Data Concurrency and Consistency ( SCN , DATA block)

Question

Hi guys, i am getting very very very confused about how oracle implement consistency / multiversioning with regards to SCN in a data block and transaction list in the data block..
I will list out what i know so you guys can gauge me on where i am..
When a SELECT statement is issued, SCN for the select query is determined. Then Blocks with higher SCN are rebuilt from the RBS.
Q1) The SCN in the block implied here - is it different from the SCNs in the transaction list of the block ? where is this SCN store ? where is the transaction list store ? how is the SCN of the block related with the SCNs in the transaction list of the block ?
Q2) can someone tell me what happen to the BLOCK SCN and the transaction list
of the BLOCK when a transaction start to update to a row in the block occurs.
Q3) If the BLOCK SCN reflects the latest change made to the block and If the SCN of the block is higher then the SCN of the SELECT query, it means that the block has change since the start of the SELECT query, but it DOESNT mean that the row (data) that the SELECT query requires has changed.
Therefore why cant ORACLE just check to see whether the row has changed and if it has, rebuilt a block from the RBS ?
Q4) when ORACLE compares the BLOCK SCN, does it only SCAN for the BLOCK SCN or does it also SEARCH through the TRANSACTION LIST ? or it does both ? and why ?
Q5) is transaction SCN same as Transaction ID ? which is store in the RBS , the transaction SCN or ID ?
Q6) in short i am confuse with the relationship between BLOCK SCN, transaction list SCN, their location, their usage and relationship of the BLOCK SCN and transaction list when doing a SELECT, their link with RBS..
any gurus clear to give me a clearer view of what is actually happening ?

Answer 1

Hi Aman
Hmm agreed.So when commit is issued , what happens at that time?Simply put:
- The SCN for the transaction is determined.
- The transaction is marked as committed in the undo header (the commit SCN is also stored in the undo header).
- If fast cleanout takes place, the commit SCN is also stored in the ITL. If not, the ITL (i.e. the modified data blocks) are not modified.
So at commit, Oracle will replace the begin scn in the ITL with this scn
and this will tell that the block is finally committed is it?The ITL does not contain the begin SCN. The undo header (specifically the transaction table) contains it.
I lost here.In the ITL , the scn is transaction SCN or commit scn?As I just wrote, the ITL contains (if the cleanout occured) the commit SCN.
This sounds like high RBA information?What is RBA?
Commit SCNThis is the SCN associated with a committed transaction.
Begin SCNThis is the SCN at which a transaction started.
Transaction SCNAs I wrote, IMO, this is the same as the commit SCN.
Also please explain that what exactly the ITL stores?If you print an ITL slot, you see the following information:
BBED> print ktbbhitl[0]
struct ktbbhitl[0], 24 bytes     @44
      struct ktbitxid, 8 bytes    @44
         ub2 kxidusn              @44       0x0009
         ub2 kxidslt              @46       0x002e
         ub4 kxidsqn              @48       0x0000fe77
      struct ktbituba, 8 bytes    @52
         ub4 kubadba              @52       0x00800249
         ub2 kubaseq              @56       0x3ed6
         ub1 kubarec              @58       0x4e
      ub2 ktbitflg                @60       0x2045 (KTBFUPB)
      union _ktbitun, 2 bytes     @62
         b2 _ktbitfsc             @62       0
         ub2 _ktbitwrp            @62       0x0000
      ub4 ktbitbas                @64       0x06f4c2a3- ktbitxid --> XID, the transaction holding the ITL slot
- ktbituba --> UBA, used to locate the undo information
- ktbitflg --> flags (active, committed, cleaned out, ...)
- _ktbitfsc --> free space generated by this transaction in this block
- _ktbitwrp+ktbitbas --> commit SCN
HTH
Chris