AWS · GCP · Azure

Serverless Geospatial Processing, done right.

Authoritative engineering references for designing, deploying, and operating serverless raster and vector workflows. Built for cloud GIS engineers, Python backend devs, and platform architects who care about constraint-aware architecture, cost, and observability.

Event-Driven Patterns Object-storage triggers, queue routing, batch-vs-stream, chunked I/O, ... Architecture & Platform Limits Memory and CPU sizing, ephemeral storage, GDAL cold-starts, concurrenc... Packaging & Dependencies Python layers, native compilation, Docker optimization, PROJ and GDAL ... Pipeline Recipes End-to-end, runnable serverless recipes: 10 GB GeoTIFF tiling with Ste...

What this site is

Serverless platforms reshape how spatial data is ingested, transformed, and served. Functions replace fleets, queues replace cron, and chunked I/O replaces monolithic loads — but only if the architecture respects platform limits, package size budgets, and IAM boundaries.

Each reference here distils production patterns: event-driven shapefile ingestion, raster cold-start mitigation, dead-letter routing for vector jobs, packaging GDAL into Lambda layers, and least-privilege IAM across AWS, GCP, and Azure. Side-by-side platform comparisons — cold starts, timeout ceilings, ephemeral storage — help you pick the right provider per constraint.

The focus throughout is constraint-aware: memory, ephemeral disk, timeout, concurrency, and cost. End-to-end pipeline recipes tie the concepts together into runnable builds — tiling a 10 GB GeoTIFF with Step Functions, real-time AIS tracking, Sentinel-2 NDVI, Cloud Run vector tiles, and publishing the result as a STAC catalogue. Examples lean on Python, GDAL, and the major-cloud event primitives, and emphasise observability and deployment automation over toy demos.

Explore the references

Event-Driven Patterns

Object-storage triggers, queue routing, batch-vs-stream, chunked I/O, idempotency, and durable orchestration for spatial pipelines.

Read the guide

Architecture & Platform Limits

Memory and CPU sizing, ephemeral storage, GDAL cold-starts, concurrency ceilings, cost modelling, and least-privilege IAM across AWS, GCP, and Azure.

Read the guide

Packaging & Dependencies

Python layers, native compilation, Docker optimization, PROJ and GDAL runtime configuration, and CI/CD synchronization for geospatial dependencies.

Read the guide

Pipeline Recipes

End-to-end, runnable serverless recipes: 10 GB GeoTIFF tiling with Step Functions, real-time AIS tracking, Sentinel-2 NDVI, Cloud Run vector tiles, and STAC cataloguing.

Read the guide

New in this release

  • Concurrency and Throttling for Tile Fan-Out Architecture & Limits

    What happens when one scene fans out into thousands of concurrent invocations, which quota gives way first on each cloud, and how to put back-pressure between the orchestrator and the workers.

  • Cost Modelling for Serverless Raster Pipelines Architecture & Limits

    The cost-per-tile equation end to end, why a larger memory setting is often cheaper, and the charges that quietly dominate a real raster pipeline.

  • Idempotency and Exactly-Once Spatial Processing Event-Driven Patterns

    Every cloud event source is at-least-once. The three places to enforce idempotency in a tiling pipeline, and why check-then-write is not one of them.

  • PROJ and GDAL Runtime Configuration Packaging & Dependencies

    The environment a GDAL stack actually needs at runtime — GDAL_DATA, PROJ_LIB, datum grids — and the configuration options that measurably change COG read performance.

  • STAC Cataloging and Metadata Publishing Pipeline Recipes

    The stage most pipelines skip: turning finished tiles into a catalogue someone can actually find, as static JSON on object storage or behind a STAC API.

Start here — featured pages