Glossary
Every abbreviation used across this documentation, with what it means in skaidb specifically — several of these have a different sense elsewhere.
SQL keywords (SELECT, ON CONFLICT, NEAREST, RANK BY RRF, …) are not
abbreviations and are not listed here; they are in
QUERY_SYNTAX.md.
Storage engine
| Stands for | In skaidb | |
|---|---|---|
| LSM | Log-Structured Merge tree | The storage shape: writes land in an in-memory table, flush to immutable files, and compact in the background. |
| SSTable | Sorted String Table | One immutable on-disk file of sorted rows produced by a flush or compaction. |
| WAL | Write-Ahead Log | The durability journal a write enters before it is acknowledged. |
| TOAST | The Oversized-Attribute Storage Technique | Borrowed from Postgres. Value-TOAST stores fields above toast_threshold out-of-line, leaving a manifest in the row. |
| TTL | Time To Live | Per-table row expiry. It bounds visibility immediately; space returns when compaction or the TTL reclaim sweep rewrites the data. |
| GC | Garbage Collection | Reclaiming tombstones and superseded versions. Held back while a witness is behind, so deletes are not dropped before it has seen them. |
| OOO | Out-Of-Order | A time-series table's bounded late-arrival window (OOO 10m); samples older than it are rejected. |
| TS | Time-Series | A time-series table — series-keyed, Gorilla-compressed samples. See TIMESERIES.md. |
Distribution and consistency
| Stands for | In skaidb | |
|---|---|---|
| RF | Replication Factor | How many members hold a copy of a row. Set cluster-wide or per table. |
| CL | Consistency Level | Per-operation ONE / QUORUM / ALL. Reads and writes each have one; read CL + write CL > RF is the strong-consistency condition. |
| LWW | Last-Writer-Wins | The conflict rule: the write with the higher HLC stamp wins, deterministically on every replica. |
| HLC | Hybrid Logical Clock | The timestamp every write carries — wall-clock plus a logical counter, so causality survives clock skew. Not unique across nodes on its own; ties break by (hlc, node). |
| CAS | Compare-And-Set | A linearizable per-key consensus round. Backs clustered UPDATE, unique-value claims, and the job scheduler's ownership lease. |
| IWM | Intelligent Workload Management | The admission and pacing subsystem: sheds or slows work under memory and CPU pressure rather than falling over. |
| OOM | Out Of Memory | The kernel killing a process for exceeding available memory — what most of the pacing above exists to prevent. |
Query and indexing
| Stands for | In skaidb | |
|---|---|---|
| PK | Primary Key | The row's identity, possibly composite. A PK pinned by =/IN becomes a point read rather than a scan. |
| DDL | Data Definition Language | Schema statements (CREATE, ALTER, DROP). HLC-stamped and quorum-applied across the cluster. |
| DML | Data Manipulation Language | Row statements (INSERT, UPDATE, DELETE). |
| FTS | Full-Text Search | The search index built on embedded Tantivy. See SEARCH.md. |
| BM25 | Best Matching 25 | The relevance ranking function behind MATCH() / score(). |
| HNSW | Hierarchical Navigable Small World | The graph index behind vector search. See VECTOR.md. |
| ANN | Approximate Nearest Neighbour | What a vector index does — near-exact neighbours far faster than a full scan. |
| k-NN | k-Nearest Neighbours | The query shape NEAREST answers: the k closest rows. |
| RRF | Reciprocal Rank Fusion | RANK BY RRF — fuses a vector leg and a full-text leg into one hybrid ranking. |
| HLL | HyperLogLog | The sketch behind the opt-in APPROX_COUNT_DISTINCT(). Plain COUNT(DISTINCT) is exact and does not use it. |
| DIM | Dimension | The width of the vectors a vector index accepts; required at CREATE VECTOR INDEX. |
Security and authentication
| Stands for | In skaidb | |
|---|---|---|
| TLS | Transport Layer Security | Encryption for client and internode connections. |
| mTLS | mutual TLS | Both sides present certificates — how internode links authenticate each other. |
| CA | Certificate Authority | The issuer whose signature a node trusts. Internode peers and witnesses must share one. |
| SAN | Subject Alternative Name | The certificate field carrying the names a certificate is valid for. |
| X.509 | — | The certificate standard itself (also written x509 in config keys). |
| PEM | Privacy-Enhanced Mail | The base64 certificate/key file encoding, despite the mail-era name. |
| SCRAM | Salted Challenge Response Authentication Mechanism | The password authentication handshake; the password never crosses the wire. |
| RBAC | Role-Based Access Control | Privileges granted to roles, checked per statement. |
| ACL | Access Control List | Topic-level permissions in the MQTT broker. |
| GSSAPI | Generic Security Services API | The interface Kerberos authentication is offered through. See KERBEROS.md. |
| KDC | Key Distribution Center | The Kerberos server issuing tickets. |
| SPN | Service Principal Name | The Kerberos identity a skaidb server is known by. |
| SSO | Single Sign-On | Logging in once against an external identity provider. |
| CSP | Content Security Policy | The browser header restricting what the web UI may load. |
| XSS | Cross-Site Scripting | Injection of hostile script into a page — what output escaping in the UI prevents. |
| CSRF | Cross-Site Request Forgery | A forged authenticated request from another origin. |
Protocols and interfaces
| Stands for | In skaidb | |
|---|---|---|
| SQL | Structured Query Language | The primary interface. Dialect in QUERY_SYNTAX.md. |
| API | Application Programming Interface | Any programmatic surface — the drivers, REST, or the ES subset. |
| REST | Representational State Transfer | The HTTP query endpoint (POST /query). |
| HTTP | HyperText Transfer Protocol | Transport for REST, the UI, the ES subset and PromQL. |
| TCP | Transmission Control Protocol | The transport under the binary and internode ports. |
| QUIC | — | A name, not an acronym: the transport the binary client port is built on. |
| RPC | Remote Procedure Call | The internode request/response calls between members. |
| JSON | JavaScript Object Notation | The wire format for REST, the ES subset and stream events. |
| CSV | Comma-Separated Values | An export/import format. |
| TOML | Tom's Obvious Minimal Language | The server configuration file format. |
| DSN | Data Source Name | A single-string connection descriptor accepted by the drivers. |
| URL / URI | Uniform Resource Locator / Identifier | Endpoint addresses in config and driver connection strings. |
| DNS | Domain Name System | Hostname resolution — relevant to seeds and certificate names. |
| MQTT | Message Queuing Telemetry Transport | The built-in broker, also used to deliver change streams. See MQTT.md. |
| ES | Elasticsearch | The compatibility subset that answers _search for existing ES clients. |
| PromQL | Prometheus Query Language | Supported for querying time-series tables. |
| TEI | Text Embeddings Inference | A HuggingFace embedding/rerank server; one supported [inference] backend. |
| LLM | Large Language Model | The audience for LLM.md, the condensed machine-readable reference. |
| CLI | Command-Line Interface | skaidbsh, the shell and admin client. |
| UI | User Interface | The built-in web console. See UI.md. |
| PHP / .NET | — | Two of the eight supported driver languages, alongside Python, Node.js, Go, Java, Ruby and Rust. |
| RFC | Request For Comments | An IETF standards document — cited where skaidb implements one (e.g. RFC 4559 for SPNEGO, RFC 6455 for WebSocket). |
| IP | Internet Protocol | As in IP address — what binds and seeds are written as. |
| DB | Database | Shorthand for a database namespace within a cluster. |
Operations, platforms and units
| Stands for | In skaidb | |
|---|---|---|
| CPU | Central Processing Unit | Reported per node and used as a pacing input. |
| RAM | Random Access Memory | Physical memory; the budget the storage targets size themselves against. |
| RSS | Resident Set Size | A process's resident memory — the number to watch for leaks and OOM risk. |
| SSD / NVMe / HDD | Solid-State Drive / Non-Volatile Memory express / Hard Disk Drive | Storage classes. First-cycle and compaction timings differ sharply between them. |
| KB / MB / GB | Kilobyte / Megabyte / Gigabyte | Sizes in config and metrics. |
| VM | Virtual Machine | A virtualised host. |
| LXC | Linux Containers | The OS-level container runtime some deployments use. |
| UTC | Coordinated Universal Time | The timezone every stored timestamp is in. Jobs accept fixed offsets, not named zones. |
| ISO | International Organization for Standardization | As in ISO 8601, the timestamp string format accepted in SQL. |
| UUID | Universally Unique Identifier | A random unique id. |
| ULID | Universally Unique Lexicographically Sortable Identifier | A unique id that sorts by creation time — used where ordering matters. |
| ARM64 / x86-64 | — | The CPU architectures release binaries are built for. |
| RHEL | Red Hat Enterprise Linux | One of the packaged Linux targets. |
| SHA256SUMS | — | The checksum file published with each release. |
| ACID | Atomicity, Consistency, Isolation, Durability | The classical transaction guarantees. Single-node transactions and BEGIN ATOMIC PARTITION are the scopes offering them. |
| CRUD | Create, Read, Update, Delete | The four basic row operations. |
| OLTP | Online Transaction Processing | Many small reads and writes — the workload the row path is tuned for. |