Skip to main content

RAC gv$ v$ view of redo log

http://asktom.oracle.com/pls/asktom/f?p=100:11:0::::P11_QUESTION_ID:18183400346178753

from Arup Nanda
Most V$ views work by selecting information from the corresponding GV$ view with a predicate "where instance_id = <that instance>". So V$SESSION in Instance 1 is actually "SELECT * FROM GV$INSTANCE WHERE INST_ID = 1". On a three node RAC database, if you select from v$session, you get sessions from that instance only. Selecting from GV$SESSION creates parallel query slaves on the other instances and gets the information back to your session. 

This works fine in almost all cases. There are few exceptions: in case of redo logs, the RAC instance must see all the redo logs of other instances as they become important for its recovery. Therefore, V$LOG actually shows all the redo logs, of all the instances, not just of its own. Contrast this with V$SESSION, which shows only sessions of that instance, not all. So, if there are 3 log file groups per instance (actually, per "thread") and there are 3 instances, V$LOG on any instance will show all 9 logfile groups, not 3. 

When you select form GV$LOG, remember, the session gets the information from other instances as well. Unfortunately, the PQ servers on those instances also get 9 records each, since they also see the same information seen by the first instance. On a three instance RAC, you will get 3X9 = 27 records in GV$LOG! 

The same explanation applies to GV$LOGFILE as well as GV$THREAD. 

In your case, you have 2 instances and there are 2 groups in each instance, so you have 4 groups in all. When you select from GV$LOG, the output shows all groups against all instances. Note your output: 

  INST_ID  GROUP#  THREAD# SEQUENCE#    BYTES  MEMBERS ARC STATUS      FIRST_CHANGE# FIRST_TIM      
                                                      
---------- ---------- ---------- ---------- ---------- ---------- --- ---------------- 
------------- ---------                                                            
      1      1      1      17  52428800      1 NO CURRENT          951208 31-DEC-06                 
                                           
      1      2      1      15  52428800      1 YES INACTIVE          940808 31-DEC-06               
                                             
      1      4      2      13  52428800      1 YES INACTIVE          948841 31-DEC-06            


This shows thread# 1 and 2 both under instance# 1. This is obviously incorrect. 

To avoid this: 

(1) Always select from V$LOG, V$LOGFILE and V$THREAD in a RAC instance. GV$ views are misleading. 

OR 

(2) add a predicate to match THREAD# with INST_ID. (Beware: thread numbers are by default the same as the instance_id; but you may have defined a different thread number while creating the database): 

SELECT * FROM GV$LOG WHERE INST_ID = THREAD# 

But I see no advantage in doing so. Your best bet is to use V$LOG. 

Comments

Popular posts from this blog

Opatch apply/lsinventory error: oneoff is corrupted or does not exist

I am applying the quarterly patch for 19c RDBMS, I tried using napply but failed, but somehow it corrupted the inventory though nothing applied. further apply and lsinventory command ran into error like this: $ ./OPatch/opatch lsinventory Oracle Interim Patch Installer version 12.2.0.1.21 Copyright (c) 2020, Oracle Corporation.  All rights reserved. Oracle Home       : /u02/app/oracle/19.0.0 Central Inventory : /u01/app/oraInventory    from           : /u02/app/oracle/19.0.0/oraInst.loc OPatch version    : 12.2.0.1.21 OUI version       : 12.2.0.7.0 Log file location : /u02/app/oracle/19.0.0/cfgtoollogs/opatch/opatch2020-09-08_13-35-59PM_1.log Lsinventory Output file location : /u02/app/oracle/19.0.0/cfgtoollogs/opatch/lsinv/lsinventory2020-09-08_13-35-59PM.txt -------------------------------------------------------------------------------- Inventory load failed... OPatch cannot load inventory ...

oracle dba_hist_sysmetric_summary

found this blog is helpful to get CPU and IO statistics on oracle database. http://shob-dbadmin.blogspot.ca/2012/12/how-to-find-total-io-of-database.html courtesy to  Shomil Bansal , below are hist writing, not mine. How to find total IO of the database instance Total IO of database instance is sum of the physical reads, physical writes and redo writes. There are several views to find these values. v$sysmetric  - Reports metric values for only the most current time sample 60 secs. v$sysmetric_summary  - Reports metric values for time sample of 1 hour. v$sysmetric_history  - Reports metric values every 60 sec from the time instance is up. Better way to analyse IO using this view to take deltas between two time periods. dba_hist_sysmetric_history  - All the above views are refreshed when the instance is restarted. This view, part of AWR, stores the historical stats. I have used this view for my report. Query: ====== set lines 350...

Oracle ASM DISK IO performance

Oracle OEM show a chart of disk performance, as shown in the chart below. when the OEM target database is 10gr2, the number of average disk response time (ms) is 0.01 ms, which unreasonable low. when the target database is 11gr2, the number is about 10ms, which I think is reasonable. it's time to find out where the discrepency comes from. The backend query is this one: set linesize 200 col name format a9 col path format a19 --col MB_per_sec format select t3.name,t2.name,t2.path,t2.reads,t2.read_time,round(t2.read_time/t2.reads*1000,3) rd_rspd_ms,t2.writes,round(t2.write_time/t2.writes*1000,3) wr_rspd_ms, round((t2.read_time+t2.write_time)/(t2.reads+t2.writes)*1000,3) dsk_rspd_ms,round((t2.bytes_read+t2.bytes_written)/1024/1024/(t2.read_time+t2.write_time))  MB_per_sec from V$asm_disk t2, v$asm_diskgroup t3 where t3.group_number=t2.group_number order by 1,2 / NAME      NAME      PATH               ...