Struct AgentManager 

pub struct AgentManager { /* private fields */ }

Agent manager handling all external agents.

Supports both v1 and v2 protocol agents. V1 agents use simple request/response, while v2 agents support bidirectional streaming with capabilities, health reporting, metrics export, and flow control.

Implementations

impl AgentManager

pub async fn new(agents: Vec<AgentConfig>) -> SentinelResult<Self>

Create new agent manager.

Each agent gets its own semaphore for queue isolation, preventing a slow agent from affecting other agents (noisy neighbor problem). The concurrency limit is configured per-agent via max_concurrent_calls in the agent config.

Supports both v1 and v2 protocol agents based on the protocol_version field.

pub async fn process_request_headers( &self, ctx: &AgentCallContext, headers: &HashMap<String, Vec<String>>, route_agents: &[(String, FailureMode)], ) -> SentinelResult<AgentDecision>

Process request headers through agents.

Arguments

• ctx - Agent call context with correlation ID and metadata
• headers - Request headers to send to agents
• route_agents - List of (agent_id, failure_mode) tuples from filter chain

pub async fn process_request_body( &self, ctx: &AgentCallContext, data: &[u8], is_last: bool, route_agents: &[String], ) -> SentinelResult<AgentDecision>

Process request body chunk through agents.

pub async fn process_request_body_streaming( &self, ctx: &AgentCallContext, data: &[u8], is_last: bool, chunk_index: u32, bytes_received: usize, total_size: Option<usize>, route_agents: &[String], ) -> SentinelResult<AgentDecision>

Process a single request body chunk through agents (streaming mode).

Unlike process_request_body which is used for buffered mode, this method is designed for streaming where chunks are sent individually as they arrive.

pub async fn process_response_body_streaming( &self, ctx: &AgentCallContext, data: &[u8], is_last: bool, chunk_index: u32, bytes_sent: usize, total_size: Option<usize>, route_agents: &[String], ) -> SentinelResult<AgentDecision>

Process a single response body chunk through agents (streaming mode).

pub async fn process_response_headers( &self, ctx: &AgentCallContext, status: u16, headers: &HashMap<String, Vec<String>>, route_agents: &[String], ) -> SentinelResult<AgentDecision>

Process response headers through agents.

pub async fn process_websocket_frame( &self, route_id: &str, event: WebSocketFrameEvent, ) -> SentinelResult<AgentResponse>

Process a WebSocket frame through agents.

This is used for WebSocket frame inspection after an upgrade. Returns the agent response directly to allow the caller to access the websocket_decision field.

pub async fn initialize(&self) -> SentinelResult<()>

Initialize agent connections.

pub async fn shutdown(&self)

Shutdown all agents.

pub fn metrics(&self) -> &AgentMetrics

Get agent metrics.

pub fn get_agents_for_event(&self, event_type: EventType) -> Vec<String>

Get agent IDs that handle a specific event type.

This is useful for pre-filtering agents before making calls, e.g., to check if any agents handle WebSocket frames.

pub async fn get_v2_pool_metrics(&self) -> Vec<(String, Arc<MetricsCollector>)>

Get pool metrics collectors from all v2 agents.

Returns a vector of (agent_id, MetricsCollector) pairs for all v2 agents. These can be registered with the MetricsManager to include agent pool metrics in the /metrics endpoint output.

pub async fn export_v2_pool_metrics(&self) -> String

Export prometheus metrics from all v2 agent pools.

Returns the combined prometheus-formatted metrics from all v2 agent pools.

pub async fn get_v2_metrics_collector( &self, agent_id: &str, ) -> Option<Arc<MetricsCollector>>

Get a v2 agent’s metrics collector by ID.

Returns None if the agent doesn’t exist or is not a v2 agent.