Glossary
Fast definitions for the vocabulary behind distributed systems design discussions.
39 terms
A
- Anti-entropyReplication
- Background comparison and repair that brings diverged replicas back toward the same state.
- AvailabilityCAP
- The system’s ability to return a non-error response to requests that reach a healthy component.
B
- BackpressureReliability
- A signal or policy that slows producers when consumers cannot safely keep up.
- Bloom filterLSM trees
- A compact probabilistic structure that can prove an item is absent, with possible false positives but no false negatives.
C
- Causal consistencyConsistency
- A model that guarantees causes are observed before their effects while concurrent events may appear in different orders.
- Circuit breakerReliability
- A control that temporarily stops calls to a failing dependency so it and its callers can recover.
- CommitConsensus
- The point after which a successful decision is authoritative and must survive the failures covered by the protocol.
- CompactionLSM trees
- Background rewriting that merges sorted files, removes obsolete values, and reclaims tombstoned space.
- ConsensusRaft
- A protocol for agreeing on one authoritative sequence of decisions despite failures and message delays.
- Consistency modelConsistency
- A contract defining which values and operation orders clients are allowed to observe.
- Consumer groupKafka
- A set of consumers that divide log partitions so each partition is processed by one group member at a time.
D
- DurabilityFoundations
- The promise that acknowledged data survives the specified failures.
E
- Epoch / termRaft
- A monotonically increasing leadership generation used to distinguish current authority from stale actors.
- Eventual consistencyConsistency
- A guarantee that replicas converge after updates stop, without specifying when each read becomes current.
F
- Fault domainFoundations
- A set of components likely to fail together, such as one host, rack, zone, region, or provider.
- Fencing tokenCoordination
- A monotonically increasing value that lets a resource reject commands from an old lock or leader holder.
- FollowerRaft
- A replica that accepts the current leader’s ordered log and may become a candidate after timeout.
I
- IdempotencyReliability
- The property that repeating an operation with the same identity has the same effect as applying it once.
J
- JitterReliability
- Random variation added to retry timing so many clients do not retry in synchrony.
L
- LeaderReplication
- The node currently authorized to order writes or coordinate a replicated protocol.
- LinearizabilityConsistency
- A consistency model where operations appear atomic and respect real-time ordering.
- Log sequence / offsetKafka
- A position that identifies an ordered record within a log or partition.
- LSM treeStorage
- A write-optimized storage structure that buffers updates, flushes sorted immutable files, and compacts them later.
N
- Network partitionCAP
- A condition where groups of healthy nodes cannot exchange messages reliably or within the required deadline.
P
- Partial failureFoundations
- A failure where one component or communication path breaks while the rest of the system continues operating.
- Partition keyPartitioning
- The value used to choose where data or events are stored and therefore where ordering and hot spots occur.
Q
- QuorumConsensus
- A subset large enough to make progress while guaranteeing the overlap required by a protocol.
R
- Read repairReplication
- Updating stale replicas after a read detects older versions among its responses.
- Replication lagReplication
- The time or version distance between an authoritative replica and a follower.
- Retry budgetReliability
- A limit on retry traffic that prevents recovery attempts from overwhelming a struggling dependency.
- RPOMulti-region
- Recovery Point Objective: the maximum acceptable amount of data loss measured backward in time.
- RTOMulti-region
- Recovery Time Objective: the target duration for restoring an acceptable service level.
S
- SagaTransactions
- A sequence of local transactions with explicit compensating actions instead of one global atomic transaction.
- Split brainReplication
- A condition where multiple nodes believe they hold authority and accept conflicting work.
- SSTableLSM trees
- An immutable file containing keys in sorted order, commonly used by LSM storage engines.
T
- TombstoneStorage
- A deletion marker retained long enough to propagate and remove older copies during compaction.
- Two-phase commitTransactions
- An atomic commitment protocol that first prepares every participant, then announces commit or abort.
V
- Vector clockTime & ordering
- Version metadata that captures causal relationships and identifies concurrent updates.
W
- Write-ahead logStorage
- A durable sequential record written before an in-memory change is considered safe.
No matching term.Try a broader word or search the topic library.