database or here:

http://www.oracle.com/technology/documentation/oracle8i.html (Or http://www.oracle.com/technology/documentation/oracle9i.html, etc...) and download the Administrator Guide for HP-UX.

To multiplex the control files:

1º - Shutdown database

2º - Copy one of the existant control files to the new location

3º - Give it a different name than others

4º - Include it in init.ora file like this:

control_files = /disc1/oradata/TST/cntrl01.ctl,
/disc2/app/oracle/TST/cntrl02.ctl

Best Regards,

Eric Antunes

[Yogeeraj_1]

hi jawed,

allow me to add this too!

To create a multiplex redo log files environment, you simply create groups with more then 1 member in them.

E.g.
system@ORA9i.mu> alter database add logfile
group 5 ( '/u02/oradata/ora9i/redo05.log',
'/u04/oradata/ora9i/redo05.log' )
size 10m;

Database altered.

The above sql command added a group with 2 members. Thus, they are now multiplexed. You can just add members to existing groups as well.

MOST IMPORTANT:
Make sure the members of the groups are on DIFFERENT physical devices -- because of otherwise, you would be defeating the purpose of multiplexing them in the first place (recovery from media failure)

hope this helps too!

kind regards
yogeeraj

[Yogeeraj_1]

hi again,

another useful link:
http://download-west.oracle.com/docs/cd/B10501_01/server.920/a96524/c04space.htm#10193

regards
yogeeraj

[Jawed Akhtar]

Hi Yogeeraj,
Yes we did multiplexing of redo logs by putting them on differnet physical device. Even these physical devices are in different storage arrays.
What I wanted to assure that in case of failure of one copy of redo log file the data base will not hang.

Thanks

[Volker Borowski]

Hi,

first of all: If you want to feel safe, test it yourself. So get yourself a testbox, install a second database just for you and may be your dba. Get a testversion for a windows-pc if nothing else is available, but you need to DO this kind of action, just to feel safe when it comes to the point where you HAVE to do it.

Kick away a redolog member and watch the alertlog what happens while you force logswitches. Bounce the DB while the log is defective. Replace the member with a valid one... Lot of things to find out, just to be confident you can handle it when it happens in production.

Allthough I did not a complete test since 8.0 I had some real defectives to fix in 8.1 and 9.2 and here goes....

Losing a redolog member will let the DB stay up and running. Been there, done that.

I only lost a controlfilecopy (one of three) once and that did crash the instance.
Startup was impossible until I replaced the defective copy (I mind to remember that this was on release 8.0). I did not try to train that error out for some releases, so may be 9.x is doing diffrent.
But I'll try this on my testdb next week, seems to be something that one should test once in a while :-)

Volker