Start Here: Core Patterns

Read/Write Files to Cloud Storage

Use /workspace/shared to read and write files through Google Cloud Storage.

Use custom Docker images & GPUs

Run workers with custom images, native tools, CUDA dependencies, and GPU resources.

Run jobs in the background

Detach long jobs so they keep running if your laptop sleeps or disconnects.

Limit parallelism for APIs or databases

Keep Burla jobs inside external service limits with chunking and max_parallelism.

Combine many results/files into one

Use a simple map-reduce pattern to merge parallel outputs into one final result.

Decide how to split your work

Choose the right input unit before scaling a job across many workers.

Process thousands of files

Run one function call per file, write one compact report per file, and reduce the reports.

Process one giant file

Stream one large JSONL file into chunks, process each chunk, and combine chunk reports.

Process database rows

Split indexed rows into ID ranges, process each range, and cap live database connections.