RE: Fwd: Snapshot too old from READ-ONLY table (data pump export)

  • From: <rajendra.pande@xxxxxxx>
  • To: <Laimutis.Nedzinskas@xxxxxx>, <oracle-l@xxxxxxxxxxxxx>
  • Date: Tue, 19 Jul 2011 09:12:16 -0400

I would also be interested to know if rowdependencies changes this in

-----Original Message-----
From: oracle-l-bounce@xxxxxxxxxxxxx
[mailto:oracle-l-bounce@xxxxxxxxxxxxx] On Behalf Of
Sent: Tuesday, July 19, 2011 7:05 AM
To: oracle-l@xxxxxxxxxxxxx
Subject: Re: Fwd: Snapshot too old from READ-ONLY table (data pump


But I had to add this to



to Jonatahn Lewis comments at:.

Jonathan is probably the first in the above mentioned links to talk
COMMIT scn not just CHANGE scn - who cares about change SCN anyway ?
But he makes one interesting statement:

"When all the ITL entries show SCNs no later than your query SCN the
is read-consistent and the data can be used by the query."

I'd like to know what was ment by that. My interpretation is:

one can reconstruct a consistent read of a particular row at query start
SCN. Or one can reconstruct a consistent read of the whole block at
start SCN.

As oracle operates "in blocks" then I assume this is how oracle
reconstructs the consistent read of a block: ALL ITL's have to be delt
to have every row in a block rollbacked to commit SCN just before query
start SCN.

and finally

Mark J Bobak tells what is really needed to complete the picture:

"Then, at the row level, in the row directory, there's a lock byte. The
value of the lock byte will be either 0, if there is no lock on the row,
it will be the number of the slot in the ITL that corresponds to the
transaction that's making a change, or just holding a lock, on that
particular row. So, the locked row in the block points to the ITL slot,
the ITL slot points back at the undo segment/slot/sequence. "

Here, IMHO, the structure is complete:

Row Lock byte points to ITL, ITL points to transaction commit SCN and

Then everything adds up: row level locking, transaction isolation, all
ITL's traversal.

The block oriented nature of oracle is explained by Tom Kyte:
"o we cache blocks
o we read blocks
o blocks have transaction tables on them, this is what drives the

relational databases tend to do things "to blocks", it is basically the
they work.

We do not multi-version ROWS.
We multi-version BLOCKS.

Quite a list to have somehow complete picture :) , isn't it  ?

Brgds, Laimis N


Please consider the environment before printing this e-mail


  From:       Ron Chennells <ron.chennells@xxxxxxxxxxxxxx>


  To:         oracle-l@xxxxxxxxxxxxx


  Date:       2011.07.19 11:58


  Subject:    Re: Fwd: Snapshot too old from READ-ONLY table (data pump


Some details here


Quoting Laimutis.Nedzinskas@xxxxxx:

> Since the question was raised again, does anyone know the latest,
> up-do-date explanation on how oracle multi versioning works.
> ITL's, commit SCN's, row lock byte all put together and explained :)
> The problem is that many very clever people reverse engineered one or
> another aspect of the process but one can hardly find an all-out
> explanation.
> I dare to claim that some statements one finds on internet made by
> experienced oracle gurus are doubtful to say the least.
> Actually, oracle's own metalink is known to provide hmmm.. strange
> statements. That happens.
> The whole multi versioning is easy to explain at a high level but
> implementation is quite tricky.
> Why is implementation that important? Because knowing how it works one
> answer some questions right away how fast oracle can perform in
> situations.
> It's less of a black box approach and more of algorithm analysis then.
> Thank you in advance,
> Laimis N
> --



Please visit our website at 
for important disclosures and information about our e-mail 
policies. For your protection, please do not transmit orders 
or instructions by e-mail or include account numbers, Social 
Security numbers, credit card numbers, passwords, or other 
personal information.

Other related posts: