How to backup only key data sets on z/OS

I’ve been backing up my datasets on z/OS, and wondered what the best way of doing it was.

I wanted to backup datasets containing data I wanted to keep, but did not want to backup other data sets which could easily be recreated, such as IPCS dump dataset, the output of compiles, or the SMF records.

DFDSS has a backup and restore program which is very powerful.  With it you can

• Process data sets under a High Level Qualifier – include or exclude data sets.
• Backup only changed data sets
• Backup individual files in a ZFS or USS – but this is limited, you have to explicitly specify the files you want to backup.   You cannot backup a directory

You cannot backup individual members of a PDS(E).   You have to backup the whole PDS(E),   If you need to restore a member, restore the backup with a different HLQ and select the members from that.

What should I use?

I tend to use XMIT and DFDSS – the Storage Management component on z/OS. This tends to be used by the data managers as it can backup groups of data sets, volumes, etc..

Backing up using XMIT.

This has the advantage that the output file is a card image, which is a portable format.

I have a job

//MYLIBS1 JCLLIB ORDER=USER.Z24A.PROCLIB 
// SET TODAY='D201224' 
//S1 EXEC PROC=BACKUP,P=USER.Z24A.PARMLIB,DD=&TODAY. 
//S1 EXEC PROC=BACKUP,P=USER.Z24A.PROCLIB,DD=&TODAY. 

Where

• P is the name of the dataset
• TODAY  – is where I set today’s date.

The backup procedure has

//BACKUP PROC P='USER.Z24A.PROCLIB',DD='UNKNOWN' 
//S1 EXEC PGM=IKJEFT01,REGION=0M, 
// PARM='XMIT A.A DSN(''&P'') OUTDSN(''BACKUP.&P..&DD'')' 
//SYSPRINT DD SYSOUT=* 
//SYSTSPRT DD SYSOUT=* 
//SYSTSIN DD * 
// PEND 

The command that gets generated when P is USER.Z24A.PROCLIB and DD=D201224 is

XMIT A.A DSN('USER.Z24A.PROCLIB') OUTDSN('BACKUP.USER.Z24A.PROCLIB.D201224') 

This makes it easy to find the backups for a file, and a particular data.

To restore a file you use command TSO RECEIVE INDSN(‘BACKUP.USER.Z24A.PROCLIB.D201224’)  .

Using DFDSS to backup

This is a powerful program, and it is worth taking baby steps to understand it.

The basic job is

//IBMDFDSS JOB 1,MSGCLASS=H 
//S1 EXEC PGM=ADRDSSU,REGION=0M,PARM='TYPRUN=NORUN' 
//TARGET DD DSN=COLIN.BACKUP.DFDSS,DISP=(MOD,CATLG), 
// SPACE=(CYL,(50,50),RLSE) 
//SYSPRINT DD SYSOUT=* 
//SYSIN DD * 
 DUMP - 
   DATASET(INCLUDE(COLIN.JCL,COLIN.WLM,COLIN.C) - 
   BY(DSCHA,EQ,YES)) - 
   OUTDDNAME(TARGET) - 
   COMPRESS 
/* 

• For the syntax of the dump data set command see here.
• This dumps the specified data sets, COLIN.JCL. COLIN.WLM, COLIN.C, takes them and puts them in one file through TARGET.   TARGET is defined a dataset (COLIN.BACKUP.DFDSS).
• This does not actually do the backup because it has TYPRUN=NORUN.
• You can specify many filter criteria, in the BY(…) such as last reference, size, etc.  See here.
• The BY(DSCHA,EQ,YES) says dump datasets only if they have the “changed flag” set.  The Changed flag is set when a data set was open for output.  Using ADRDSSU with the RESET option resets the changed flag.   This allows you to backup only data sets which have changed – see below.
• It compresses the files as it backs up the files.

I did have

DATASET(INCLUDE(COLIN.**) - 
   EXCLUDE(COLIN.O.**,COLIN.SMP*.**,COLIN.DDIR ) - 
    BY(DSCHA,EQ,YES)) - 

Which says backup all data sets with the High Level Qualifier COLIN.**, but exclude the listed files.  I ran this using TYPRUN=NORUN, and this listed 100+ datasets.   Whoops, so I changed it to explicitly include the files I wanted to backup.  Once  I had determined the files I wanted to backup I removed the TYPRUN=NORUN, and backed up the datasets.

Using DFDSS to restore

You can restore from the DFDSS backups using a job like

//S1 EXEC PGM=ADRDSSU,REGION=0M,PARM='TYPRUN=NORUN' 
//TARGET DD DSN=COLIN.BACKUP.DFDSS,DISP=SHR
//SYSPRINT DD SYSOUT=* 
//SYSIN DD * 
  RESTORE - 
    DATASET(INCLUDE(COLIN.C) ) - 
    RENAME(COLINN) - 
    INDDNAME(TARGET) 
/* 

This says restore the files specified in the INCLUDE…  rename the HLQ to be COLINN.  From the dataset via //TARGET.

Initially I specified PARM=’TYPRUN=NORUN’ so it did not actually try to restore the files.   It reported

THE INPUT DUMP DATA SET BEING PROCESSED IS IN LOGICAL DATA SET FORMAT AND WAS CREATED BY Z/OS DFSMSDSS 
VERSION 2 RELEASE 4 MODIFICATION LEVEL 0 ON 2020.359 17:16:44 
DATA SET COLINN.C WAS SELECTED 
PROCESSING BYPASSED DUE TO NORUN OPTION 
THE FOLLOWING DATA SETS WERE SUCCESSFULLY PROCESSED 
COLIN.C 

From the time stamp 2020.359 17:16:44 we can see I was using the expected backup.

Once you are happy you have the right backup, and list of data sets, you can remove the PARM=’TYPRUN=NORUN’ to restore the data.

If you have backed up COLIN.JCL, and SUE.JCL, and try to rename on restore ( so you do not overwrite existing files) it would fail because if would create COLINN.JCL and then try to create COLINN.JCL from the other file!   To get round this using INCLUDE(COLIN.**) RENAMEN(COLINN) and INCLUDE(SUE.*) renamen(SUEN) .

 

What’s in the backup?

You can use the following to list the contents  (with TYPRUN=NORUN)

RESTORE - 
  DATASET(INCLUDE(**) ) - 
  INDDNAME(TARGET) 

Note: that because this job does not have REPLACE, it will not overwrite any files.

Using advanced backup facilities.

Each dataset has a changed-flag associated with it.   If this bit is on, the data set has been changed.  You can display this in the data set section of ISMF.  Field 24 – CHG IND, or if you have access to the DCOLLECT output, it is in one of the flags.

If you use

DUMP - 
   DATASET(INCLUDE(COLIN.JCL,COLIN.WLM,COLIN.C) - 
   BY(DSCHA,EQ,YES)) - 
   RESET - 
   OUTDDNAME(TARGET) - 
   COMPRESS 

it will backup the data sets, and reset the changed flag.  In my case it backed up the 3 data sets I had specified.

When I reran the same job, it backup up NO data sets, giving me a message

ADR383W (001)-DTDSC(01), DATA SET COLIN.JCL NOT SELECTED, 01.
Where  01 means  The fully qualified data set name did not pass the INCLUDE, EXCLUDE, and/or BY filtering criteria.

This is because I had specified BY(DSCHA,EQ,YES)) which says filter by Data Sets with the CHAnge flag on (DSCSHA) flag on.  The first DUMP request RESET the flag, the second DUMP job skipped the data sets.

You can exploit this by backing up all data sets once a week, and just changed data sets during the week.

You might need to keep the output of the dump job in member of a PDS, so you can search for your dataset name to find the date when a backup was done which included the file.

How many backups should I keep?

This depends on if you are backing up all, or just changed files.  You can use GDG (see here) where you use a generation of dataset.  If you specify 3 generations, then when you create the 4th copy, it deletes copy 1 etc.