Dienstag, 4. September 2018

TFA reports full dba_feature_usage_statistics to MyOracleSupport

Those who deal with MyOracleSupport (MOS) regularly will have noticed their demand for Trace File Analyzer (TFA) diagnostic collections.
Few will know they are also providing a full content of their DBs DBA_FEATURE_USAGE_STATISTICS view; at least when running the parameters
tfactl diagcollect -database <DB_UNIQUE_NAME>

The report calls tfa_home/resources/sql/db_feature_usage.sql and it's result is stored in the <hostname>.tfa_<timestamp>.zip file.

I did not find any mention about this behavior in the documentation, so it might be worth to post about it.

This is at least true for version (latest version available at 2018-09-04)

RBAL process needs to release FORMER disk

Oracles Automatic Storage Management (ASM) has many nice advantages against other volume managers. One of my favorite is to add and remove or rebalance disks without any interruption for the database. Unfortunately the remove of devices after the rebalance is completed is not as straight forward as expected: 
In a current environment (Grid Infrastructure:, rdbms: and others) after the ALTER DISKGROUP xxx DROP DISK yyy specific discs are in status FORMER
Due to  a specific schema of directories & symlinks it takes some effort to come to the real /dev/dm-xxx device. 
Still several RBAL processes of different DBs (I did not find a derterministic pattern which processes or DBs) hold a file handle on this specific device. To make it more interesting is the fact that the diskgroup (DG) or disk was never used by any of those DBs. We have a strict mapping of DBs to DGs, so it's easy for me to claim that. 

To tell RBAL processes to release this "stale" filehandle, I have to run a
Any DG does the trick, it is not required to be related to the DB of RBAL process at all. 
As diskgroups are normally balanced, this is a kind of NOP. Still it's required to wake up RBAL and makes it release unused filehandles. 

I still have no idea why these RBAL processes open a device it's DB never needs at all; 
but as I have my proper workaround, I don't investigate any further. 

Might this help someone out there, or at least me when I hit this odd behavior again ;-)