Data locality, sovereignty, bandwidth, and egress patterns

Data locality is a design decision about where inputs are accessed and work runs. Data sovereignty is broader: it concerns legal, contractual, operational, and technical controls around data. Bacalhau can help express compute placement through node labels and job constraints, but it cannot certify compliance or infer the location of a remote source.

Choose the pattern that matches the boundary

Pattern	Use it when	What to verify
Site-local processing	A site hosts the data and compute	Node labels, local access path, output location, and offline behavior
Region-scoped processing	Data and workers must stay within an approved region	Storage endpoint, DNS and network routes, credentials, and failover region
Object-store-adjacent processing	Large objects are reachable from selected workers	Actual byte transfer, IAM scope, result publisher, and cost model
Centralized processing	A controlled central copy is acceptable	Ingest policy, egress cost, retention, and recovery objectives

Make placement explicit

Start a compute node with operator-maintained labels, then require them in the job:

bacalhau serve --compute -c Labels="residency=ca,site=toronto"

bacalhau docker run \
  --constraints "residency=ca" \
  --constraints "site=toronto" \
  --input https://example.org/public-records.csv:/input/records.csv \
  --output locality-results:/outputs \
  --publisher local \
  alpine:3.21 sh -c 'sha256sum /input/records.csv > /outputs/checksum.txt'

The command is intentionally simple: it proves that a constrained job can execute and publish an output. It does not prove that an HTTPS endpoint, container registry, log system, or operator workstation stayed in the same jurisdiction.

Controls that belong together

Goal	Bacalhau configuration	Complementary control
Direct work to suitable nodes	Labels and constraints	Change control for labels and capacity
Describe a data path	Input sources and tasks	Storage IAM, network policy, and credential rotation
Retain artifacts safely	Result paths and a publisher	Encryption, retention, and downstream authorization
Bound a workload lifecycle	Job type and timeouts	Monitoring, incident response, and business continuity

Limitations and audit evidence

For each production job, retain the submitted specification, selected node, input and output endpoints, identities used, timing, and transfer measurements. Review these against your jurisdictional and contractual requirements. Bacalhau's architecture explains the scheduling components, while your organization remains responsible for the surrounding enforcement and evidence.