COBOL to BigQuery

Start with a representative workload

Use this guide to prepare a COBOL to BigQuery assessment. The first decision is whether the intended target preserves the behavior your business relies on, at an acceptable operating cost.

What to collect from COBOL

• Include copybooks, called programs, file layouts, and JCL or other batch dependencies.
• Sample packed decimals, fixed-width records, date fields, and file-status/error handling.
• Keep record ordering, numeric precision, and restart expectations explicit; select representative batch jobs first.

Plan the BigQuery implementation

Agree which transformations run in BigQuery SQL and which need a separate Python or Spark runtime. Check query cost, partition pruning, permissions, and load/retry behavior.

A small transformation example

Use this excerpt to discuss the mapping, not as a complete executable job. Verify source-specific null, type, and ordering behavior with the surrounding workflow.

Before
05 AMT-X PIC X(12).
05 AMT-N REDEFINES AMT-X PIC S9(10)V99.

After (BigQuery)
-- Typed fields → BigQuery SQL (conceptual)

Agree what a passing result means

1. Fix the baseline. Use the same input snapshot and record source parameters and expected outputs.
2. Compare the data. Check row counts, keys, duplicates, nulls, aggregates, and row-level values with agreed precision tolerances.
3. Review exceptions. Keep a list of behavior that needs manual work, an owner, and a repeatable test.
4. Check operations. Measure runtime and cost, rehearse retries and recovery, and confirm who owns the production job.

What to decide after the pilot

Review generated code, test results, unresolved exceptions, and measured delivery effort together. Expand only after the sample meets your acceptance criteria. Agree whether your team leads the next wave, needs engineering support, or wants managed delivery.