Joined: 06 Oct 2009 Posts: 45 Location: Melbourne, Australia
Hi,
I'm sure it's because I'm about to take a week off, but I've got a problem I don't seem to be able to work out. Essentially, I want to run a step at the end of a job, whether or not the rest of the job has an abend or high return code(s).
Normally, I've been able to do this using COND=ONLY, but for this particular job (failing with a wrong length record in a COBOL program), that doesn't seem to work. The info on the IEF450I message is ABEND=S000 U4038 REASON=00000001. Nor does an IF/THEN/ELSE/ENDIF construct using the same step included twice:
Code:
//CHKIF IF (ABEND | RC > 0) THEN
//CHKSTEPA EXEC PGM=IKJEFT01,DYNAMNBR=1635
...
// ELSE
//CHKSTEPB EXEC PGM=IKJEFT01,DYNAMNBR=1635
...
//CHKIF ENDIF
Both steps (in the above case) come up as "SMFSI2I CHKSTEP? STEP BYPASSED" without saying exactly why. I'm a bit at a loss, as I cannot see why it wouldn't work. I know I'm missing something, I just don't know what. And to make matters worse, there's a couple of steps in the original JCL that use the "IF (ABEND | RC > 0) THEN" to run a step to write to the console - and they execute successfully!?!
Oh, the reason for this is that I'm looking to allow our testers to run their inputs files in, using the scheduler authority, but without them having access to either the region files or the scheduler (OPC/Tivoli). So the step that I want to run, at the end of every job, will check the job condition codes and submit the next job in the list, A sort of a poor man's scheduler. It works for just about every execution type except this one job and its particular failure type.
I must need a couple of days off - maybe the rest of the year! I'll see if I can solve it by not thinking about ot for a while
System abends can’t be handled as per me in subsequent steps. However, you can try to change the cobol program to set specific RC when dataset attributes mismatch occurs and this way it will not have system abend and you can continue to run all steps including last one without IFTHEN.
Joined: 06 Oct 2009 Posts: 45 Location: Melbourne, Australia
Rohit Umarjikar wrote:
System abends can’t be handled as per me in subsequent steps. However, you can try to change the cobol program to set specific RC when dataset attributes mismatch occurs and this way it will not have system abend and you can continue to run all steps including last one without IFTHEN.
Unfortunately it's a production program and a generic IO routine at that. So changing it would be too much trouble.
But I had a thought this morning on a different approach (just in case there are abends of this type) and I might try working on that instead.
FYI, anyone interested in the Rexx used to grab the return codes from the job (in the last step), it's here:
Code:
/* */
GetHighStep:procedure expose x.;trace "O";cvt=storage(10,4)
ascb=storage(224,4);ascb_jbni=c2x(storage(d2x(c2d(ascb)+172),4))
ascb_jbns=c2x(storage(d2x(c2d(ascb)+176),4))
assb=storage(d2x(c2d(ascb)+336),4);asxb=Storage(D2x(c2d(ascb)+108),4)
acee=Storage(D2x(c2d(asxb)+200),4);jsab=storage(d2x(c2d(assb)+168),4)
asid=storage(d2x(c2d(ascb)+36),2);psatold=storage(21C,4)
jscb=storage(d2x(c2d(psatold)+180),4);jct=storage(d2x(c2d(jscb)+260),4)
sct=storage(d2x(c2d(jct)+48),3);pnt=0;sctd=c2d(sct)
do forever while sctd<>0;pnt=pnt+1
sctansct=storage(d2x(c2d(sct)+36),3)
sctsname.pnt=storage(d2x(c2d(sct)+68),8)
sctpgmnm.pnt=storage(d2x(c2d(sct)+124),8)
sctabcnd.pnt=x2b(c2x(storage(d2x(c2d(sct)+176),1)))
sctstend.pnt=x2b(c2x(storage(d2x(c2d(sct)+188),1)))
if \substr(sctstend.pnt,1,1)then sctsexec.pnt=0
else sctsexec.pnt=c2d(storage(d2x(c2d(sct)+24),2))
sct=sctansct;sctd=c2d(sct)
if substr(sctstend.pnt,1,1)&\substr(sctstend.pnt,2,1) then leave;end
if sct<>0 then do;OKRC=-1;x.0RC=0;abend=0;step=;message.=;message=0
do while pnt>1;pnt=pnt-1;if substr(sctabcnd.pnt,6,1) then do
abend=1;step=strip(sctsname.pnt);message=message+1
message.message="Step "step" abended.";end
else if sctsexec.pnt>OKRC then do;message=message+1
message.message="Step "strip(sctsname.pnt)" ended with RC="||,
sctsexec.pnt".";end
if sctsexec.pnt>x.0RC&\abend then do;x.0RC=sctsexec.pnt
step=strip(sctsname.pnt);end;end;message.0=message
if abend then return="ABEND";else return=x.0RC
sysid=storage(d2x(c2d(cvt)+340),4);x.0id=storage(d2x(c2d(jsab)+20),8)
x.0us=Storage(d2x(c2d(acee)+21),7)
if ascb_jbns=0 then x.0jb=storage(ascb_jbni,8)
else x.0jb=storage(ascb_jbns,8)
if x.0jb='INIT' then x.0jb=storage(d2x(c2d(jsab)+28),8)
if message>0 then do i0=1 to message;say message.i0;end;end;return sctd
/* */
I like the REXX solution for locating latest rc. You can also get it using the REXX/SDSF API, like so:
Code:
/* REXX */
n=isfcalls('ON')
isfprefix=mvsvar('symdef','jobname')
isfowner ='*'
Call XWait 2 /* (sometimes) required for synch */
Address SDSF "ISFEXEC ST"
do n=1 to jname.0 until queue.n='EXECUTION'
end
if queue.n<>'EXECUTION' then exit xmsg('Job not found')
/* show return codes for job till now */
Address SDSF "ISFACT ST TOKEN('"token.n"') PARM(NP SA)"
"execio * diskr" isfddname.3 "(stem msgs. finis)"
do n=1 to msgs.0
if word(msgs.n,1)='IEF142I' then say msgs.n
end
exit 0
XMsg: if arg(1)<>'' then say arg(1);return word(arg(2) 0,1)
XWait:
call syscalls 'ON'
address syscall 'sleep' arg(1)
call syscalls 'OFF'
return 0
Joined: 07 Feb 2009 Posts: 1317 Location: Vilnius, Lithuania
JPVRoff wrote:
FYI, anyone interested in the Rexx used to grab the return codes from the job (in the last step), it's here:
Got something similar in PL/I, but we had to change something in our system to get them run differently, as some control blocks were above the line or something like that, if I remember correctly, and that would have required the use of SWAREQ to convert pointers, or pseudo-pointers. Your code might suffer from the same problem.
put skip edit(ssib.ssibjbid,
jct.jctjname, 'run on',
jct.jmrjmrjd, 'by',
jct.jctuser)
(a, x(1),
a, x(1), a, x(1),
p'99v.999', x(1), a, x(1),
a);
call print_accounting();
sct_ptr = char3_to_ptr(jct.jctsdkad);
do while(sct_ptr ^= ptrvalue(0));
sctx_ptr = char3_to_ptr(sct.sctx_pch);
if unspec(substr(sctcdent(8), 1, 1)) & '04'bx then
do;
if ^have_had_title then
call print_title();
put skip edit(sctsclpc,
sctsname,
sctpgmnm)
(a, x(1),
a, x(1),
a, x(1));
put edit('ABENDed')
(a, x(1));
if flag_desc then
call print_desc();
if flag_parm then
call print_parm();
end;
else
if sctstend & '80'bx then /* started */
if sctstend & '40'bx then /* ended */
do;
if flag_before then
if (flag_zero & sctsexec = 0) |
(flag_nonzero & sctsexec ^= 0) then
do;
if ^have_had_title then
call print_title();
put skip edit(sctsclpc, sctsname, sctpgmnm)
(a, x(1), a, x(1), a, x(3));
put edit(sctsexec)
(p'zzzz9');
if flag_desc then
call print_desc();
if flag_parm then
call print_parm();
end;
if sctsexec > max_cc then
max_cc = sctsexec;
end;
else
do;
if flag_current then
do;
if ^have_had_title then
call print_title();
put skip edit(sctsclpc, sctsname, sctpgmnm)
(a, x(1), a, x(1), a, x(9));
if flag_desc then
call print_desc();
if flag_parm then
call print_parm();
end;
have_had_current = '1'b;
end;
else /* not started */
if have_had_current then
if flag_after then
do;
if ^have_had_title then
call print_title();
put skip edit(sctsclpc, sctsname, sctpgmnm)
(a, x(1), a, x(1), a, x(9));
if flag_desc then
call print_desc();
if flag_parm then
call print_parm();
end;
else
do;
end;
else
if flag_flush then
do;
if ^have_had_title then
call print_title();
put skip edit(sctsclpc, sctsname, sctpgmnm)
(a, x(1), a, x(1), a, x(1));
put edit('Flushed') (a, x(1));
if flag_desc then
call print_desc();
if flag_parm then
call print_parm();
end;
sct_ptr = char3_to_ptr(sct.sctansct);
end;
if have_had_title then
put skip edit((69)'-')(a);
put skip edit('Maximum completion code was', max_cc)
(a, x(2), p'zzzz9');
if jctjstat & '08'bx then
do;
put skip edit('The job ABENDed', abend_code(jctacode))
(a, x(1), a);
end;
print_parm: proc;
dcl ix fixed bin (31);
ix = index(sctxparm, '00'x);
if ix = 0 then
sctxparm_v = sctxparm;
else
sctxparm_v = substr(sctxparm, 1, ix);
put skip list('PARM=' || sctxparm_v);
end print_parm;
print_desc: proc;
dcl ix fixed bin (31);
do ix = 1 to desc_used while(sct.sctsname ^= desc_array(ix).stepname);
end;
if ix <= desc_used then
put edit(desc_array(ix).text) (x(1), a);
end print_desc;
load_desc_array: proc;
dcl sysin file input record sequential;
dcl (undf,eof) bit (1) init('0'b);
on undefinedfile(sysin) undf = '1'b;
on endfile(sysin) eof = '1'b;
dcl sysin_rec char (80);
dcl ix fixed bin (31);
open file(sysin);
if undf then
do;
put skip list('Unable to open SYSIN');
stop;
end;
read file(sysin) into(sysin_rec);
if eof then
do;
put skip list('No records in SYSIN');
stop;
end;
alloc desc_array;
do while(^eof);
ix = index(sysin_rec, ',');
if ix > 1 & ix <= 9 then
do;
desc_used = desc_used + 1;
desc_array(desc_used).stepname = substr(sysin_rec, 1, ix - 1);
desc_array(desc_used).text = substr(sysin_rec, ix + 1, 64);
end;
else
do;
put skip list('Invalid record in SYSIN');
stop;
end;
Tuens out that the REXX/SDSF API solution can be improved by using DA instead of ST:
Code:
/* REXX */
n=isfcalls('ON')
isfprefix=mvsvar('symdef','jobname')
isfowner ='*'
Call XWait 2 /* required for synch */
Address SDSF "ISFEXEC DA"
Address SDSF "ISFACT DA TOKEN('"token.1"') PARM(NP SA)"
"execio * diskr" isfddname.3 "(stem msgs. finis)"
do n=1 to msgs.0 /* show return codes for job till now */
if word(msgs.n,1)='IEF142I' then say msgs.n
end
exit 0
XMsg: if arg(1)<>'' then say arg(1);return word(arg(2) 0,1)
XWait:
call syscalls 'ON'
address syscall 'sleep' arg(1)
call syscalls 'OFF'
return 0
Joined: 06 Oct 2009 Posts: 45 Location: Melbourne, Australia
Willy Jensen wrote:
Tuens out that the REXX/SDSF API solution can be improved by using DA instead of ST
prino wrote:
Got something similar in PL/I, but we had to change something in our system to get them run differently, as some control blocks were above the line or something like that, if I remember correctly, and that would have required the use of SWAREQ to convert pointers, or pseudo-pointers. Your code might suffer from the same problem.
Thanks for both the ideas! I'm actually using SDSF in the last step to read in all the outputs to search for "run information" created during all of the job processes. But for the initial checking in the jobs, I've found that the Rexx is a fair bit quicker.
And it's also pretty robust, as I don't think we've had to change it for more than a decade (other than how it returns results to the calling Rexx).
P.S. I'm now using COND=EVEN and that's been working fine, plus I have a catch-all process in case the job fails with a JCL or system issue.
