tar crashed oracle

[CJENSEN_1]

I was doing a tar on my controlfiles and redo logs (as a test) while the database was running. It caused the database to crash. The alert log mentions a corrupt block relative... and so on. My question about TAR - does unix do something with the file during the TAR process that prevents Oracle from using the file? Just curious. We are using fbackup/frestore, but are looking at tar to do some incrementals.

[Ross Zubritski]

I appears that tar "locked" the control file. tar was never meant to be a "open file" backup solution.

Regards,

RZ

[Patrick Wallek]

As far as I know tar does not lock a file. It just reads it.

I was tar'ing up some files today and had a file change size while I was tar'ing it.

I would not really suspect tar. I suspect it may have just been a case of bad timing.

[Ross Zubritski]

Good info Patrick

Was the file that changed bin or ascii? I can see tar handling an ascii file on the fly, bin is another story.

Just my 2 cents.

RZ

[Patrick Wallek]

It was an binary file. In fact it was one of Measurewares data files in /var/opt/perf/datafiles/log*.

[Ross Zubritski]

Ah, the myteries of Oracle ;)~

[Ross Zubritski]

The plot thickens.. An oracle control file changes each time time SCN (transaction) number changes, this could be nanoseconds. The files in the perf directory changes based on the time you set for data collection, i.e. 5, 10.

Just my 2 cents once again,

Regards,

RZ

[Patrick Wallek]

That is true. I would still wonder if tar is the culprit here though.

I would be tempted to do a 'dd' on the disk that the LV is part of and see what happens. If it completes successfully, then maybe tar did do it. If not, you've got a disk going bad, which what I suspect.

[Ross Zubritski]

Excellent point sir.

[Stan_17 · Accepted solution]

Hi,

Just curious, why would you want to backup redo logs when the databases are open. never backup online redologs during hotbackup, cuz you might restore the backed up redolog file on top of current redolog thereby losing all the redo's that are not yet archived or not being checkpointed and you end up in-complete recoveries.

do not tar controlfiles too...use 'alter database backup controlfile to backup_location;' also, have ascii version of it 'alter database backup controlfile to trace;'

Stan