Data Storage Architecture
Acceso’s storage layer is built for read-heavy workloads. It also supports selective persistence and strict domain isolation.
Acceso does not store the full blockchain. It persists derived, indexed, and operational data needed for performance and reliability.
Storage goals (what the layer optimizes for)
Low-latency reads for API requests.
Strong consistency for identity, metering, and config.
Isolation between tenants and data domains.
Separate paths for transactional and analytical workloads.
Predictable lifecycle management and retention enforcement.
What we persist (and what we don’t)
Persisted:
Operational state: API keys, account metadata, configs, usage counters.
Derived indices: normalized views that speed up reads.
Aggregates and signals: metrics rollups and analytics outputs.
Not persisted:
A complete replica of chain state.
Raw upstream data that is not required for product behavior.
Storage planes
The system separates “serve requests” data from “analyze trends” data. This prevents analytical queries from starving transactional workloads.
Operational data backs authentication, authorization, and metering. It is optimized for small reads and writes with strict correctness.
Typical contents:
API keys and key status.
Account and plan metadata.
Quotas, counters, and enforcement state.
Configuration and feature flags.
Guarantees and expectations:
Strong consistency for write-followed-by-read workflows.
Low-latency reads and writes.
Strict access controls and auditable mutations.
Analytical data is separated from operational state. It is optimized for scanning, aggregation, and historical queries.
Typical contents:
Historical usage metrics.
Aggregated statistics and rollups.
Derived signals for analytics layers.
Guarantees and expectations:
Read-heavy access patterns.
Query isolation from OLTP workloads.
Retention can vary by data category and plan tier.
Some data is intentionally short-lived. It exists to accelerate request paths or computations.
Typical contents:
Temporary computation artifacts.
Intermediate normalization outputs.
Short-lived caches and precomputations.
Guarantees and expectations:
Explicit TTLs and automatic eviction.
No reliance on long-term persistence.
Safe to regenerate when missing.
Isolation and access boundaries
Isolation is enforced by design, not convention.
Identifiers are scoped to accounts and API keys.
Data domains are separated to prevent cross-contamination.
Access paths are constrained by backend authorization rules.
Last updated