If you used a formatted cdb_extents query for a single object and automated the
compare with the mutually sorted and formatted output of od on the
file(s)-blocks that could tend to validate your supposition or at least fail to
find a contrary example.
I want to be sure I understand the question: you’re being handed a block
without concurrent access to the database and need to identify the root
container row block address.
I wonder whether that’s a trade secret (sarcasm) or whether an Oracle lurker
can verify the details. My search fu is too lazy for a comprehensive search at
the moment. We’re not missing something obvious like did they add that to
concepts or something? (It should be there, right?)
mwf
From: oracle-l-bounce@xxxxxxxxxxxxx [mailto:oracle-l-bounce@xxxxxxxxxxxxx] On ;
Behalf Of Jonathan Lewis
Sent: Wednesday, October 06, 2021 1:16 PM
To: Oracle L; Seth Miller
Subject: Re: Block Identifiers in Multitenant Databases
Thinking about it a bit, the individual blocks are unlikely to hold any
information about their tablespace or absolute file number, otherwise
a) for tablespace number - every block would have to be updated to avoid a
collision when you did a 'transport tablespace'
b) for absolute file number - every block would have to be updated when you
unplugged a PDB and plugged it in somewhere else.
Regards
Jonathan Lewis
On Wed, 6 Oct 2021 at 16:55, Seth Miller <sethmiller.sm@xxxxxxxxx> wrote:
Yes, I noticed this as well and it is pretty easy to identify the file as a
whole. Unfortunately, I am dealing with individual blocks. Specifically, I need
to identify each block that comes through an rman stream which means I can have
blocks from multiple datafiles in a single stream.
Seth
On Wed, Oct 6, 2021 at 10:37 AM Jonathan Lewis <jlewisoracle@xxxxxxxxx> wrote:
Sorry,
I didn't intend my previous reply to go to the individual rather than the group.
I've just done
od -o {filename} | head -64 on the system tablespace file for the root, seed
and pdb of a multi tenant database.
Then I've flagged each line of the file (changed the first space in each line
to a letter surrounded by spaces).
The sorted the three files together.
In v$datafile root system is file 1, seed system is file 5, and the pdb system
is file 11.
The following may be a coincidence (the first 6 digit number is the byte number
in the file, in Octal:
0020060 p 020000 000000 000011 000003 000000 000000 000000 000000
prod system = file# 9
0020060 r 020000 000000 000001 000003 000000 000000 000000 000000
root system = file# 1
0020060 s 020000 000000 000005 000003 000000 000000 000000 000000
Seed system = file# 5
Note the value of byte 002065
Regards
Jonathan Lewis
On Wed, 6 Oct 2021 at 10:03, Jonathan Lewis <jlewisoracle@xxxxxxxxx> wrote:
Where are you starting from that gives you the RDBA ?
I don't think that there's anything in the block that identifies the CON_ID/pdb
id, but when Oracle is handling the block it has the relative file number,
tablespace number and absolute file number of the block in the block handle
structure (x$bh), and two of those three would have been derived (or deduced)
information. WIth the absolute file number you just need to check v$datafile
to identify the container. Bottom line - if you can infer the tablespace number
you're done.
Brute force method: check v$datafile for every absolute file# that matches your
relative file number (which isn't goint to eliminate many pdbs if you're using
bigfile tablespaces across the board), then dump the matching block number from
each file. (That's if you really want to find out for the occasional special
case, but probably too much of a nuisance in general).
Regards
Jonathan Lewis
On Tue, 5 Oct 2021 at 20:29, Seth Miller <sethmiller.sm@xxxxxxxxx> wrote:
I am attempting to isolate the identification attributes of an individual
block. Prior to multitenant, this was easy to do using the RDBA to get the file
and block. However, in a PDB, the RDBA does not tell me how the root container
identifies the block either by the file number at the root level or by the
container ID.
I've done a lot of digging and debugging and can't figure out how the root
container identifies a block or where it might be in the block header. Any help
would be appreciated.
Seth
