Glossary

A

Actor. An entity with its own state and behavior that processes messages one at a time. The basic unit of computation in the actor model. See Actor.

ActorPath. The unique hierarchical name for an actor in a system, e.g. actor-ts://my-app/user/api/sessions/user-42. See Actor paths.

ActorRef. A typed handle to an actor — the value you tell. See Actor.

ActorSelection. A lookup description for an actor by path string; resolved on demand to an ActorRef. See Actor paths.

ActorSystem. The top-level container for actors; one (or a few) per process. Owns the dispatcher, scheduler, supervisor tree, and extensions. See ActorSystem.

AllForOneStrategy. A supervisor strategy where one child’s failure causes the same directive to apply to every sibling. See Supervision.

Ask. Request/response over actors — ask(ref, msg, timeout) returns a Promise for the reply. See Ask pattern.

At-least-once delivery. A message-delivery guarantee where duplicates are possible but losses are not. See Delivery.

B

Backoff supervisor. A supervisor that restarts its child with exponential backoff between attempts. See Backoff supervisor.

Become. Replace the current message handler with a new one. See Become and stash.

Behavior. In the typed API, a value describing how to handle the next message. Composed with Behaviors.* combinators. See Behaviors.

Broadcast. A router strategy (or a Broadcast<T> wrapper) that sends a message to every routee. See Strategies.

C

Circuit breaker. A three-state wrapper (closed/open/half-open) that fails fast after sustained downstream failures. See Circuit breaker.

Cluster. A group of actor systems that gossip membership and route messages across the network. See Cluster overview.

ClusterSharding. Distributes per-key actors across cluster nodes with automatic rebalancing. See Sharding overview.

ClusterSingleton. Exactly one actor cluster-wide, hosted on the leader. See Singleton overview.

Command. In event sourcing, a request to do something — distinct from the event that records what happened. See PersistentActor.

CoordinatedShutdown. Phased shutdown DSL with task registration per named phase. See Coordinated shutdown.

CRDT. Conflict-free Replicated Data Type — a data type whose merge operation is commutative, associative, and idempotent, so gossip convergence works without coordination. See Distributed data.

D

DeadLetter. The synthetic recipient for messages with no live target. See ActorSystem.

Death watch. An actor’s mechanism for being notified when another actor stops. See Death watch.

Directive. The supervisor’s decision for a child failure: Restart / Resume / Stop / Escalate. See Supervision.

Discriminated union. A TypeScript type pattern where each variant has a tag field (kind) — the framework’s convention for messages. See Messages.

Dispatcher. Schedules an actor’s message processing onto the JavaScript event loop. See Dispatchers.

DistributedData. CRDT-replicated state across the cluster. See Distributed data overview.

DistributedPubSub. Topic-based publish/subscribe across the cluster. See PubSub.

Downing strategy. Logic that decides which side of a network partition wins and which downs itself. See Downing strategies.

DurableStateActor. Persists a single state snapshot per persistenceId (vs. PersistentActor’s event log). See Durable state.

E

Entity. In sharding, a single actor identified by an entity ID. See Sharding overview.

Envelope. The internal wrapper around a message in transit: { message, sender, context?, trace? }.

Event. In event sourcing, the persisted fact recording that something happened. Distinct from a command. See PersistentActor.

Event adapter. Schema-evolution upcaster for events. See Migration overview.

Event sourcing. A persistence model where state is derived by replaying a log of events. See Persistence overview.

Event stream. System-wide pub/sub bus, scoped to one ActorSystem. See Event stream.

Extension. Plugin mechanism — lazy-initialized, system-scoped services (cluster, persistence, metrics, …). See ActorSystem.

F

Failure detector. Phi-accrual detector that flags unreachable peers based on missing heartbeats. See Failure detector.

FIFO. First-in-first-out — the default mailbox order. See Mailboxes.

FSM. Finite-state-machine pattern; both in-memory and event-sourced variants exist. See FSM overview.

G

Gossip. The cluster’s membership-propagation protocol — members exchange views periodically until convergence. See Cluster overview.

Guardian. Top-level system actors: /user, /system, /deadLetters. See ActorSystem.

H

HOCON. Human-Optimized Configuration Object Notation — the config format actor-ts uses. See Configuration.

J

Journal. The append-only event store for PersistentActors. See Persistence overview.

K

Kill. System message that raises ActorKilledError — triggers the supervisor strategy. See Poison pill and Kill.

L

Lease. Coordination primitive — exactly-one actor holds a named lease at a time. Used to prevent split-brain for singletons and sharding coordinators. See Coordination.

LogContext / MDC. Mapped Diagnostic Context — per-async-stack key-value store that propagates with tell and across cluster hops. See Logging.

M

Mailbox. The per-actor message queue. Default unbounded FIFO; bounded and priority variants ship. See Mailboxes.

Message. Any value tell’d to an actor. Convention: discriminated union tagged by kind. See Messages.

N

Node. In a cluster, a single ActorSystem instance bound to a host:port. See Cluster overview.

O

onReceive. An actor’s method that handles each incoming message. See Actor.

OneForOneStrategy. The default supervisor strategy — a failing child gets the directive; siblings are unaffected. See Supervision.

P

Passivation. Stopping an idle sharded entity to free memory; next message for the same ID re-spawns it. See Sharding overview.

persist. In PersistentActor, the call that writes an event to the journal and invokes the callback after the event has been applied via onEvent. See PersistentActor.

persistenceId. The unique identifier for an entity’s event stream in the journal.

PersistentActor. Event-sourced actor base class. See PersistentActor.

Phi accrual. The failure detector’s underlying algorithm — suspicion accumulates based on heartbeat history.

pipeTo. Route a Promise’s eventual result into an actor’s mailbox as a message. See Future patterns.

PoisonPill. System message that triggers a graceful stop-after-drain. See Poison pill and Kill.

Pool. A router that creates its routees from Props. See Pool vs group.

Projection. Read-side view built from a PersistentActor’s event stream. See Projections.

Props. Immutable description of how to construct an actor — factory + supervisor strategy + dispatcher + mailbox. See Props.

R

Reachability. Whether the failure detector currently sees a peer as heartbeating.

Receive timeout. Timer that fires when no user message has arrived in N ms. See Receive timeout.

Rebalance. Sharding’s process of moving shards between nodes when membership changes. See Rebalance.

Remember entities. Sharding option that persists the active entity-ID set so they re-spawn after a coordinator failover. See Remember entities.

replyTo. Convention: a field on a request message that carries the ActorRef to reply to. See Messages.

Region. The per-node sharding actor that hosts entities for shards it owns and routes others. See Sharding overview.

Restart. Supervisor directive — throw away the broken instance, build a fresh one. See Supervision.

Resume. Supervisor directive — keep the state, skip the failing message. See Supervision.

Routee. An actor that receives messages routed by a Router.

Router. An actor that forwards each message to one (or all) of its routees per a strategy. See Router.

S

Scheduler. System-level scheduler — runs delayed and periodic tasks. Distinct from per-actor TimerScheduler. See Timers and scheduling.

Seed node. An address a joining node contacts to enter the cluster. See Joining and seeds.

Shard. A subset of entity IDs hashed together. Owned by exactly one node at a time. See Sharding overview.

Snapshot. Periodic state dump that shortens recovery for PersistentActors. See Snapshots.

Split-brain. When a network partition splits the cluster into two halves that both think the other is gone. See Downing strategies.

Stash. Park the currently-handled message for later replay. See Become and stash.

Supervisor strategy. Per-parent logic that maps a child’s error to a directive. See Supervision.

T

tell. Fire-and-forget send. The default actor verb. See Messages.

Terminated. System message delivered to watchers when a watched actor stops. See Death watch.

Tombstone. In gossip, a record kept after a member is removed to prevent stale gossip from resurrecting it.

Tracer. Observability abstraction for distributed tracing — typically wraps an OpenTelemetry SDK. See Tracing.

ts-pattern. Library used for match().exhaustive() dispatch on discriminated unions. See Pattern matching.

Typed actor. The functional alternative to subclassing Actor — actors as Behavior<T> values. See Typed overview.

U

Unbecome. Restore the previous behavior after a become that stacked it. See Become and stash.

Unreachable. A cluster member the failure detector flags as not heartbeating. Not yet officially downed. See Failure detector.

V

Vector clock. Logical clock used in replicated event sourcing to track concurrent updates. See Replicated ES — Vector clocks.

W

Watch / unwatch. Subscribe (or unsubscribe) to the Terminated notification for another actor. See Death watch.

Weakly-up. A transient member state — gossip-visible but not yet leader-confirmed. See Weakly-up.