Overview

HydraDB's authority map separates canonical graph truth, query snapshots, index generations, WAL overlays, coordination records, and local caches by what each may decide. It explains safe degradation: use an indexed base only when its tail can be reconciled to the pinned view; otherwise fall back to canonical adjacency. This supports replaceable compute and bounded operation without proving universal latency, availability, security, or compatibility.

Learning outcomes

Durable SlateDB records define committed graph content; snapshots define query visibility, while generations and overlays only accelerate or reconcile reads.
Caches, reader handles, and compiled matrices can be rebuilt, but their loss causes warm-up, fallback, or bounded failure rather than graph change.
A missing or oversized index tail must trigger canonical fallback or explicit failure, never an unmarked stale result.

Keywords

HydraDBAuthorityRebuildabilitySnapshotsOverlaysCaches

The Authority Map: What Can Be Rebuilt, and What Cannot

HydraDB's first four design questions now meet in one judgment: which pieces of a graph system are allowed to lag, disappear, or be duplicated, and which piece must decide what the graph means? This synthesis connects scopes and canonical records with pinned snapshots, traversal generations, WAL overlays, sparse kernels, and local caches. The practical result is a way to predict HydraDB's behavior under change without mistaking a fast route for an authoritative one.

The distinction is easy to state and easy to lose. Durable graph records define committed vertices, relationships, properties, topology, and deletion state. A query snapshot defines the version a particular read may see. An immutable generation accelerates topology. A WAL overlay brings that generation forward to the query's read point. Coordination records help compute discover scopes or choose a contender. Local caches avoid repeated work. All may be stored near one another, but they do not answer the same question.

Six roles, one boundary

The most important synthesis is not a list of components; it is a map of permissions. Each role can answer some questions and is forbidden to answer others.

classDiagram class DurableGraphTruth class QuerySnapshot class IndexGeneration class WALOverlay class CoordinationRecords class LocalCache DurableGraphTruth --> QuerySnapshot : authoritative records at a sequence IndexGeneration --> QuerySnapshot : accelerates topology within the view WALOverlay --> QuerySnapshot : reconciles the base to the read point CoordinationRecords --> IndexGeneration : guides discovery and publication LocalCache --> IndexGeneration : retains rebuildable working state QuerySnapshot --> DurableGraphTruth : constrains visibility

Durable graph truth is the authority. In HydraDB, that truth is held in SlateDB's durable records on S3-compatible object storage. The record may be an endpoint, a relationship, a tombstone, a property, or a topology entry. Its authority comes from the storage and transaction model, not from the fact that a particular process currently has it in memory.

A query snapshot is a bounded view over that truth. It does not create a second copy of the graph's meaning. It says which committed sequence the query will interpret. If the query reads a relationship, its endpoints, its properties, and its deletion state, those reads are constrained by the same snapshot. The snapshot is therefore authoritative for the query's visibility, but not a new durable record that other queries must adopt.

An index generation is a materialized topology view. It can be immutable and durable in object storage while still remaining subordinate to canonical records. Its job is to make repeated traversal cheaper or more compact. It may be built at an earlier sequence than the query needs, in which case a visible WAL overlay can reconcile it. If that reconciliation is unavailable or too expensive, the generation loses its place in the execution path; the canonical snapshot does not lose its authority.

The overlay is a bridge between two times, not a third source of truth.

Published by DeepMox

FAQ

What is HydraDB's authority map?

HydraDB's authority map assigns durable graph records the right to define content, while snapshots define visibility and indexes, overlays, coordination records, and caches serve narrower roles.

Can HydraDB rebuild a lost query node's state?

Yes, HydraDB can rebuild local reader state, hydrated adjacency, compiled matrices, and caches from durable records, though replacement incurs reopening, remote reads, and warm-up.

Why is a durable HydraDB index not the source of graph truth?

A durable HydraDB index is derived from canonical records and may lag their sequence, so it accelerates traversal but cannot define visibility or relationship meaning.

Previous11 minHydraDB Sparse Traversal and Paths

Next11 minHydraDB Writer Placement and Fencing

HydraDB Authority and Rebuildability

The Authority Map: What Can Be Rebuilt, and What Cannot

Six roles, one boundary

FAQ