Status: Accepted

Context

Analyzing video or converting presentations can be slow, resource-intensive, and dependent on native binaries.

Decision

Use BullMQ with Redis and run conversions in a separate Docker worker.

Rationale

• fast web responses after upload
• job retries and status tracking
• isolated FFmpeg, Poppler, and LibreOffice dependencies
• independently scalable workers
• visible processing states in the media library.

Consequences

Redis is required by the processing pipeline. Jobs must tolerate retries without producing inconsistent variants. Concurrency limits must account for memory and temporary disk use.