Key Concepts Glossary

A

ACID - Atomicity, Consistency, Isolation, Durability. Properties guaranteeing database transaction correctness.

Anti-entropy - Process of comparing replicas and synchronizing data to resolve inconsistencies.

Atomicity - Property that all operations in a transaction succeed or none do.

Availability - System's ability to respond to requests, even when some nodes fail.

B

B-tree - Balanced tree data structure for indexing, maintaining sorted key-value pairs. Standard in most relational databases.

Backpressure - Mechanism where downstream system signals upstream to slow down.

BASE - Basically Available, Soft state, Eventual consistency. Contrasts with ACID.

Bloom filter - Probabilistic data structure to test set membership, with false positives but no false negatives.

Broadcast hash join - Join strategy where small table is broadcast to all nodes.

Byzantine fault - Node providing incorrect or malicious data.

C

CAP theorem - Consistency, Availability, Partition tolerance - can only guarantee two in presence of network partitions.

Causal consistency - Ordering guarantee where causally related operations appear in order.

Change Data Capture (CDC) - Technique to capture and stream database changes.

Clustered index - Index where row data is stored directly in the index.

Column-oriented storage - Storage format where each column is stored separately, optimized for analytics.

Consensus - Process of getting multiple nodes to agree on a value.

Consistent hashing - Hash function that minimizes redistribution when nodes added/removed.

Consistent prefix reads - Guarantee that reads see causally consistent prefix of writes.

CRDT - Conflict-free Replicated Data Type. Data structure that automatically resolves conflicts.

D

Data-intensive - Application where challenges are data volume, complexity, and change rate (not CPU).

Distributed transaction - Transaction spanning multiple nodes, requiring coordination.

Durability - Property that committed data survives system failures.

E

Event sourcing - Storing all events that led to current state, not just current state.

Eventual consistency - System will become consistent if no new updates are made.

F

Fault - One component deviating from spec (vs failure: system as a whole stops working).

Fault tolerance - System's ability to continue operating despite faults.

Fencing token - Monotonically increasing token to reject stale operations.

G

Gossip protocol - Protocol for nodes to share information about cluster state.

H

Happens-before - Relationship defining causal ordering of events.

Hash index - Index using hash table for O(1) key lookups.

Head-of-line blocking - When slow requests hold up subsequent fast requests.

I

Immutability - Property that data cannot be changed after creation.

Isolation - Property that concurrent transactions don't interfere with each other.

J

Join - Combining rows from two or more tables based on related columns.

L

Latency - Time a request waits to be handled (excluding service time).

Linearizability - Strongest consistency model where operations appear atomic and instant.

Log-structured storage - Storage approach using append-only logs and compaction.

LSM-tree - Log-Structured Merge-tree. Storage engine using sorted runs and background compaction.

M

Materialized view - Pre-computed query result stored as a table.

Memtable - In-memory sorted data structure in LSM-trees.

Multi-Version Concurrency Control (MVCC) - Concurrency control using multiple versions of data.

N

Normalization - Database design to reduce data duplication through relationships.

O

OLAP - Online Analytical Processing. Read-heavy analytics workloads.

OLTP - Online Transaction Processing. Write-heavy transaction workloads.

Optimistic concurrency control - Allow conflicts, detect at commit time, retry.

P

Partition tolerance - System's ability to continue operating despite network partitions.

Percentile - Statistical measure (p50, p95, p99) describing distribution of values.

Pessimistic concurrency control - Lock before accessing to prevent conflicts.

Primary key - Unique identifier for a row/document/vertex.

Q

Quorum - Minimum number of nodes that must respond for operation to succeed.

R

Read skew - Reading different objects at different times, getting inconsistent view.

Replication - Copying data across multiple nodes for redundancy.

Replication lag - Delay between leader write and follower replication.

S

Schema-on-read - Data structure interpreted when read (vs schema-on-write enforced at write).

Secondary index - Index on columns other than primary key.

Snapshot isolation - Transaction sees consistent snapshot of data at start time.

SSTable - Sorted String Table. Immutable, sorted file format for LSM-trees.

Star schema - Schema with central fact table and dimension tables.

Synchronous replication - Leader waits for follower acknowledgment before confirming write.

T

Tail latency - High percentiles of response time (p95, p99, p99.9).

Three-phase commit (3PC) - Attempt to improve 2PC with reduced blocking.

Total order broadcast - Guarantee that all nodes receive messages in same order.

Transaction - Group of reads and writes as single logical operation.

Two-phase commit (2PC) - Coordinator-based atomic commit protocol.

V

Vector clock - Data structure for capturing causality in distributed systems.

Vertical scaling - Using more powerful machine (vs horizontal scaling).

W

Write-ahead log (WAL) - Append-only file for crash recovery and replication.

Write amplification - Multiple disk writes for single logical write.

Write skew - Two transactions read same data, write different objects, violating constraints.