Golden Dataset Management

Operator baseline for maintaining persistent Golden DB. Canonical source: this runbook in angarabook/src/operations/.

Goal

Use a stable large dataset for:

• release closure validation;
• upgrade rehearsal;
• performance drift tracking;
• soak scenarios under realistic load.

Canonical sources

• RFC: RFC-2026-380-continuous-validation-infrastructure-v0
• Tooling: tools/golden_db/manage.sh

Infrastructure baseline

• Storage: .fastio/golden_db (NVMe).
• Separate data/txlog paths.
• Production-like durability and binary WAL.

Main commands

• tools/golden_db/manage.sh init
• tools/golden_db/manage.sh start
• tools/golden_db/manage.sh stop
• tools/golden_db/manage.sh status
• tools/golden_db/manage.sh grow --rows <n>
• tools/golden_db/manage.sh upgrade-check --binary <path>

Routine release flow

1. Stop Golden DB.
2. Run upgrade-check with the new binary on a snapshot.
3. Verify startup/connectivity/row-count oracle.
4. Record artifacts and the final verdict.

Validation tiers

• Tier 1: read compatibility (required).
• Tier 2: write compatibility (planned).
• Tier 3: performance canary (planned).

Next

• Testing and validation baseline — how the golden dataset is connected to the validation pipeline.
• CI reproducibility contract — fixture reproducibility requirements.