Engine Manual
Glossary
Terms used throughout the RushDB manual.
| Term | Definition |
|---|---|
| Base backup | A physical copy of the store files plus a manifest.json with sizes and CRC32 checksums, pinned at one commit index. Taken with rushdb serve --backup-base. The starting point for point-in-time restore and for attaching a follower. See Backup and restore. |
| Boot token | The secret given with --token / RUSHDB_TOKEN. Grants full access on every transport and is the pgwire password. Not bound to a tenant. |
| Capability token | A signed, expiring, permission-scoped (and optionally tenant-bound) token accepted on HTTP when --capability-secret is set. See Security. |
| Checkpoint | The point at which applied commits become durable in the store file, after which WAL segments can be reclaimed. Runs on every maintenance tick and at clean shutdown. |
| Commit index (CI), LSN | The sequence number of a durable commit. RushDB uses the same number as the WAL log sequence number, so --to-lsn takes a commit index. |
| Compaction | Rewriting the store file to release free pages and retained history. Runs automatically when the store is idle and has grown, or on demand with rushdb serve --compact. See Storage and compaction. |
config.yml | The configuration record inside a data directory. Written on first start and authoritative for that directory. Never contains secrets. |
| Cursor | A server-side, snapshot-pinned position in a records query, continued with query.cursor. Lives only in the server process and expires. |
| Data directory | The directory given with --data-dir, holding the store file(s), their WAL directory, and config.yml. |
| Declared index | An ordered, text, or vector index you create explicitly. RushDB does not create property indexes automatically. See Indexes. |
| Derived state | Counters, presence lists, and dictionaries the engine maintains from records (for example O(1) label counts). Recomputed from data, never replicated. |
| Durability mode | local (fsync before acknowledging, default), async (acknowledge before fsync), or memory (in-memory nodes). |
| Follower / leader | In streaming replication, the follower polls the leader's WAL and replays it; the follower is read-only until promoted. See Replication. |
| Group commit | Several concurrent commits written to the WAL with one fsync by the single publisher. |
| History floor (GC watermark) | The oldest commit whose snapshot can still be read. Older MVCC versions are reclaimed. |
| Label | The type name of a record (for example User). A record has one primary label. |
| LMPG | Labeled Meta-Property Graph, RushDB's data model and its storage layout: properties are first-class typed definitions, and records, property values, and relationships are stored as key families in an ordered key-value store. See LMPG layout. |
| Maintenance mode | A startup switch (--maintenance) that rejects writes and fails readiness while reads continue. |
| Maintenance tick | The background cycle (every 5 s by default) that runs GC, WAL retention, index builds, vector segment maintenance, and the durability checkpoint. |
| Managed envelope | The memory budget (managed-limit) from which caches, the transaction pool, and query phases reserve bytes. Not the process RSS. |
| MVCC | Multi-version concurrency control: every read sees one commit's snapshot while writers continue. |
| Point-in-time restore (PITR) | Restoring a base backup and replaying archived WAL up to a chosen commit index. |
| Protocol v1 | RushDB's transport-neutral request/response envelope, served at POST /protocol/v1/execute and over gRPC and Bolt. |
| Publisher | The single component that appends each group of commits to the WAL, fsyncs, and applies them to the store. All tenants share it. |
| Purge (tenant) | purgeTenant: delete every key of one tenant. The freed space is returned by the next compaction. |
| Read view | A pinned snapshot at one commit index used by a query, cursor, transaction, or export. |
| Record | A node in the graph: an id, a label, and properties. |
| Relationship | A typed, directed edge between two records. |
| Retention consumer | Something that still needs WAL history: a backup, the WAL archiver, a follower, or a cursor. Consumers protect the WAL from pruning up to the hard caps and expire after their lease. |
| Retry receipt | The stored outcome of a commit, kept for the idempotency window so a retried commit returns the original result. |
| Scope (tenant scope) | The string a request presents to select its tenant in multitenant mode: the X-RushDB-Tenant header, pgwire database, gRPC x-rushdb-tenant, or Bolt db. |
| Search query | RushDB's native JSON query document (query.records), always enabled. |
| Shedding | Refusing a request immediately with a retryable 503 when the scheduler or the in-flight body budget is full, instead of queueing without bound. |
| Statement language | An optional text query language: SQL, GQL, Cypher, Gremlin, SQL/PGQ, or MongoDB query documents. Enabled with --languages. |
| Tenant | An isolated keyspace inside one server. In multitenant mode every request selects one by scope; nothing identifying the tenant is stored on records. See Tenants. |
| WAL | Write-ahead log: the durable, ordered log of commits, stored beside the store file in rushdb.redb.wal/. |
| WAL archive | A directory of WAL segment files (wal-<from>-<to>.seg) written continuously when --wal-archive-dir is set, used for point-in-time restore. |