Agents

Agents are similar to Angular’s Services (but without dependency injection), and provide Yew with an Actor Model. Agents can be used to route messages between components independently of where they sit in the component hierarchy, or they can be used to create shared state between different components. Agents can also be used to offload computationally expensive tasks from the main thread which renders the UI. There is also planned support for using agents to allow Yew applications to communicate across tabs (in the future).

In order for agents to run concurrently, Yew uses web-workers.

Lifecycle

agent lifecycle diagramagent lifecycle diagram

Types of Agents

Reaches

  • Context - There will exist at most one instance of a Context Agent at any given time. Bridges will spawn or connect to an already spawned agent on the UI thread. This can be used to coordinate state between components or other agents. When no bridges are connected to this agent, the agent will disappear.

  • Job - Spawn a new agent on the UI thread for every new bridge. This is good for moving shared but independent behavior that communicates with the browser out of components. (TODO verify) When the task is done, the agent will disappear.

  • Public - Same as Context, but runs on its own web worker.

  • Private - Same as Job, but runs on its own web worker.

  • Global (WIP)

Communication between Agents and Components

Bridges

A bridge allows bi-directional communication between an agent and a component. Bridges also allow agents to communicate with one another.

A use_bridge hook is also provided to create bridges in a function component.

Dispatchers

A dispatcher allows uni-directional communication between a component and an agent. A dispatcher allows a component to send messages to an agent.

Overhead

Agents that use web workers (i.e. Private and Public) will incur a serialization overhead on the messages they send and receive. They use bincode to communicate with other threads, so the cost is substantially higher than just calling a function. Unless the cost of computation will outweigh the cost of message passing, you should use agents running on the UI thread (i.e. Job or Context).

Further reading

  • The pub_sub example shows how components can use agents to communicate with each other.