Skip to content

Glossary of Terms

A

ACID
Atomicity, Consistency, Isolation, Durability: the four guarantees of a reliable transaction — all or nothing, rules hold, no interference, and committed data survives crashes.

B

BASE
Basically Available, Soft state, Eventually consistent — the looser guarantees many NoSQL systems accept in exchange for availability and scale.
Bloom filter
A compact probabilistic structure that answers 'definitely not present' or 'maybe present'; LSM-tree engines use it to skip files that cannot hold a key.
B+tree
The balanced tree structure behind most database indexes; all values sit in linked leaf pages, so lookups and range scans need only a few page reads.
Buffer pool
The database's own memory cache of disk pages; reads and writes go through it, and its hit ratio strongly affects performance.

C

CAP theorem
During a network partition a distributed system must choose between consistency and availability; PACELC adds the latency-versus-consistency trade-off when there is no partition.
Change data capture (CDC)
Streaming every insert, update and delete from a database's log to other systems, for example with Debezium into Kafka.
Columnar storage
Storing each column's values together instead of whole rows, so analytical queries read only the columns they need and compress very well.
Common table expression (CTE)
A named temporary result defined with WITH at the start of a query; recursive CTEs can walk hierarchies and graphs.
Connection pool
A set of open database connections shared by many application requests, avoiding the cost of reconnecting; examples are HikariCP and PgBouncer.
Consensus
Agreement among replicated servers on a single order of decisions despite failures; Raft and Paxos are the best-known algorithms.
Constraint
A rule the database enforces on data — primary key, foreign key, UNIQUE, CHECK, NOT NULL — rejecting changes that would break it.

D

Data lake
Cheap object storage holding raw files of any format (often Parquet) for later processing and analytics.
Data warehouse
A database optimised for analytical queries over large, integrated, historical data, such as Snowflake, BigQuery or Redshift.
DBMS
Database management system: the software that stores, queries, secures and recovers data, such as PostgreSQL, Oracle or MongoDB.
Deadlock
Two or more transactions each waiting for a lock the other holds; the database detects it and aborts one of them.
Denormalisation
Deliberately duplicating data to make reads faster or simpler, at the cost of extra storage and harder updates.
Document database
A NoSQL database storing self-contained JSON-like documents, such as MongoDB or Firestore.

E

Embedding
A vector of numbers produced by a machine-learning model so that similar meanings lie close together; the basis of vector search.
Entity-relationship (ER) diagram
A drawing of entities, their attributes and the relationships between them, used to design a database before building it.
ETL / ELT
Extract, transform, load — or load first and transform inside the warehouse (ELT) — the two main patterns for moving data into analytics systems.
Eventual consistency
A guarantee that replicas converge to the same value if no new writes arrive, though reads may briefly return stale data.
EXPLAIN
A command that shows the execution plan the optimiser chose for a query; EXPLAIN ANALYZE also runs it and reports real timings.

F

Foreign key
A column whose values must match a key in another table, enforcing referential integrity between related rows.

G

Graph database
A database storing nodes and relationships as first-class data, queried with languages like Cypher or GQL; Neo4j is the best known.

H

HNSW
Hierarchical Navigable Small World: a layered graph index for fast approximate nearest-neighbour vector search.

I

Index
An extra data structure that lets the database find rows without scanning the whole table, at the cost of storage and slower writes.
Isolation level
How strongly concurrent transactions are separated: read uncommitted, read committed, repeatable read or serializable.

J

Join
Combining rows from two tables based on a related column; inner, left, right, full outer, cross and self joins are the main kinds.

K

Key-value store
A database that stores values under unique keys with very fast lookups, such as Redis, Valkey or DynamoDB.

L

Lakehouse
An architecture combining data-lake storage with warehouse features such as ACID tables and SQL, using open table formats like Iceberg or Delta Lake.
Least privilege
Giving every user and application only the permissions it needs, limiting the damage from mistakes or stolen credentials.
LSM-tree
Log-structured merge-tree: a write-optimised storage structure that buffers writes in memory and merges sorted files in the background.

M

Materialized view
A stored result of a query that can be refreshed, trading freshness for much faster reads.
Model Context Protocol (MCP)
An open protocol that lets AI applications call external tools and data sources, including databases, through standard servers.
MVCC
Multi-version concurrency control: keeping several versions of rows so readers see a consistent snapshot without blocking writers.

N

Normalisation
Organising tables to remove redundancy and update anomalies, following normal forms such as 1NF, 2NF, 3NF and BCNF.
NoSQL
A broad family of non-relational databases — key-value, document, wide-column and graph — built for particular data shapes and scale.
NULL
A marker for a missing or unknown value; comparisons with NULL yield 'unknown', which is why SQL uses three-valued logic.

O

OLAP
Online analytical processing: complex queries scanning large volumes of historical data for reporting and analysis.
OLTP
Online transaction processing: many small, fast reads and writes from applications, such as orders and payments.
Open table format
A specification such as Apache Iceberg, Delta Lake or Hudi that adds tables, transactions and schema evolution on top of files in object storage.
ORM
Object-relational mapper: a library that maps programming-language objects to database rows and generates SQL, such as Hibernate or SQLAlchemy.

P

Parameterised query
A query sent with placeholders and separate values, so user input can never change the SQL — the main defence against SQL injection.
Parquet
An open columnar file format with compression and statistics, the standard storage format for data lakes and lakehouses.
Partitioning
Splitting a large table into smaller pieces by range, list or hash, within one server or across servers.
Point-in-time recovery (PITR)
Restoring a database to any chosen moment by replaying archived transaction logs on top of a base backup.
Primary key
The column or columns that uniquely identify each row in a table; it cannot be NULL or duplicated.

Q

Query optimiser
The DBMS component that estimates the cost of alternative plans using statistics and picks the cheapest way to run a query.
Quorum
The minimum number of replicas that must acknowledge a read or write; with W + R > N, reads overlap the latest write.

R

Relational model
E. F. Codd's 1970 model representing data as relations (tables) of tuples (rows) with attributes (columns), manipulated declaratively.
Replication
Keeping copies of the same data on several servers for availability, durability and read scaling; it can be synchronous or asynchronous.
Retrieval-augmented generation (RAG)
Retrieving relevant data, often by vector search, and passing it to a language model so its answer uses current, private facts.
Row-level security (RLS)
Database policies that automatically filter which rows each user may see or change.
RPO / RTO
Recovery point objective (how much data you can afford to lose) and recovery time objective (how long you can afford to be down).

S

Schema
The defined structure of a database — tables, columns, types, constraints and relationships — or a named namespace grouping such objects.
Sharding
Splitting data horizontally across several servers, each holding a subset of rows chosen by a shard key.
Snapshot isolation
An isolation level in which each transaction reads a consistent snapshot as of its start; it prevents most anomalies but allows write skew.
SQL
Structured Query Language: the standard declarative language for defining, querying and changing relational data, standardised by ISO/IEC 9075.
SQL injection
An attack in which untrusted input concatenated into SQL text changes the query's meaning; prevented by parameterised queries.
Star schema
A dimensional model with a central fact table of measurements surrounded by dimension tables describing who, what, where and when.
Stored procedure
Named code stored and run inside the database, written in languages such as PL/pgSQL, T-SQL or PL/SQL.
Surrogate key
An artificial identifier, such as an identity integer or UUID, used as a primary key instead of a natural business value.

T

Transaction
A group of database operations treated as one unit that either commits completely or rolls back completely.
Transparent data encryption (TDE)
Engine-level encryption of data and log files at rest, invisible to applications.
Two-phase commit (2PC)
A protocol in which a coordinator first asks all participants to prepare, then tells them all to commit, keeping a distributed transaction atomic.

V

Vector database
A database specialised in storing embeddings and finding nearest neighbours quickly; many general databases now offer vector search too.

W

Window function
A function such as ROW_NUMBER, RANK or SUM OVER that computes across related rows without collapsing them into groups.
Write-ahead log (WAL)
A log to which every change is written before the data pages, so committed transactions can be recovered after a crash.