Cohesive Systems logoCOHESIVE SYSTEMS

Search Cohesive Systems

Ready

Search Cohesive Systems

Find product pages, building blocks, technical articles, and graph definitions.

Application Hosts

Application Hosts are runtime containers for application code, request handling, background work, dependency management, and operational concerns.

An application host may expose HTTP endpoints, subscribe to queues, run workers, schedule jobs, host actors, execute workflow activities, or maintain projections.

Application hosts commonly provide:

  • Lifecycle management.
  • Dependency injection or component composition.
  • Configuration.
  • Request or message pipelines.
  • Logging and telemetry.
  • Health checks.
  • Graceful shutdown and restart behavior.

In the model, an application host is a concrete realization mechanism for one or more observers and their interaction edges.

In stateless web hosts, a request handler often realizes a short-lived observer for one operation. It may load an entity state observation, interpret an input as a command, and attempt a transition, but it usually does not own the entity's transition boundary across requests. Correctness therefore depends on explicit concurrency control, such as expected-version checks, rather than on the request observer itself.

Session affinity can preserve locality for cached state or repeated interactions, but it does not by itself provide exclusive transition ownership unless the host also guarantees serialized access to the entity boundary.

An application host may load, validate, and interpret canonical transition models or process graphs. Long-lived continuation must be pinned to an exact supported definition identity, semantic revision, and normalized content identity rather than resolved only by a display name or current runtime registration.

The host may also run effect dispatchers and adapters, but those components must not mutate authoritative entity state directly or redefine request, signal, reply, retry, timeout, or recovery semantics through dependency-injection wiring. Returned results re-enter through explicit observer, wait, continuation, and transition boundaries.

Lifecycle control such as inspect, signal, pause, continue, restart, cancel, or terminate may be exposed through HTTP, CLI, or another protocol. The protocol binding is a host concern; stable command identity, authorization, idempotency, fencing, safe-point, and outcome meanings belong to the modeled execution contract.

Related concepts: execution kernel, realization, runtimes, transition models, process graphs, observer, entity, effect, interaction, idempotency, recovery, network, compute, infrastructure, concurrency control.