Reference library
System Design Glossary
Use these definitions to make a design conversation more precise. Each term is intentionally concise; when a decision deserves a deeper explanation, follow the linked guide rather than treating the glossary as a substitute for practice.
A
- API gateway
- An edge service that applies shared API concerns such as authentication, routing, quotas, and request shaping before traffic reaches internal services.
- Read the related guide
- Availability
- The proportion of time a service can perform its intended function for a defined request class.
- Read the related guide
B
- Backpressure
- A mechanism that slows producers or rejects work when consumers cannot safely keep up.
- Read the related guide
C
- CAP theorem
- A distributed-systems lens: during a network partition, a replicated service cannot guarantee both every response and a single current value for every read.
- Read the related guide
- Cache stampede
- A burst of requests that all miss or refresh the same expensive value at once, overwhelming the origin.
- Read the related guide
- Circuit breaker
- A control that stops sending work to an unhealthy dependency for a period so failures do not amplify.
- Read the related guide
- Consistent hashing
- A partitioning approach that limits how many keys move when nodes join or leave a cluster.
- Read the related guide
D
- Dead-letter queue
- A separate destination for messages that repeatedly fail normal processing and need inspection or a repair workflow.
- Read the related guide
E
- Eventual consistency
- A model in which replicas may temporarily disagree but converge after propagation and conflict handling.
- Read the related guide
F
- Failover
- A controlled switch from an unhealthy component or location to a ready alternative.
- Read the related guide
- Fault tolerance
- The ability to continue a defined level of service despite component, dependency, or operator failures.
- Read the related guide
I
- Idempotency
- A property that lets a retried request produce the same intended outcome rather than duplicate work.
- Read the related guide
L
- Load balancer
- A traffic-routing layer that distributes requests based on policy and health information.
- Read the related guide
M
- Message queue
- A durable handoff between producers and consumers that supports asynchronous work and controlled acknowledgement.
- Read the related guide
- Multi-region
- An architecture that deliberately places service capacity or data across geographic failure domains.
- Read the related guide
P
- Partition
- A subset of data, requests, or nodes handled as a separately routed unit to distribute load or isolate failure.
- Read the related guide
Q
- Quorum
- A minimum overlapping set of replicas whose agreement is required before a distributed operation is accepted.
- Read the related guide
R
- Rate limiting
- A policy and enforcement mechanism that bounds how much work one identity can ask a system to perform.
- Read the related guide
- Replication
- Keeping copies of data on multiple nodes for read capacity, availability, recovery, or geographic proximity.
- Read the related guide
- RPO
- Recovery point objective: the maximum amount of acknowledged data loss an organization accepts after a recovery event.
- Read the related guide
- RTO
- Recovery time objective: the maximum time a service may take to restore its agreed function after disruption.
- Read the related guide
S
- Saga
- A distributed workflow made of local changes plus explicit compensating actions when a later step cannot complete.
- Read the related guide
- Sharding
- Splitting a dataset across independently scalable partitions selected by a routing key.
- Read the related guide
- SLA
- A customer-facing agreement that defines service expectations and often remedies when they are missed.
- Read the related guide
- SLI
- A measured indicator of a service outcome, such as successful requests within a latency threshold.
- Read the related guide
- SLO
- A target range for an SLI that guides engineering priorities and error-budget decisions.
- Read the related guide
T
- Throughput
- The amount of useful work completed per unit of time under stated conditions.
- Read the related guide
- TTL
- Time to live: a period after which stored data should expire or be refreshed.
- Read the related guide
Use the glossary as vocabulary support, then return to the broader Interview Preparation hub for behavioral and technical preparation paths.