Error codes and guides

Decimal arithmetic or a numeric compare hit bytes that are not a valid number. The program is usually fine; the data is not.

The program touched storage it is not allowed to use — often a subscript out of range, a bad parameter list or a closed file.

z/OS could not find the load module to run or load in any library the step searched.

An output dataset could not get more space. B37, D37 and E37 each fail for a different reason, so read the code before you change SPACE.

A dataset could not be opened, usually because its attributes conflict with the program or JCL, or a member does not exist.

Language Environment ended the program after an unhandled run-time error. The real cause is in the CEE and IGZ messages, not the code.

Db2 could not find a package matching the program in any collection the plan searches. Nearly always a bind or deployment problem.

Two units of work wanted the same locks. -911 means Db2 rolled your work back; -913 means it did not, and you must decide.

RACF refused an access. Read every field of the message and it tells you whose access, to what, and what is missing.

A program check inside a CICS transaction — the online equivalent of a batch S0C7 or S0C4. The interrupt code and offset lead you to the bad line.

The system rejected the JCL or its allocations before the program ran. Different from a job that ran and failed.

The processor was asked to run bytes that are not a valid instruction — usually because control went to the wrong place.

The program tried to run an instruction only the operating system may use — in application code, a sign that control went to the wrong place.

A packed-decimal divide failed — almost always because the divisor was zero; rarely the quotient was too large for the instruction's work area.

The job or step used more CPU time than it was allowed. Find out whether it was looping or just had more work before raising TIME.

An INSERT or UPDATE would create two rows with the same value in a unique index. Usually a reprocessed file or a key-generation problem.

A singleton SELECT INTO or a scalar subquery found more than one row. The data or the predicate is not as unique as the program assumed.

Db2 could not use an object or resource, often a table space in a restricted state. The reason code and resource name tell you which and why.

A date, time or timestamp was in the wrong format (-180) or out of range (-181). Usually spaces, low-values or a badly built date string.

An INSERT or UPDATE set a foreign key to a value with no matching parent row. Usually a load-order or missing-reference-data problem.

An MQI call was refused by security. Find which call failed, under which user ID, against which object.

The queue manager could not find the queue or object the program named. Usually a name, case or environment mismatch.

A put failed because the queue reached its maximum depth. The real problem is usually the consumer, not the producer.

Submitted, but nothing happens. A decision tree.

The code is fixed, the job ran, and the behaviour is unchanged.

Same job, same JCL, much longer elapsed time.

A batch job cannot open a VSAM file because a CICS region has it open for update.

Text that looks fine on one platform is gibberish on the other.

Return code zero, empty or missing output. Often worse than an abend.

A system-wide condition that can stop new work across the whole system.

Ended not OK, stuck waiting or running late: how to read the scheduler, find the cause, and choose rerun, force complete or hold safely.

A workflow for REST APIs served through z/OS Connect: read the status code, find the layer that failed, then fix the right thing.

A realistic order that gets you useful fastest.

An honest answer with the caveats attached.

The access problem and the ways around it.

Because on this platform, consumption maps directly to a bill.

What team composition is often like, and how to get the most from it.

Three quick tells in the source.

Knowing who to ask saves more time than knowing the answer.

A short answer with a genuinely useful consequence.

A rough dictionary for anyone arriving from another platform.

Finding, reading and managing jobs and system output.

Line commands, primary commands, and the combinations that matter.

Commands you can run from ISPF option 6 or the READY prompt.

Defining, copying, listing and deleting — VSAM and catalog operations.

Inspecting and controlling a CICS region.

Catalog queries and operator commands worth knowing.

Standard Unix, plus the z/OS-specific ones that matter.

Everyday Zowe CLI commands for working with datasets and jobs from a terminal or a pipeline.

The simplest useful job: IEBGENER with a new output dataset.

DFSORT control cards doing what would otherwise be a program.

How a job that runs every night produces a new output without changing its JCL.

IF/THEN/ELSE instead of COND, with a notification path on failure.

The three-step pattern, with the options that matter in test.

The shape of most batch COBOL: open, loop, accumulate, close, report.

OCCURS, INDEXED BY, SEARCH and SEARCH ALL, with the bounds problem handled.

The batch pattern: WITH HOLD, periodic commit, and a restart record.

The template most CICS screens follow.

An example convention, and how to read one you have been given.

The four allocations you will need most, with the parameters explained.

Creating the group, and what the options actually do.

What to get sorted before you can be productive.

Ninety seconds that prevent most avoidable failures.

What to look for when reviewing someone's change, or your own.

What to do, in order, when something fails at 3am.

What must be in place before a change goes to production.

What the next person needs to know, written so they do not have to ask.

What to have in place before moving a COBOL application from manual promotion to a CI/CD pipeline.