Comments on: Row Lock is Synonymous with TX Lock, and is a Lock on a Single Row of a Table – Verify the Documentation Challenge

By: Charles Hooper

Charles Hooper — Mon, 13 Jun 2011 12:47:03 +0000

Paresh,

Thanks for providing your point of view. I need to work through the supplied test cases – it might be interesting to see if the results of the test cases are different for the latest release of Oracle Database.

By: Charles Hooper

Charles Hooper — Mon, 13 Jun 2011 12:45:12 +0000

Kyle,

Thank you for your perspective.

I suppose that too much detail is just about as bad as not nearly enough detail when explaining concepts. I also like Mark Bobak’s summary. I wonder if an enhancement could be made to his summary. I had in mind something along the lines of “The lock structure which other sessions queue upon when those sessions attempt to modify uncommited changes made by other sessions.” (that might have been suggested in the OTN thread) Then we might also have to address those TX locks that do not result in “enq: TX – row lock contention” wait events: “enq: TX – allocate ITL entry”, “enq: TX – index contention”, and “enq: TX – contention”.

Tom Kyte’s comment does have the apparent intended effect – makes the reader stop and think for a couple of minutes.

By: Paresh

Paresh — Sun, 12 Jun 2011 03:37:29 +0000

There is just one gotcha with about conceptual understanding (my response above) as per demo by Tom. In his demo Tom explains that if TX was a true row lock, session B should be able to proceed when session A rolls back to save point.

I still will vote for Mark’s position (better of the 2 ‘evils’).

Paresh

By: Paresh

Paresh — Sun, 12 Jun 2011 03:21:32 +0000

I vote for Mark Bobak’s position especially with Mark’s statement in Kyle’s response above and Mark’s demo that we can disable TM – DML locks on a table and still have pseudo row level ‘locks’ in form of TX lock. Think of it as one lock protecting multiple rows (and ITL list etc.) just like many latches protect multiple data structures each e.g. latch that protects LRU buffer chain (the analogy is not perfect but to give an idea).

Paresh

By: Kyle Hailey

Kyle Hailey — Fri, 10 Jun 2011 21:14:14 +0000

The Tom Kyte discussion give some nice alternative perspectives.
I think part of the problem is explaining things in way that readers can grasp onto efficiently. If ideas are explained in full detial then it can sound like incomprehensible fog.

I like Mark Bobak’s summary in the discussion:

“A TX lock is the transaction itself, and can be interpreted as a row-level lock, in the sense that it is the lock structure that is used to “protect” uncommitted changes to rows.”

- Kyle Hailey