So, in case someone is facing the same rman performance issue for 10.2.0.3 on linux x86-64 platform, the patch is 6081556. -Rajeev #------------------------------------------------------------------------- # Interim Patch for Base Bugs: 6081556 #------------------------------------------------------------------------- # # DATE: June 14, 2007 # ------------------------------- # Platform Patch for : 226 - Linux x86-64 # Product Version # : 10.2.0.3 # Product Patched : RDBMS # # Bugs Fixed by this patch: # ------------------------- # 6081556:TRACKING BUG FOR INVENTORY OF DG RMAN BUGS ON TOP OF 10.2.0.3 # 5998546:MLR ON TOP OF 10.2.0.3 FOR INVENTORY OF RMAN BUGS # 5219484:CATALOG RESYNCS ARE VERY SLOW - ROUT TABLE HAS 6 MILLION ROWS + # 5899994:RESYNC OF CATALOG ON DATABASES WHICH WERE STANDBY BEFORE TAKE A VERY LONG TIME # 5063635:GAP OCCURED AFTER ORA-00600 [KCCDAGF_3] IN STANDBY ARC # 5742127:BACKUP SOMETIMES FAILS WITH INTERNAL ERROR IF DATAFILE GROWS DURING BACKUP On 10/12/07, Rajeev Prabhakar <rprabha01@xxxxxxxxx> wrote: > > Charles > > From what I could see, the problemetic query that you mentioned earlier > has been clearly identified as a bug 5965453 (On Apr-3rd-07) by oracle > support on rhas for 10.2.0.3. > > -Rajeev > > > On 10/12/07, Charles Schultz <sacrophyte@xxxxxxxxx> wrote: > > > > Since I was not able to get a definitive answer about what rsr_l0key is, > > I did some investigation. This is not rocket science, but I want to > > share it just in case anyone else is curious. > > > > Apparently, rsr_l0key only has relevance when row_level = 2 (also when > > row_type = 'RECURSIVE OPERATION'). When that is the case, rsr_l0key is the > > level 0 key, where rsr_pkey is the level 1 key and rsr_key is the level 2 > > key. I think level 2 is the max, at least it is in my database. This is the > > only case where nvl(rsr_pkey,-1) != nvl(rsr_l0key, -1), or to put it in > > english, when the parent key is not exactly the same as the level 0 key, > > even if they are both null. > > > > I am not exactly sure why the rc_rman_status view wants to union in all > > the null level 0 keys. It seems like you would grab all the regular rsr > > keys, and if you want parent information, go get it with a hierarchy or > > something. > > > > However, I am even more perplexed about the original query. Why select a > > session key using a very generic timestamp that has nothing to do with your > > session? > > > > Where are those rman developers when you need to ask completely inane, > > innocuous and inconsequential questions. =) > > > > On 10/11/07, Charles Schultz <sacrophyte@xxxxxxxxx > wrote: > > > > > > Thanks to those that responded. I thought to make a "group" reply to > > > answer you all at once, but also to make this info available to anyone > > > else. > > > Unfortunately, I have not learned anything else about rsr_l0key; the jury > > > is > > > still out for that one. > > > > > > However, after looking over my SR and hearing back from some of you, > > > we will be applying bug patch 5620640 over the weekend and hopefully that > > > will help. > > > > > > On 10/11/07, Charles Schultz <sacrophyte@xxxxxxxxx > wrote: > > > > > > > > We are having some serious performance problems with the RMAN > > > > catalog; Oracle suggested that we upgraded to 10.2.0.3 but I think > > > > that has only moved the bottlenecks to different parts of the > > > > architecture. > > > > Right now, we are getting killed by this rman query: > > > > SELECT NVL (MAX(SESSION_KEY ), 0 ) FROM RC_RMAN_STATUS WHERE > > > > SESSION_STAMP < :B1 > > > > > > > > Our RC_RMAN_STATUS has ~500 k rows. Looking at the view, there is a > > > > union for rows were rsr.rsr_l0key is null. A third of our rows have > > > > a null value for that column. What is rsr_l0key? It is not documented. > > > > > > > > I am sure our problem is related to volume, but I am not convinced > > > > that this should be the problem. I mean, come on, this is a measly > > > > 500,000 rows. > > > > We are in the progress of creating a colocated catalog to partition > > > > the load a little. But I am still curious about the internals of rman, > > > > especially since it is not documented. > > > > > > > > -- > > > > Charles Schultz > > > > > >