The questions you will be asked

Mainframe interviews are more predictable than most. They test whether you understand mechanisms rather than whether you have memorised syntax, and they nearly always include a scenario.

JCL and z/OS

Question	What a strong answer contains
Explain DISP fully	Three positions, the four statuses, exclusive versus shared locking, and why (NEW,CATLG,DELETE) fails on rerun
Difference between COND and IF/THEN/ELSE	COND states when to bypass; IF is clearer and preferred for new work
What is in a STEPLIB and why does order matter	Program search order, first match wins, and the stale-module trap
What does TYPRUN=SCAN do	Syntax check without execution, and when you would use it
Job is queued but never starts	Job class, initiators, held class — not a program problem
Explain a PROC override	Replaces the whole DD, must match step order, JESJCL shows the result

COBOL

Question	What a strong answer contains
COMP vs COMP-3 vs DISPLAY	Storage layout, why COMP-3 is used for money, byte sizes
What causes S0C7 and how do you find it	Invalid numeric data; offset to line via the compile listing; hex view of the field; fix the source of the data
Static vs dynamic CALL	Bind time versus run time resolution, and what you must relink after a change
What does INITIALIZE do that MOVE ZEROES does not	Respects field types instead of moving a character across the whole group
Why use 88-levels	Readability and single point of change for a condition
How do you pass a PARM into COBOL	LINKAGE SECTION, PROCEDURE DIVISION USING, and the two-byte length prefix

DB2

Question	What a strong answer contains
What does BIND do and why	Access path chosen once instead of on every execution; packages, plans, collections
Explain -805	Package not found in the collections the plan searches; a deployment problem
-911 versus -913	Deadlock rolled back versus timeout not rolled back, and the retry implications
Why commit in batch	Lock duration, concurrency, restartability — and the checkpoint pattern
What is a cursor WITH HOLD	Survives a commit; without it the next FETCH fails
When would you use isolation UR	Reporting on stable data where an approximate answer is acceptable

CICS and VSAM

Question	What a strong answer contains
What is pseudo-conversational and why	Task ends between screens; commarea or channel carries state; it is about scalability
LINK versus XCTL	LINK returns, XCTL does not
What is ASRA	A program check in the application — the online S0C7
KSDS versus ESDS	Key access with an index versus entry-sequenced append-only
What are CI and CA splits	Inserts exhausting free space; performance degrades; reorganisation is the fix
Why does batch fail when CICS has the file open	Sharing rules; RLS is the answer where both are genuinely needed

The scenario question

Almost every interview includes one. It is not about knowing the answer; it is about showing method.

A typical scenario and a strong answer

Q: A nightly job that has run for six years abended with S0C7 at 02:30.
   The business needs the output by 06:00. What do you do?

A strong answer moves in this order:
  1. Establish impact — what does it feed, is it on the critical path,
     who is affected if it is late.
  2. Find the job on the spool; read JESMSGLG for the first failing step.
  3. Get the offset from the symptom dump, map it to the source line
     using the compile listing.
  4. Find the record — the program's DISPLAY output or the input file
     around the record count.
  5. Look at the field in hex; confirm it is not valid numeric data.
  6. Ask what changed — new feed, new data value, deployment, date boundary.
  7. Decide the action: fix data and restart, or bypass the record if the
     process allows, checking whether the failed step updated anything.
  8. Tell people, raise the underlying defect, update the runbook.

The method is the answer. Saying 'I would check the data' is not.

JCL and z/OS

COBOL

DB2

CICS and VSAM

The scenario question

Common mistakes

What you will see at work

Key terms