Question abt rollback segments

Question

hi all,
I have some doubts on oracle internal architecture. When DML statements are run, the old datablocks are written to rollback segments. But it first happens in SGA and later they will be written to the file by dbwr. Any idea when they will be written? i.e., before commit or after?
Regards
Suresh

Answer 1

This has been running through my head all morning and I did some more tests. For the same demo I get this output:
7624441972          20060419 165120          INSERT     insert into "SYS"."TAB$"("OBJ#","DATAOBJ#","TS#","FILE#","BLOCK#","BOBJ#","TAB#","COLS","CLUCOLS","PCTFREE$","PCTUSED$","INITRANS","MAXTRANS","FLAGS","AUDIT$","ROWCNT","BLKCNT","EMPCNT","AVGSPC","CHNCNT","AVGRLN","AVGSPC_FLB","FLBCNT","ANALYZETIME","SAMPLESIZE","DEGREE","INSTANCES","INTCOLS","KERNELCOLS","PROPERTY","TRIGFLAG","SPARE1","SPARE2","SPARE3","SPARE4","SPARE5","SPARE6") values ('1247473','1247473','5','3','13198',NULL,NULL,'1',NULL,'10','40','1','255','1','--------------------------------',NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,'1','1','536870912','0','736',NULL,NULL,NULL,NULL,TO_DATE('19/04/06', 'DD/MM/RR'));     
7624441972          20060419 165120          INSERT     insert into "SYS"."COL$"("OBJ#","COL#","SEGCOL#","SEGCOLLENGTH","OFFSET","NAME","TYPE#","LENGTH","FIXEDSTORAGE","PRECISION#","SCALE","NULL$","DEFLENGTH","SPARE6","INTCOL#","PROPERTY","CHARSETID","CHARSETFORM","SPARE1","SPARE2","SPARE3") values ('1247473','1','1','22','0','A','2','22','0',NULL,NULL,'0',NULL,NULL,'1','0','0','0','0','0','0');     
7624441972          20060419 165120          INTERNAL          
7624441972          20060419 165120          INTERNAL          
7624441972          20060419 165120          INTERNAL          
7624441972          20060419 165120          DDL     create table T
a number
7624441973          20060419 165120          UPDATE     update "SYS"."SEG$" set "TYPE#" = '5', "BLOCKS" = '64', "EXTENTS" = '1', "INIEXTS" = '64', "MINEXTS" = '1', "MAXEXTS" = '2147483645', "EXTSIZE" = '64', "EXTPCT" = '0', "USER#" = '44', "LISTS" = '0', "GROUPS" = '0', "CACHEHINT" = '0', "HWMINCR" = '1247473', "SPARE1" = '131329' where "TS#" = '5' and "FILE#" = '3' and "BLOCK#" = '13198' and "TYPE#" = '3' and "BLOCKS" = '64' and "EXTENTS" = '1' and "INIEXTS" = '64' and "MINEXTS" = '1' and "MAXEXTS" = '2147483645' and "EXTSIZE" = '64' and "EXTPCT" = '0' and "USER#" = '44' and "LISTS" = '0' and "GROUPS" = '0' and "BITMAPRANGES" = '0' and "CACHEHINT" = '0' and "SCANHINT" = '0' and "HWMINCR" = '1247473' and "SPARE1" = '131329' and "SPARE2" IS NULL and ROWID = 'AAAAAIAABAAADxHAAY';     
7624441975     7624441975     20060419 165120     20060419 165120     COMMIT     commit;     
7624441977          20060419 165127          START     set transaction read write;     
7624441977          20060419 165127          INSERT     insert into "YJAM"."T"("A") values ('1');     
7624441979          20060419 165133          INSERT     insert into "YJAM"."T"("A") values ('2');     
7624442016     7624442016     20060419 165322     20060419 165322     COMMIT     commit;     
7624442021          20060419 165334          START     set transaction read write;     
7624442021          20060419 165334          INSERT     insert into "YJAM"."T"("A") values ('3');     
7624442025     7624442025     20060419 165341     20060419 165341     COMMIT     commit;     (I don't know if the formatting will be OK since it's a SQL Developer copy/pase and I don't have a clear formatted output there anyway)
Field 2 is the "CSCN" which I think is the Commit SCN, since it appears only on commit.
So now, I can only agree with you Mark, that the transaction SCN is actually defined when the transaction ends. But I still think that for transaction read concistency, the SCN used through the ransaction is the one of the transaction start (START - set transaction read write).
Regards,
Yoann.