Cloud Arbitrage

DistributedCloudFinOpsMulti-Cloud

Roadmap Planned · not yet usable · autonomous cross-cloud cost optimization; verify before relying on it — SQL below is illustrative

What it is

Cloud Arbitrage is an autonomous cost-optimization layer that monitors cloud spot pricing in real time and migrates eligible workload components to the cheapest provider, region, or instance class that satisfies your declared policy. Because storage and compute are decoupled, the cluster can move execution to a cheaper region without moving any data — or it can move cold data to cheaper object storage when the egress math says it’s worth it.

The same machinery is meant to make spot-instance reclaims uneventful. When a provider issues a short reclaim warning, the departing node checkpoints its in-flight work to persistent storage, surviving nodes pick the work back up, and clients see no error.

Why it matters

Cloud bills are a top cost line for data platforms, and spot pricing fluctuates by the minute. Doing this optimization by hand means writing migration scripts and absorbing disruption whenever a spot reclaim hits an unprepared workload. Arbitrage is meant to turn that into a configuration choice.

Spot instances without the fear — reclaims are graceful yields, not outages.
Cross-cloud migration — workload follows the cheapest viable provider that meets your policy.
Egress-aware — cold-data moves factor network transfer fees into the destination choice.
Policy-driven — you set the rules (regions, compliance, latency); the agent stays inside them.

flowchart LR
  classDef a fill:#1f2640,stroke:#7c5cff,color:#e8eaf0
  classDef b fill:#1f2640,stroke:#00d4ff,color:#e8eaf0
  classDef g fill:#1f2640,stroke:#4ade80,color:#4ade80

  P[Provider price feeds<br/>AWS / GCP / Azure]:::a --> A[Arbitrage agent]:::a
  Pol[Operator policy<br/>regions, SLOs, budget]:::a --> A
  A --> M[Migrate replicas<br/>to cheapest viable]:::b
  A --> T[Tier cold data<br/>to cheapest storage]:::b
  M --> C[Cluster<br/>lower bill, same SLO]:::g
  T --> C

How you use it

The intended model is: declare a budget envelope and the providers / regions you’re willing to run in, plus any hard compliance constraints. The agent optimizes spend continuously within those rules.

-- Illustrative — tell the cluster which providers and regions are in play.
ALTER SYSTEM SET arbitrage_providers = ARRAY['aws', 'gcp'];
ALTER SYSTEM SET arbitrage_regions   = ARRAY['us-east-1', 'eu-central-1'];
ALTER SYSTEM SET arbitrage_mode      = 'spot-preferred';
ALTER SYSTEM SET arbitrage_budget_per_hour_usd = 45.00;

-- Pin compliance constraints the agent must never violate.
ALTER SYSTEM SET arbitrage_data_residency = 'eu-only-for:pii_tables';

-- Observe what the agent is doing.
SELECT ts, action, from_region, to_region, projected_savings_usd
FROM   _sys_arbitrage_log
ORDER BY ts DESC
LIMIT  20;

The agent is designed never to cross your residency, latency, or budget rules. Within those rules it minimizes spend continuously and logs every decision so you can audit it.

Key concepts

Term	Meaning
Spot instance	Discounted cloud hardware the provider can reclaim with short notice.
Graceful yield	A departing node checkpoints its work; surviving nodes pick up where it left off.
Egress-aware tiering	Cold-data migration that factors network transfer fees into the destination choice.
Residency policy	Hard constraints on where specific tables or rows may be stored or processed.
Arbitrage log	Auditable record of every migration the agent has performed.

Notes & caveats

Cloud Runtime →

The native tiering that makes cross-cloud movement zero-code.

Cluster Backup & Restore →

Cheap object storage as a backup target and DR anchor.

Transactions Overview →

How surviving nodes agree on the new owner during a yield.