Oracle Integration Cloud: Building Integrations That Survive Production · Module 7 · Scheduled integrations and file handling
Archiving and idempotent file processing
Lesson 76 of 175 · 2 min
Making "process each file exactly once" true EVEN WHEN A RUN DIES HALFWAY. The move-first pattern from Lesson 4 does most of it: claim by moving out of inbound, finish by moving to processed, and a dead run leaves the file visibly in between. That gives you exactly-once for the FILE. It does not give you exactly-once for the RECORDS inside a file the run had partly committed, which is the harder half. Either the target is idempotent — a reference per record, checked before creating, which is the pattern this course states everywhere — or the flow records its position durably so a rerun resumes rather than restarts. MARKER FILES solve the partner side. The partner writes the data file, then writes a small marker; your pattern matches the marker and reads the data. The marker cannot exist before the data is complete, which is what makes the half-written…
The full lesson is part of the course
The video, the complete written lesson and the module quiz are included in Oracle Integration Cloud: Building Integrations That Survive Production, with a certificate on completion and a fourteen-day refund window.
In this module: Module 7 · Scheduled integrations and file handling
- 1Scheduling an integration
- 2Parameters and run state
- 3Reading files: patterns and directories
- 4The staging problem
- 5Large files and chunking
- 6Writing files a partner can read
- 7Archiving and idempotent file processing
- 8Catching the file that never arrived
- 9What breaks: four file failures
- 10Lab briefing · A nightly file, done properly