Kategorie: News from the Labs

This month I want to have a good look inside the SYSLGRNX – Not just for fun, but actually to review what is inside and ask the question: Is any of it “interesting” for us as DBAs?

Create your own?

Since Db2 10 the Db2 Directory table SYSIBM.SYSLGRNX has been available for normal SQL use. I wrote a newsletter all about this (Discovering hidden recovery problems in the SYSLGRNX.) This “older” newsletter has now been updated with a newer version of the SYSLGRNX private table/index and how to cross load it and RUNSTATS it, which we will be using within this newsletter. 

What is it?

The SYSLGRNX table is used by Db2 to “remember” when a physical page set (object) changed from R/O access to R/W access. This is a requirement to speed up RECOVER jobs. When this is done correctly, the RECOVER can simply look in SYSLGRNX to find any RBA/LRSN ranges that must actually be checked in the Db2 Log. Without this table, Db2 would *have* to access all the logs from the last Full Image Copy/Incremental Image Copy to see if any data changes had been done. This means that the table is pretty critical when it comes to doing a RECOVER job!

Why look in it?

Well, it is nice to know that Db2 is keeping tabs on everything, but what if it makes a mistake? Heaven forbid there could be bugs in the code or perhaps an LPAR crash… Basically, it boils down to a good chance that you have some garbage in the table. This does not have any impact at all *until* you try and RECOVER an object… then Db2 requires an archive log from 2017 and the object is marked as “unrecoverable” – Whoops!

SQL Time – LRSN or RBA Sir?

So now you have to decide: are you interested in data sharing (DS) or in non-data sharing (Non-DS) systems? DS use the Log Record Sequence Number (LRSN) to mark data on the log. It is basically a ten-byte super-duper timestamp field, with/without an offset, which points to a place on the log. If Non-DS, you use a Relative Byte Address (RBA) which is really just a ten-byte address pointing to a place on the log. All of the sample SQL assumes you are running in a DS world and that you have created the updated private version of SYSLGRNX. 

Start with Open Units of Work

Let’s have a quick review of all current Open Units of Work (OUW) that exist:

-- OPEN UNITS OF WORK IN SYSLGRNX                     
SELECT L.LGRDBID                    AS DBID           
      ,SUBSTR(TS.DBNAME, 1, 8)      AS DBNAME         
      ,L.LGRPSID                    AS ID             
      ,L.LGRNEGPSID                 AS NEG_PSID       
      ,SUBSTR(TS.NAME, 1, 8)        AS SPACE          
      ,L.LGRPART                    AS PART           
      ,L.LGRUCTS                    AS LGRX_UCTIMESTAMP
--    ,HEX(L.LGRSRBA)               AS START_RBA      
--    ,HEX(L.LGRSPBA)               AS END_RBA        
      ,HEX(L.LGRSLRSN)              AS START_LRSN     
      ,HEX(L.LGRELRSN)              AS END_LRSN       
      ,'TS'                                           
      ,L.LGRMEMB                    AS DM             
FROM BOXWELL.SYSLGRNX2    L                           
    ,SYSIBM.SYSTABLESPACE TS                          
WHERE TS.DBID     = L.LGRDBID                         
  AND TS.PSID     = L.LGRPSID                          
--AND HEX(L.LGRSPBA)  = '00000000000000000000'        
  AND HEX(L.LGRELRSN) = '00000000000000000000'        
UNION ALL                                             
SELECT L.LGRDBID                    AS DBID           
      ,SUBSTR(IX.DBNAME, 1, 8)      AS DBNAME         
      ,L.LGRPSID                    AS ID             
      ,L.LGRNEGPSID                 AS NEG_PSID       
      ,SUBSTR(IX.INDEXSPACE, 1, 8)  AS SPACE          
      ,L.LGRPART                    AS PART           
      ,L.LGRUCTS                    AS LGRX_UCTIMESTAMP
--    ,HEX(L.LGRSRBA)               AS START_RBA      
--    ,HEX(L.LGRSPBA)               AS END_RBA        
      ,HEX(L.LGRSLRSN)              AS START_LRSN     
      ,HEX(L.LGRELRSN)              AS END_LRSN       
      ,'IX'                                           
      ,L.LGRMEMB                    AS DM             
FROM BOXWELL.SYSLGRNX2    L                           
    ,SYSIBM.SYSINDEXES    IX                           
WHERE IX.DBID   = L.LGRDBID                           
  AND IX.ISOBID = L.LGRPSID                           
--AND HEX(L.LGRSPBA)  = '00000000000000000000'        
  AND HEX(L.LGRELRSN) = '00000000000000000000'        
ORDER BY 7                                             
WITH UR                                               
  ;                                                   

This SQL is written for DS. To convert to Non-DS just comment out the LRSN lines and uncomment out the RBA lines not forgetting the WHERE statements!

An OUW is one where the Stop RBA/End LRSN is not yet filled (so all zeroes), by the way. These are all “in use” objects and I have seen ranges from six to 17,000 of these. The Stop RBA / End LRSN is updated when the page set gets pseudo, or physically, closed again. How many of these you can/should have all depends on your workload, your choice of ZPARM PCLOSEN and when you copy the SYSLGRNX of course.

Here on my test system I only get these:

---------+---------+---------+---------+---------+---------+---------+---------+
  DBID  DBNAME        ID  NEG_PSID  SPACE       PART  LGRX_UCTIMESTAMP         
---------+---------+---------+---------+---------+---------+---------+---------+
     6  DSNDB06     2210  N         SYSTSTSS       1  2020-07-29-13.53.27.740000
     6  DSNDB06     2208  N         SYSTSISS       1  2020-08-26-12.46.56.690000
     6  DSNDB06     2068  N         SYSTSPKG       1  2020-08-26-13.46.57.710000
     6  DSNDB06     2018  N         SYSTSDBA       1  2020-08-26-14.13.13.310000
     6  DSNDB06      371  N         SYSSEQ         0  2020-08-26-14.13.13.310000
     6  DSNDB06     1994  N         SYSTSCOL       1  2020-08-26-14.13.13.320000
     6  DSNDB06     2014  N         SYSTSTAB       1  2020-08-26-14.13.13.320000
     6  DSNDB06     2016  N         SYSTSTSP       1  2020-08-26-14.13.13.320000
     6  DSNDB06     2012  N         SYSTSTPT       1  2020-08-26-14.13.13.320000
     6  DSNDB06     2020  N         SYSTSDBU       1  2020-08-26-14.13.13.320000
     6  DSNDB06     2010  N         SYSTSTAU       1  2020-08-26-14.13.13.340000
   505  DSN00171       2  N         SYSLGRNX       1  2020-08-26-14.13.33.340000
DSNE610I NUMBER OF ROWS DISPLAYED IS 12                                        

But here you can see there is something odd in that first row! I do not believe we have an OUW from July as it is the end of August at time of writing.

Grouping the output

Now to get a view of how many records per object you have. With this output you can instantly see where you have *way* too many records and need at least a MODIFY RECOVERY and/or a full image copy!

-- GROUP RECORDS IN SYSLGRNX                              
SELECT L.LGRDBID                    AS DBID               
      ,SUBSTR(TS.DBNAME, 1, 8)      AS DBNAME             
      ,L.LGRPSID                    AS ID                 
      ,L.LGRNEGPSID                 AS NEG_PSID           
      ,SUBSTR(TS.NAME, 1, 8)        AS SPACE              
      ,L.LGRPART                    AS PART               
      ,MIN(L.LGRUCTS)               AS MIN_LGRX_UCTIMESTAMP
      ,MAX(L.LGRUCTS)               AS MAX_LGRX_UCTIMESTAMP
--    ,MIN(HEX(L.LGRSRBA))          AS MIN_START_RBA      
--    ,MAX(HEX(L.LGRSRBA))          AS MAX_START_RBA      
--    ,MIN(HEX(L.LGRSPBA))          AS MIN_END_RBA        
--    ,MAX(HEX(L.LGRSPBA))          AS MAX_END_RBA        
      ,MIN(HEX(L.LGRSLRSN))         AS MIN_START_LRSN     
      ,MAX(HEX(L.LGRSLRSN))         AS MAX_START_LRSN     
      ,MIN(HEX(L.LGRELRSN))         AS MIN_END_LRSN       
      ,MAX(HEX(L.LGRELRSN))         AS MAX_END_LRSN       
      ,'TS'                                               
      ,COUNT(*)                                           
FROM BOXWELL.SYSLGRNX2    L                               
    ,SYSIBM.SYSTABLESPACE TS                              
WHERE TS.DBID = L.LGRDBID                                 
  AND TS.PSID = L.LGRPSID                                 
--AND NOT L.LGRSRBA  = L.LGRSPBA                          
  AND NOT L.LGRSLRSN = L.LGRELRSN                         
GROUP BY L.LGRDBID                                        
       , TS.DBNAME                                         
       , L.LGRPSID                                        
       , L.LGRNEGPSID                                     
       , TS.NAME                                          
       , L.LGRPART                                        
UNION ALL                                                 
SELECT L.LGRDBID                    AS DBID               
      ,SUBSTR(IX.DBNAME, 1, 8)      AS DBNAME             
      ,L.LGRPSID                    AS ID                 
      ,L.LGRNEGPSID                 AS NEG_PSID           
      ,SUBSTR(IX.INDEXSPACE, 1, 8)  AS SPACE              
      ,L.LGRPART                    AS PART               
      ,MIN(L.LGRUCTS)               AS MIN_LGRX_UCTIMESTAMP
      ,MAX(L.LGRUCTS)               AS MAX_LGRX_UCTIMESTAMP
--    ,MIN(HEX(L.LGRSRBA))          AS MIN_START_RBA      
--    ,MAX(HEX(L.LGRSRBA))          AS MAX_START_RBA      
--    ,MIN(HEX(L.LGRSPBA))          AS MIN_END_RBA        
--    ,MAX(HEX(L.LGRSPBA))          AS MAX_END_RBA        
      ,MIN(HEX(L.LGRSLRSN))         AS MIN_START_LRSN     
      ,MAX(HEX(L.LGRSLRSN))         AS MAX_START_LRSN
      ,MIN(HEX(L.LGRELRSN))         AS MIN_END_LRSN 
      ,MAX(HEX(L.LGRELRSN))         AS MAX_END_LRSN 
      ,'IX'                                         
      ,COUNT(*)                                     
FROM BOXWELL.SYSLGRNX2    L                          
    ,SYSIBM.SYSINDEXES    IX                        
WHERE IX.DBID   = L.LGRDBID                         
  AND IX.ISOBID = L.LGRPSID                         
--AND NOT L.LGRSRBA  = L.LGRSPBA                    
  AND NOT L.LGRSLRSN = L.LGRELRSN                    
GROUP BY L.LGRDBID                                  
       , IX.DBNAME                                  
       , L.LGRPSID                                  
       , L.LGRNEGPSID                               
       , IX.INDEXSPACE                              
       , L.LGRPART                                  
ORDER BY 14 DESC                                    
WITH UR                                             
  ;                                                 

Again, this is a DS version – same rules apply as the first SQL and, in fact, for all following SQL as well!

Output looks like:

---------+---------+---------+---------+---------+---------+---------+---------+
  DBID  DBNAME        ID  NEG_PSID  SPACE       PART  MIN_LGRX_UCTIMESTAMP     
---------+---------+---------+---------+---------+---------+---------+---------+
     6  DSNDB06     2208  N         SYSTSISS       1  2019-09-28-15.02.57.790000
     6  DSNDB06     2210  N         SYSTSTSS       1  2019-09-28-15.02.57.780000
     6  DSNDB06     2068  N         SYSTSPKG       1  2019-09-28-13.51.16.650000
     6  DSNDB06     2228  N         SYSTSSFB       1  2019-09-28-15.02.58.840000
     6  DSNDB06     2010  N         SYSTSTAU       1  2019-09-28-13.51.18.800000
   348  MVNXTEST     352  N         MVNXS89        1  2020-03-11-07.21.38.830000
     1  DSNDB01      127  N         SPT01          1  2019-09-28-13.51.16.620000
-------+---------+---------+---------+---------+---------+---------+--
MAX_LGRX_UCTIMESTAMP        MIN_START_LRSN        MAX_START_LRSN     
-------+---------+---------+---------+---------+---------+---------+--
2020-08-26-12.46.56.690000  00D6CADBE78BCD209000  00D86990DB7295A35400
2020-07-29-13.53.27.740000  00D6CADBE77162E15600  00D84A37CE9CCF71C400
2020-08-26-13.46.57.710000  00D6CACBE16C6D6D2800  00D867213DA095E82400
2020-08-26-09.16.50.090000  00D6CADBE7FB6E5FCC00  00D8670666C5C4344600
2020-08-26-14.13.13.340000  00D6CACBE2DBBC140000  00D866E87FD0328A4600
2020-07-02-08.42.53.340000  00D7A00B986CE7398600  00D82800085E355D1800
2020-08-26-10.18.22.240000  00D6CACBE1285CB92800  00D866E87F9F228F0A00
-----+---------+---------+---------+---------+---------+---

MIN_END_LRSN          MAX_END_LRSN                         
-----+---------+---------+---------+---------+---------+---
00D6CADE5D36D0BD2000  00000000000000000000  TS        22519
00D6CADE5D34CB2B6C00  00000000000000000000  TS        21092
00D6CACCE2EF932A2400  00000000000000000000  TS          698
00D6CADE5D381865B400  00D86710F530EDF11800  TS          568
00D6CACCE2F4C7A05800  00000000000000000000  TS          487
00D7A00A120AD792D800  00D8280A9A34E82F9800  TS          450
00D6CACCE344A2021200  00D866F33992A0EDD600  TS          448

Here you can see that the RTS tablespaces are in desparate need of a Full Image Copy and a MODIFY RECOVER! I hope you have no numbers as high as mine!

Oh, for the poor orphans

Yes indeed, orphans also exist in Db2 SYSLGRNX sometimes… Run this SQL to check: 

-- ORPHAN RECORDS IN SYSLGRNX - SHOULD BE NONE        
SELECT L.LGRDBID                    AS DBID           
      ,L.LGRPSID                    AS ID             
      ,L.LGRNEGPSID                 AS NEG_PSID       
      ,L.LGRPART                    AS PART           
      ,L.LGRUCTS                    AS LGRX_UCTIMESTAMP
--    ,HEX(L.LGRSRBA)               AS START_RBA      
--    ,HEX(L.LGRSPBA)               AS END_RBA        
      ,HEX(L.LGRSLRSN)              AS START_LRSN     
      ,HEX(L.LGRELRSN)              AS END_LRSN       
      ,L.LGRMEMB                    AS DM             
FROM BOXWELL.SYSLGRNX2    L                           
WHERE NOT EXISTS                                      
 (SELECT 1                                            
  FROM SYSIBM.SYSTABLESPACE TS                        
  WHERE TS.DBID     = L.LGRDBID                        
    AND TS.PSID     = L.LGRPSID                       
 )                                                    
  AND NOT EXISTS                                      
 (SELECT 1                                            
  FROM SYSIBM.SYSINDEXES    IX                        
  WHERE IX.DBID   = L.LGRDBID                         
    AND IX.ISOBID = L.LGRPSID                         
 )                                                    
ORDER BY LGRX_UCTIMESTAMP                              
WITH UR                                               
  ;                                                   

In my DS there are none, but when I run in my non-DS system I get:

---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+----
  DBID      ID  NEG_PSID    PART  LGRX_UCTIMESTAMP            START_RBA             END_RBA            
---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+----
   414       2  N              1  2020-01-06-13.31.11.100000  00000000008EB1D404EC  00000000008EB1D419EC
   414       2  N              1  2020-01-06-13.31.11.110000  00000000008EB1D438A7  00000000008EB1D49074

Now I happen to remember this! A colleague was “playing” with CLONE Objects and caused a bit of a problem back in the first week of the year… I have to now figure out how to get rid of these bad boys really…REPAIR should do it for me, but in a productive system I would probably contact IBM for assistance.

SYSLGRNX for Dummies?

You may have noticed that in most of the queries there is a predicate:

--AND NOT L.LGRSRBA  = L.LGRSPBA                    
  AND NOT L.LGRSLRSN = L.LGRELRSN                   

So, the question is: What??? Why would the SYSLGRNX have entries where the Start RBA was equal to the Stop RBA or the Start LRSN was equal to the End LRSN? These are special “events” or the so-called “dummy” entries. To see how many you have, run this little query:

-- SPECIAL/DUMMY ENTRIES IN SYSLGRNX                  
SELECT L.LGRDBID                    AS DBID           
      ,SUBSTR(TS.DBNAME, 1, 8)      AS DBNAME         
      ,L.LGRPSID                    AS ID             
      ,L.LGRNEGPSID                 AS NEG_PSID       
      ,SUBSTR(TS.NAME, 1, 8)        AS SPACE          
      ,L.LGRPART                    AS PART           
      ,L.LGRUCTS                    AS LGRX_UCTIMESTAMP
--    ,HEX(L.LGRSRBA)               AS START_RBA      
--    ,HEX(L.LGRSPBA)               AS END_RBA        
      ,HEX(L.LGRSLRSN)              AS START_LRSN     
      ,HEX(L.LGRELRSN)              AS END_LRSN       
      ,'TS'                                           
      ,L.LGRMEMB                    AS DM             
FROM BOXWELL.SYSLGRNX2    L                           
    ,SYSIBM.SYSTABLESPACE TS                          
WHERE TS.DBID    = L.LGRDBID                           
  AND TS.PSID    = L.LGRPSID                          
  AND L.LGRSLRSN = L.LGRELRSN                         
--AND L.LGRSRBA  = L.LGRSPBA                          
UNION ALL                                             
SELECT L.LGRDBID                    AS DBID           
      ,SUBSTR(IX.DBNAME, 1, 8)      AS DBNAME         
      ,L.LGRPSID                    AS ID             
      ,L.LGRNEGPSID                 AS NEG_PSID       
      ,SUBSTR(IX.INDEXSPACE, 1, 8)  AS SPACE          
      ,L.LGRPART                    AS PART           
      ,L.LGRUCTS                    AS LGRX_UCTIMESTAMP
--    ,HEX(L.LGRSRBA)               AS START_RBA      
--    ,HEX(L.LGRSPBA)               AS END_RBA        
      ,HEX(L.LGRSLRSN)              AS START_LRSN     
      ,HEX(L.LGRELRSN)              AS END_LRSN       
      ,'IX'                                           
      ,L.LGRMEMB                    AS DM             
FROM BOXWELL.SYSLGRNX2    L                            
    ,SYSIBM.SYSINDEXES    IX                          
WHERE IX.DBID    = L.LGRDBID                          
  AND IX.ISOBID  = L.LGRPSID                          
  AND L.LGRSLRSN = L.LGRELRSN                         
--AND L.LGRSRBA  = L.LGRSPBA                          
ORDER BY 7                                            
WITH UR                                               
  ;                                                   

I get data like this here on my test DS system:

---------+---------+---------+---------+---------+---------+---------+---------+
  DBID  DBNAME        ID  NEG_PSID  SPACE       PART  LGRX_UCTIMESTAMP         
---------+---------+---------+---------+---------+---------+---------+---------+
   320  R510D0DE     122  N         R510S23        1  2019-09-30-07.53.56.670000
   320  R510D0DE     122  N         R510S23        1  2019-09-30-07.53.58.880000
   348  MVNXTEST     122  N         MVNXS23        1  2019-11-07-11.24.59.920000
   348  MVNXTEST     122  N         MVNXS23        1  2019-11-07-11.24.59.950000
   348  MVNXTEST     389  N         MVNXS001       1  2019-11-07-14.48.37.780000
   348  MVNXTEST     389  N         MVNXS001       1  2019-11-07-14.48.38.890000
   348  MVNXTEST     398  N         MVNXS002       1  2019-11-07-14.48.40.050000
   348  MVNXTEST     398  N         MVNXS002       1  2019-11-07-14.48.40.090000
   348  MVNXTEST     429  N         MVNXS007       1  2019-11-07-14.48.42.200000
   348  MVNXTEST     389  N         MVNXS001       1  2020-02-21-11.25.54.480000
   348  MVNXTEST     389  N         MVNXS001       1  2020-02-21-11.25.55.560000
   348  MVNXTEST     398  N         MVNXS002       1  2020-02-21-11.25.56.680000
   348  MVNXTEST     398  N         MVNXS002       1  2020-02-21-11.25.57.700000
-------+---------+---------+---------+---------+------
START_LRSN            END_LRSN                      DM
-------+---------+---------+---------+---------+------
00D6CCFFBDC159BB2E00  00D6CCFFBDC159BB2E00  TS       2
00D6CCFFBFC00FF01200  00D6CCFFBFC00FF01200  TS       2
00D6FCF5C92F2D4F5200  00D6FCF5C92F2D4F5200  TS       2
00D6FCF5C976B5E74600  00D6FCF5C976B5E74600  TS       2
00D6FD234DC040BFE200  00D6FD234DC040BFE200  TS       1
00D6FD234EBD47CE8400  00D6FD234EBD47CE8400  TS       1
00D6FD235042B0392800  00D6FD235042B0392800  TS       2
00D6FD2350B1394C5800  00D6FD2350B1394C5800  TS       2
00D6FD2351B0B0B72A00  00D6FD2351B0B0B72A00  TS       2
00D7823BB0FED94EA800  00D7823BB0FED94EA800  TS       1
00D7823BB1C7253F4A00  00D7823BB1C7253F4A00  TS       1
00D7823BB2F61E0A3400  00D7823BB2F61E0A3400  TS       2
00D7823BB32BFC34CE00  00D7823BB32BFC34CE00  TS       2

What I can see from this, is that for any DDL CREATE with LOGGED attribute, or any LOAD REPLACE LOG(NO) utility, you get an entry that you can also find in SYSCOPY. The RECOVER utility might well need this data as well if you are rolling forward. 

The oldest are the bestest!

Finally, a simple “review” of the SYSLGRNX data sorted by LGRNX_UCTIMESTAMP. Just so you get an overview, this SQL has a FETCH FIRST 500 ONLY but feel free to change that number!

-- OLDEST VALID RECORDS IN SYSLGRNX                   
SELECT L.LGRDBID                    AS DBID           
      ,SUBSTR(TS.DBNAME, 1, 8)      AS DBNAME         
      ,L.LGRPSID                    AS ID             
      ,L.LGRNEGPSID                 AS NEG_PSID       
      ,SUBSTR(TS.NAME, 1, 8)        AS SPACE           
      ,L.LGRPART                    AS PART           
      ,L.LGRUCTS                    AS LGRX_UCTIMESTAMP
--    ,HEX(L.LGRSRBA)               AS START_RBA      
--    ,HEX(L.LGRSPBA)               AS END_RBA        
      ,HEX(L.LGRSLRSN)              AS START_LRSN     
      ,HEX(L.LGRELRSN)              AS END_LRSN       
      ,L.LGRMEMB                    AS DM             
      ,'TS'                                           
FROM BOXWELL.SYSLGRNX2    L                           
    ,SYSIBM.SYSTABLESPACE TS                          
WHERE TS.DBID     = L.LGRDBID                         
  AND TS.PSID     = L.LGRPSID                         
--AND NOT L.LGRSRBA  = L.LGRSPBA                      
  AND NOT L.LGRSLRSN = L.LGRELRSN                     
UNION ALL                                             
SELECT L.LGRDBID                    AS DBID           
      ,SUBSTR(IX.DBNAME, 1, 8)      AS DBNAME         
      ,L.LGRPSID                    AS ID             
      ,L.LGRNEGPSID                 AS NEG_PSID       
      ,SUBSTR(IX.INDEXSPACE, 1, 8)  AS SPACE          
      ,L.LGRPART                    AS PART           
      ,L.LGRUCTS                    AS LGRX_UCTIMESTAMP
--    ,HEX(L.LGRSRBA)               AS START_RBA      
--    ,HEX(L.LGRSPBA)               AS END_RBA        
      ,HEX(L.LGRSLRSN)              AS START_LRSN     
      ,HEX(L.LGRELRSN)              AS END_LRSN       
      ,L.LGRMEMB                    AS DM             
      ,'IX'                                           
FROM BOXWELL.SYSLGRNX2    L                           
    ,SYSIBM.SYSINDEXES    IX                          
WHERE IX.DBID   = L.LGRDBID                           
  AND IX.ISOBID = L.LGRPSID                           
--AND NOT L.LGRSRBA  = L.LGRSPBA                      
  AND NOT L.LGRSLRSN = L.LGRELRSN                     
ORDER BY LGRX_UCTIMESTAMP                             
FETCH FIRST 500 ROWS ONLY                              
WITH UR                                               
  ;                                                   

Here you can see that I rarely ever image copy the Db2 catalog…

---------+---------+---------+---------+---------+---------+---------+---------+
  DBID  DBNAME        ID  NEG_PSID  SPACE       PART  LGRX_UCTIMESTAMP         
---------+---------+---------+---------+---------+---------+---------+---------+
     6  DSNDB06     2002  N         SYSTSIPT       1  2019-09-28-13.45.32.200000
     6  DSNDB06     2000  N         SYSTSIXS       1  2019-09-28-13.45.32.250000
     6  DSNDB06     2016  N         SYSTSTSP       1  2019-09-28-13.46.59.970000
     6  DSNDB06     2012  N         SYSTSTPT       1  2019-09-28-13.46.59.970000
     6  DSNDB06     2034  N         SYSTSSTG       1  2019-09-28-13.48.02.210000
     1  DSNDB01      148  N         DSNSPT02       0  2019-09-28-13.51.16.620000


-------+---------+---------+---------+---------+------
START_LRSN            END_LRSN                  DM   
-------+---------+---------+---------+---------+------
00D6CACA98D8486B6C00  00D6CACCE2F5E2B14600       1  TS
00D6CACA99560741CC00  00D6CACCE2F9B5CCD800       1  TS
00D6CACAEC7D724A9600  00D6CACCE2F31722C800       1  TS
00D6CACAEC7C8EA08A00  00D6CACCE2F7C7553800       1  TS
00D6CACB27FFD7A40200  00D6CACCE2F214B70E00       1  TS
00D6CACBE127037D0A00  00D6CACCE31529801400       1  IX
00D6CACBE1279F5EA000  00D6CACCE31662377800       1  IX

What’s in a name?

Looking at the outputs you can see the NEG_PSID column. This is set based on whether or not the original PSID in the SYSLGRNX was negative or not. If it is negative it means you are dealing with a CLONE object, so buyer beware on those entries!

The DM column is the Data-sharing member number, it is 0 (zero) for Non-DS and 1 to 32 for DS systems. Sometimes you can get a clue about why an entry is there when you can link it to a member. Use the -DISPLAY GROUP command to get the number for your member:

DSN7100I  -SC10 DSN7GCMD                                            
*** BEGIN DISPLAY OF GROUP(GSC10C11) CATALOG LEVEL(V12R1M505)       
                  CURRENT FUNCTION LEVEL(V12R1M505)                 
                  HIGHEST ACTIVATED FUNCTION LEVEL(V12R1M505)       
                  HIGHEST POSSIBLE FUNCTION LEVEL(V12R1M506)        
                  PROTOCOL LEVEL(2)                                  
                  GROUP ATTACH NAME(SC1 )                           
---------------------------------------------------------------------
DB2          SUB                     DB2    SYSTEM    IRLM          MEMBER   ID  SYS  CMDPREF   STATUS   LVL    NAME      SUBSYS IRLMPROC
-------- --- ---- --------  -------- ------ --------  ----   --------
MEMSC10    1 SC10 -SC10     ACTIVE   121507 S0W1      JC10   SC10IRLM
MEMSC11    2 SC11 -SC11     QUIESCED 121507 S0W1      JC11   SC11IRLM
---------------------------------------------------------------------

Here the ID column is the member number. It is possible you might have DM ids for members that no longer exist… these must also be cleared out as well!

As always, any questions or comments would be most welcome!

TTFN,

Roy Boxwell

As noted in my last newsletter, the virtual world is influencing us all now. The IDUG North America 2020 had to “go virtual” and here are a few points I picked up over the last weeks while manning our virtual booth and attending virtual sessions. These are just randomly sorted and there were a ton of other interesting IDUG presentations that I recommend you try and read!

Always on

IFCID 376 for incompatible changes should be on all the time in all systems. The reasoning behind this, is quite simple: The longer ahead you are seeing problems the more time you have to get an application fix. Naturally, you can and should use the APPLCOMPAT feature to lock down applications, but when you wish to use newer features you cannot rely on this anymore. This leads to the paradox of don’t go there yet, but go there as fast as you can! Judicious use of this IFCID could well save you an embarrassment or two!

PBGs forever

Well, only in small doses! The idea with these, is to use them as a plug-replacement for segmented and simple spaces, but only with MAXPARTITIONS set to 1 – A bit strange though, as that caps your maximum size quite a bit, but the reasoning here is to use the DSSIZE to actually control the size of your objects and not let them grow and grow and grow.

FTB switch off

Sadly, the use of FTB Fast Traversal Blocks is not recommended at this time due to ongoing issues. In fact, a Red Alert came during the IDUG noting that we should also switch off Insert Algorithm 2 (IAG2) – there is even an APAR out there to change the IAG2 default to “switch off by default”. This is a real shame, as these two features were always touted as the next best thing after sliced bread. Oh well…

New Db2 Version

Yes indeed! – Spotted in at least one IBM presentation were tantalizing hints about a new release coming out. Remember when Db2 12 came out, the list of reasons for a new version/release were just “Major control block changes or a New PL/X compile to take advantage of any new architecture level” – Well, a few months ago, a new reason appeared: “Changes to terms and conditions” which I quite liked, as that is pretty transparent! Now I have seen a new one: “Non-UTS Tablespaces will be retired in the future as part of a future function level or VNEXT“

Yep, VNEXT is back – Yoo-hoo!

RUNSTATS you like

Not really learnt by me, but a few people mentioned that when you REORG only collect stats at the TS level if things have really changed. If you REORG Indexes *never* collect inline stats. Why? These days the reasons for REORG can be for actioning DDL changes and, in these cases, spending all the CPU to re-get the stats data is probably pointless. On the other hand, with indexes you can get statistical data drift, and so it makes *no* sense to do inline stats with index reorgs at all.

RTS Time travel

In Db2 12 we got a whole bunch of SYSIBM.SYSxxxxx_H style tables. We got two for the RTS tables, and you must “activate” them as they are not active by default. There are also no indexes and no automatic deletion, so when you do the ALTERs make sure you have thought about, and created, performance indexes and a purge process for old, dead data. These tables can give very good insights into the overall performance and usage of all the objects in a Db2 system.

Distributed Level

For access Db2 12 we got a little update about exactly which levels are required:

Need V11.1 FP1 (JCC driver level 3.72 to 4.22) to exploit beyond V12R1M500 (or NFM mode), otherwise ANY supported level should work

• Need clientApplCompat driver parameter to exploit features in M501+

• Made optional with APAR PH08482 for down-level clients and Db2 Connect Servers

This is still a major cause of confusion even today, nearly three years after Db2 12 was first released!

Buzzword Bingo

Then I started the Jenkins, Pipeline learning curve, but I must admit, as an old COBOL guy, all the new stuff is a bit strange. GiT is still a term of abuse for me! But the presentation went into near epic depth about how to actually get it all working… I added this to my bucket list! I especially liked the SCORCHING JIT compiler optimization setting.

REORG Rules

The use of REORGINSERTS is no longer recommended as a reason for a REORG. A few releases ago, the REORGUPDATES reason was also removed. Now we only have REORGDELETES from the classic counters in the RTS that should be used as a trigger for a TP/TS level REORG.

So, did you learn anything new at the IDUG? Is there anything I missed?

As always, any questions or comments would be most welcome and I would love to “virtually” meet you all soon!

TTFN,

Roy Boxwell

Due to the Corona virus a whole bunch of us IT folk have been getting a lot of “virtual” recently! We are viewed as being “system relevant” and some of us *must* work from home to keep the systems rolling. This month I thought I would walk down the road we at Software Engineering and SEGUS have all about virtual, and simulated, Db2.

Virtual, really?

The first thing is to think what do we mean by “virtual”, when I hold a live webinar I am talking live and you are listening live, there is nothing really virtual about it is there? The same is true about Db2 sub-systems. If you wish to test a major upgrade of your software (or even just a FL switch) you really should get it all tested in a virtual world before letting it loose in the real world!

How to get there?

You need a complete copy of your Db2 Catalog and Directory. That is enough – User data is not required and actually could be classified as dangerous from the Audit point of view! How you get this done is your business but I use ICE our InstantCloning Expert to get the job done really really fast!

Enough?

Nope, once you have got this data copied across you need to “virtualize” the ZPARMS, bufferpools and even the hardware you currently have in production. IBM came up with an interesting way of doing this with hideously complicated HEX updates of EXPLAIN columns but I use VOX our VirtualOptimizer Expert to do all this for me!

Done yet?

Of course not! What is the good of having a virtual production system when you have no SQL to actually run on it? Just doing an EXPLAIN is ok of course – see later – but actually running the SQL that runs on the productive system should be your goal. For this I use the WLX (SQL WorkLoadExpert) tool to collect as much SQL as I like on production and then use this “workload” on my virtual production to actually check what happens when software is running.

Finished?

Still not yet! The next part of the puzzle is to also EXPLAIN all of the dynamic and static SQL both before and after the “change” whatever that change may be! You then compare the outputs of about 120,000 explains to see where anything diverges in its access plan. I use our BIX BindImpact Expert to do all this automatically for dynamic and static SQL.

You must be done by now!

Never! The final piece of the virtual puzzle is now to switch on the IFCID 376 while running the SQL workloads to give you a heads up if any application changes are rolling down the road towards you. We are all aware that application changes take a while to get approved and done and the further out you can see the obstacle the easier it is to plan and avoid the crash!

It’s all too much!

Well, you can even do a “cheaper” variant – Instead of cloning the whole Db2 subsystem just copy the DDL, using DEFINE NO of course, and the production Statistics using our ProductionSimulator tool to a sand-box style Db2 subsystem. You can also rename the objects on the fly and this all enables EXPLAIN processing to be done without a real workload. This will only show you changing access plans for your static and dynamic SQL but is clearly much better than nothing!

Sounds Cool!

It is! You have a nearly 100% automated method to completely check any Db2 subsystem for any changes and using our advanced checking algorithms we can weed out all the background noise to let you see just “the facts ma’am”

But Virtual with each other?

As you are probably aware the IDUG NA 2020 has gone Virtual on us all and so it has a bunch of live sessions and a whole raft of pre-recorded sessions. Happily they also Include mine – released on the 3^rd August – ”A DBA’s Got to Know Their Limitations!” I am also holding a VSP about Zowe (The zGUI r(evolution) – What is ZOWE going to do for me?) on Monday 10^th August at 4pm EDT (22:00 Berlin) with live Q&A.

Please visit our virtual Booth at

https://www.idug.org/page/expo-hall

Click on the SEGUS booth, middle row right hand side,  then scroll down to register and get the chance to win an iPad!!! A real one as well…

As always, any questions or comments would be most welcome and I would love to “virtually” meet you all soon!

TTFN,

Roy Boxwell

I thought I would create a newsletter out of all the stuff I never knew before that I have learnt over the last few years. Maybe you’re very clever and know all of this already? But if you don’t, that’s OK too.

Sliding Scale

We have a nice piece of software called Space AssuranceExpert for Db2 z/OS (SAX), that guarantees that you do not hit any of the various limits within space management, and beyond, on Db2. When IBM introduced their sliding scale of secondary allocations it collided with our variant. (We use “seed” values to get, what I believe, is a much better utilization of available space, but I digress…) so we changed our system to check the MGEXTSZ. This, by the way, is one of the worst documented ZPARMs I have ever tried to understand! The documentation states “MGEXTSZ default YES – If SECQTY is greater than 0 then the sliding scale will be used”. So, when you create an object using SECQTY -1 or just not using SECQTY (also gives you -1) it would appear that the sliding scale is switched off… Well what IBM really meant was the SECQTY of the allocated VSAM dataset. Naturally this is either zero (no secondaries for you!) or a positive number! If you look at the Db2 Catalog tables you will also see that the fields you use as PRIQTY and SECQTY are actually stored as PQTY, SQTY, and SQTYI so, technically speaking, IBM are correct. The only case where MGEXTSZ is actually used is if you *do* have a positive value in the SQTY/SQTYI field. If MGEXTSZ is YES then this value is ignored and the sliding scale is used, if MGEXTSZ is NO then the value of SQTY/SQTYI, converted back to kilobytes of course, will be used.

An extent by any other name

Talking about SAX: I also saw something quite amazing as a customer allocated a PBG space with 11 maxpartitions and a DSSIZE of 8GB. He managed to hit maxpartitions without getting warned by SAX. “Impossible”, I said! There is no way it can allocate one dataset at 8GB with *no* extents… Yet this customer had somehow managed it… Their EAVs are so huge, they could allocate without a *single* extent being requested. Now, I always thought that every dataset that is allocated gets one extent, and that this extent would trigger the IFCID 258. I was wrong! The creation of an LDS, if it does not require more than one extent, does not trigger this IFCID. It does, however, trigger the IFCID 97 (LDS Creation).So I added this IFCID into our SAX system too and now we are able to catch LDS problems even with *massive* disks!

Index for DGTT

This brings me to the next item. If you create a DGTT:

DECLARE GLOBAL TEMPORARY TABLE T (COL1 CHAR(1) NOT NULL);

You get *no* dataset being created after all why? It just goes in DSNDB07 etc. so a create makes no sense, right? 

SAX has seen nothing either:

O2RT-SU04-011I: 10:01:09 - Datasets will be processed now                  

Now create an index on that DGTT

DECLARE GLOBAL TEMPORARY TABLE T (COL1   CHAR(1) NOT NULL);
CREATE INDEX I ON SESSION.T (COL1 ) ;        

Now you do get a dataset being allocated and deleted!

O2RT-SU04-011I: 10:01:09 - Datasets will be processed now                 
O2RT-SU04-024I: LDS creation for DB2DC1.DSNDBD.DSNDB07.TIX12768.I0001.A001 
O2RT-SU04-011I: 14:41:58 - Datasets will be processed now                 

Notice the Database name “DSNDB07”?- This tells you that it is creating a dataset in the work database but this is not really documented anywhere. Naturally, if you think about it for a while, it *must* create a VSAM LDS to simply hold the data, as it cannot use another one of the DSNDB07 tablespaces for that purpose! This is something to bear in mind when using indexes on DGTTs. You get dataset create/delete elapsed time…

RUNSTATS for XML

In Db2 12, FL100 RUNSTATS was enhanced to enable Frequency and Histogram data to be collected into the SYSKEYTGTDIST catalog table. I created an XML Object:

DROP DATABASE      "BOXWELLX" ;                                     
COMMIT ;                                                           
  CREATE DATABASE "BOXWELLX"                                        
         BUFFERPOOL BP0                                             
         INDEXBP    BP0                                             
         STOGROUP   SYSDEFLT                                        
         CCSID UNICODE                                              
;                                                                  
COMMIT ;                                                           
  CREATE       TABLESPACE "XMLCUSTO"                                
         IN "BOXWELLX"                                             
         USING STOGROUP SYSDEFLT                                    
         PRIQTY      720                                            
         SECQTY      720                                            
         ERASE NO                                                  
         DEFINE YES                                                
         FREEPAGE    0                                              
         PCTFREE     5                                              
         GBPCACHE CHANGED                                           
         TRACKMOD YES                                               
         LOG      YES                                               
         MAXPARTITIONS  6                                           
         BUFFERPOOL BP0                                             
         LOCKSIZE ROW                                               
         LOCKMAX SYSTEM                                             
         CLOSE YES                                                 
         COMPRESS NO                                               
         CCSID UNICODE                                              
         MAXROWS  255                                               
         SEGSIZE    4                                               
;                                                                  
COMMIT ;                                                           
  CREATE TABLE "BOXWELL"."XMLCUSTOMER"                              
  ("CID"                            BIGINT                  NOT NULL
  ,"INFO"                           XML                             
  ,CONSTRAINT CID                                                  
   PRIMARY KEY                                                     
  ("CID"                                                           
  )                                                                 
  )                                                                
  IN  "BOXWELLX"."XMLCUSTO"                                         
  AUDIT NONE                                                       
  DATA CAPTURE NONE                                                
  CCSID UNICODE                                                    
      VOLATILE CARDINALITY                                          
  APPEND NO                                                         
;                                                                  
  COMMIT ;                                                   
    CREATE UNIQUE INDEX                "BOXWELL"."XML_PRIMARY"
           ON "BOXWELL"."XMLCUSTOMER"                         
           ("CID"                            ASC              
           )                                                 
           CLUSTER                                            
           USING STOGROUP SYSDEFLT                            
           PRIQTY      720                                    
           SECQTY      720                                    
           ERASE NO                                           
           FREEPAGE    0                                      
           PCTFREE    10                                      
           GBPCACHE CHANGED                                   
           DEFINE YES                                         
           COMPRESS NO                                        
           BUFFERPOOL BP0                                     
           CLOSE NO                                           
           PIECESIZE    2 G                                   
           COPY NO                                            
  ;                                                          
  COMMIT ;                                                   
    CREATE INDEX                       "BOXWELL"."CUST_TYPE"  
           ON "BOXWELL"."XMLCUSTOMER"                         
           ("INFO"                                            
           )                                                 
    GENERATE KEY USING XMLPATTERN                             
/customerinfo/phone/@type'                                    
    AS SQL VARCHAR( 128)                                      
           NOT CLUSTER                                        
           NOT PADDED                                         
           USING STOGROUP SYSDEFLT                            
           PRIQTY        4                                    
           SECQTY        4                                    
           ERASE NO                                           
           FREEPAGE    0                                      
           PCTFREE    10                                      
           GBPCACHE CHANGED                                   
           DEFINE YES                                         
           COMPRESS NO                                        
           BUFFERPOOL BP0                                     
           CLOSE YES                                          
           PIECESIZE    2 G                                   
           COPY NO                                           
  ;                                                          
    CREATE INDEX                       "BOXWELL"."CUST_ZIP"   
           ON "BOXWELL"."XMLCUSTOMER"                         
           ("INFO"                                            
           )                                                
    GENERATE KEY USING XMLPATTERN                            
//pcode-zip'                                                
    AS SQL DECFLOAT(34)                                      
           NOT CLUSTER                                       
           NOT PADDED                                        
           USING STOGROUP SYSDEFLT                           
           PRIQTY        4                                   
           SECQTY        4                                   
           ERASE NO                                          
           FREEPAGE    0                                     
           PCTFREE    10                                     
           GBPCACHE CHANGED                                  
           DEFINE YES                                        
           COMPRESS NO                                       
           BUFFERPOOL BP0                                    
           CLOSE YES                                         
           PIECESIZE    2 G                                  
           COPY NO                                           
  ;                                                         
  COMMIT ;                                                  
    CREATE INDEX                       "BOXWELL"."CUST_PHONE"
           ON "BOXWELL"."XMLCUSTOMER"                        
           ("INFO"                                           
           )                                                
    GENERATE KEY USING XMLPATTERN                            
/customerinfo/phone'                                         
    AS SQL VARCHAR( 128)                                     
           NOT CLUSTER                                       
           NOT PADDED                                        
           USING STOGROUP SYSDEFLT                           
           PRIQTY        4                                   
           SECQTY        4                                   
           ERASE NO                                          
           FREEPAGE    0                                     
           PCTFREE    10                                     
           GBPCACHE CHANGED                                  
           DEFINE YES                                        
           COMPRESS NO                                       
           BUFFERPOOL BP0                                    
           CLOSE YES                                         
           PIECESIZE    2 G                                  
           COPY NO                                           
  ;                                                         
  COMMIT ;                                                

Phew!

Then I needed to insert a few rows. Here are the five I used for my tests:

SET CURRENT SQLID = 'BOXWELL' ;                                  
COMMIT ;                                                         
DELETE FROM BOXWELL.XMLCUSTOMER ;                                
COMMIT ;                                                         
INSERT INTO BOXWELL.XMLCUSTOMER VALUES (06001, XMLPARSE(DOCUMENT '
<customerinfo><name>Justin Horovitz</name>                        
<addr country="United States"><street>327 Ramses Ave</street>    
<city>Glendale</city><prov-state>California</prov-state>         
<pcode-zip>91208-06001</pcode-zip></addr>                        
<phone type="work">818-956-06001</phone></customerinfo>'));      
INSERT INTO BOXWELL.XMLCUSTOMER VALUES (06002, XMLPARSE(DOCUMENT '
<customerinfo><name>Matthew Broad</name>                         
<addr country="United States"><street>808 Mayo St</street>       
<city>Burbank</city><prov-state>California</prov-state>          
<pcode-zip>9150600-06002</pcode-zip></addr>                      
<phone type="work">818-541-06002</phone></customerinfo>'));      
INSERT INTO BOXWELL.XMLCUSTOMER VALUES (06003, XMLPARSE(DOCUMENT '
<customerinfo><name>Laura McCarthy</name>                        
<addr country="United States"><street>5224 Grover Court</street> 
<city>San Jose</city><prov-state>California</prov-state>         
<pcode-zip>95123-06003</pcode-zip></addr>                        
<phone type="work">408-956-06003</phone></customerinfo>'));      
INSERT INTO BOXWELL.XMLCUSTOMER VALUES (06004, XMLPARSE(DOCUMENT '
<customerinfo><name>Mira Glass</name>                            
<addr country="United States"><street>444 Valencia St</street>   
<city>San Francisco</city><prov-state>California</prov-state>    
<pcode-zip>94110-06004</pcode-zip></addr>                        
<phone type="work">415-762-06004</phone></customerinfo>'));      
INSERT INTO BOXWELL.XMLCUSTOMER VALUES (06005, XMLPARSE(         
'<customerinfo><name>Amir Malik</name>                           
<addr country="United States"><street>555 Bailey Ave</street>    
<city>San Jose</city><prov-state>California</prov-state>         
<pcode-zip>95141-06005</pcode-zip></addr>                        
<phone type="work">408-555-06005</phone></customerinfo>'));      
INSERT INTO BOXWELL.XMLCUSTOMER VALUES (06995, XMLPARSE(         
'<customerinfo><name>Amir Malik</name>                           
<addr country="United States"><street>555 Bailey Ave</street>    
<city>San Jose</city><prov-state>California</prov-state>         
<pcode-zip>95141-06995</pcode-zip></addr>                         
<phone type="work">408-555-06995</phone></customerinfo>'));      INSERT INTO BOXWELL.XMLCUSTOMER VALUES (06996, XMLPARSE(         
'<customerinfo><name>Kathy Smith</name>                          
<addr country="Canada"><street>25 EastCreek</street>             
<city>Toronto</city><prov-state>Ontario</prov-state>             
<pcode-zip>M8X-3T6-06996</pcode-zip></addr>                      
<phone type="work">416-555-06996</phone></customerinfo>'));      
INSERT INTO BOXWELL.XMLCUSTOMER VALUES (06997, XMLPARSE(     
'<customerinfo><name>Jim Noodle</name>                       
<addr country="Canada"><street>25 EastCreek</street>         
<city>Markham</city><prov-state>Ontario</prov-state>         
<pcode-zip>N9C-3T6-06997</pcode-zip></addr>                  
<phone type="work">905-555-06997</phone></customerinfo>'));  
INSERT INTO BOXWELL.XMLCUSTOMER VALUES (06998, XMLPARSE(     
'<customerinfo><name>Anant Jhingran</name>                   
<addr country="United States"><street>555 Bailey Ave</street>
<city>San Jose</city><prov-state>California</prov-state>     
<pcode-zip>95141-06998</pcode-zip></addr>                    
<phone type="work">408-555-06998</phone>                     
<phone type="home">416-555-06998</phone>                     
<phone type="cell">905-555-06998</phone>                     
<phone type="cottage">613-555-06998</phone></customerinfo>'));
INSERT INTO BOXWELL.XMLCUSTOMER VALUES (06999, XMLPARSE(     
'<customerinfo><name>Bert and Ernie Inc.</name>              
<addr country="Canada"><street>1 Yonge Street</street>       
<city>Toronto</city><prov-state>Ontario</prov-state>         
<pcode-zip>M5W-IE6-06999</pcode-zip></addr>                  
<phone type="work">416-555-06999</phone></customerinfo>'));  
COMMIT ;                                                     

If you run a RUNSTATS on that tablespace:

RUNSTATS TABLESPACE BOXWELLX.XMLCUSTO         
   TABLE(ALL)                                 
   INDEX(ALL                                  
   KEYCARD                                    
         FREQVAL   NUMCOLS 0001 COUNT   10 MOST
         FREQVAL   NUMCOLS 0002 COUNT   10 MOST
         HISTOGRAM NUMCOLS 0001               
         HISTOGRAM NUMCOLS 0002               
        )                                     
   SHRLEVEL CHANGE                            
   REPORT YES                                 
   UPDATE ALL                                  
   SORTDEVT SYSALLDA                          
   SORTNUM  0004                              

In Db2 11 and Db2 12 I got *exactly* the same output. Even though I have created three XML indexes that *should* have got the SYSKEYTGTDIST statistics! The Db2 Catalog entries for these three indexes also look a bit “odd” as the UNIQUERULE column is set to “D” (I was expecting an “X” for XML!) and the IX_EXTENSION_TYPE is set to “V” (XML). Interestingly the implicit I_DocId Index *is* a UNIQUERULE ‘X’ and IX_EXTENSION_TYPE blank entry even though, in my eyes, it is *not* an XML index!

Now, when you create the test objects listed above, you also get a whole bunch of implicit objects. The tablespace (mine was called XXML0000), which contains the implicit table BOXWELL.XXMLCUSTOMER with five columns and an implicit I_NodeId index.

If you now run this RUNSTATS:

RUNSTATS TABLESPACE BOXWELLX.XXML0000          
   TABLE(ALL)                                   
   INDEX(ALL                                   
   KEYCARD                                     
         FREQVAL   NUMCOLS 0001 COUNT   10 MOST
         FREQVAL   NUMCOLS 0002 COUNT   10 MOST
         HISTOGRAM NUMCOLS 0001                
         HISTOGRAM NUMCOLS 0002                
        )                                      
   SHRLEVEL CHANGE                             
   REPORT YES                                  
   UPDATE ALL                                   
   SORTDEVT SYSALLDA                           
   SORTNUM  0004                             

In Db2 11 it moans about the FREQVAL and HISTOGRAM keywords:

DSNU1354I -QB1A 120 09:14:41.60 DSNUSIIX - KEYWORD KEYCARD IS NOT SUPPORTED FOR XML OBJECTS. THE KEYWORD IS IGNORED 
DSNU1354I -QB1A 120 09:14:41.60 DSNUSIIX - KEYWORD FREQVAL IS NOT SUPPORTED FOR XML OBJECTS. THE KEYWORD IS IGNORED 
DSNU1354I -QB1A 120 09:14:41.60 DSNUSIIX - KEYWORD HISTOGRAM IS NOT SUPPORTED FOR XML OBJECTS. THE KEYWORD IS IGNORED

And it does not create the SYSKEYTGTDIST entries. In Db2 12 it moans, but a bit differently:

DSNU1354I -DC10 120 09:13:07.35 DSNUSIIX - KEYWORD KEYCARD IS NOT SUPPORTED FOR CERTAIN TYPES OF XML OBJECTS. 
THE KEYWORD IS IGNORED.                                                                                        DSNU1354I -DC10 120 09:13:07.35 DSNUSIIX - KEYWORD FREQVAL IS NOT SUPPORTED FOR CERTAIN TYPES OF XML OBJECTS. 
THE KEYWORD IS IGNORED.                                                                                       DSNU1354I -DC10 120 09:13:07.35 DSNUSIIX - KEYWORD HISTOGRAM IS NOT SUPPORTED FOR CERTAIN TYPES OF XML OBJECTS.
THE KEYWORD IS IGNORED.                                                                                        

In my case, right at the end of the job, were the successful results:

DSNU1353I -DC10 120 09:13:07.43 DSNUSUKD - SYSKEYTGTDIST CATALOG STATISTICS FOR                                 

                                       CUST_TYPE  KEYSEQ 0001                                                  

                              FREQUENCY           COLVALUE                                                     

                              ---------           --------                                                     

                              7,6923076923076E-01 X'776F726B0000000000000000000000000000000000000000000000000000

                                                    000000000000000000000000000000000000000000000000000000000000

                                                    000000000000000000000000000000000000000000000000000000000000

                                                    000000000000000000000000000000000000000000000000000000000000

                                                    00000000000000000004'                                      

And

DSNU1356I -DC10 120 09:13:07.44 DSNUSUKD - SYSKEYTGTDIST CATALOG HISTOGRAM STATISTICS FOR CUST_TYPE KEYSEQ 1 QUANTILE 1

                              LOWVALUE          = X'63656C6C0000000000000000000000000000000000000000000000000000      

                                                    000000000000000000000000000000000000000000000000000000000000      

                                                    000000000000000000000000000000000000000000000000000000000000      

                                                    000000000000000000000000000000000000000000000000000000000000      

                                                    00000000000000000004'                                             

                              HIGHVALUE         = X'63656C6C0000000000000000000000000000000000000000000000000000      

                                                    000000000000000000000000000000000000000000000000000000000000      

                                                    000000000000000000000000000000000000000000000000000000000000      

                                                    000000000000000000000000000000000000000000000000000000000000      

                                                    00000000000000000004'                                             

                              CARDF             = 1,0E+00                                                              

                              FREQUENCYF        = 7,6923076923076E-02                                                 

Most of the other columns were also reported, apart from one set of data which was strangely missing. Can you guess which one and why? Scroll back up to the creation of the indexes and you will see that I made one resolve into a DECFLOAT. This is*not* supported and so is simply ignored!

Where’s the beef?

Well, the interesting point about this “voyage of XML discovery” is you *must* run a RUNSTATS on a different tablespace than where the indexes were created. Does your current RUNSTATS creation software handle this? Or are you missing out on possible XMLEXISTS exploitation due to missing RUNSTATS? From the “What’s New?”:

Db2 can use frequency and histogram statistics for XML indexes to estimate the filter factor of XMLEXISTS predicates when the following conditions are true:

The predicate is one of the following types:

• Equality predicates: =
• Range predicates: >, >=, >=, <=
• The right side of the predicate is a literal value.

Our Freeware StatisticsHealthCheck was just upgraded to version 3.4 to handle exactly this “problem”.

Naturally, if you can read, it is an advantage and the IBM Db2 documentation clearly states:

“XML indexes are related to XML tables, and not to the associated base tables. If you specify a base table space and an XML index in the same RUNSTATS control statement, Db2 generates an error. When you run RUNSTATS against a base table, RUNSTATS collects statistics only for indexes on the base table, including the document ID index.”

Db2 11/12 – Utilities – RUNSTATS – Collection of statistics on XML objects

I had never noticed that paragraph before… Shame on me!

Caught in a BIND

Back in Db2 10, IBM introduced the BIND PACKAGE(xxx) EXPLAIN(ONLY) syntax to fill the PLAN_TABLE and its buddies, but without actually doing a BIND. Very cool feature – until you find out that it jolly well *does* do a BIND, with all of the locking implications that we all know and hate, before casually doing a ROLLBACK as if “these are not the droids you are looking for”. One of my pet bugbears I can tell you!

And this can really hurt in production!

Learning Curve

What have you learned from Db2 over the years that was strange/obtuse or just downright weird? Don’t be shy! We’d love to hear about it.

As always, any questions or comments would be most welcome!

TTFN,

Roy Boxwell