Struct FallbackEvaluator 

pub struct FallbackEvaluator<'a> { /* private fields */ }

Evaluates fallback conditions and selects alternative upstreams.

The evaluator checks both pre-request conditions (health, budget) and post-response conditions (error codes, latency) to determine if a fallback should be triggered.

Implementations

impl<'a> FallbackEvaluator<'a>

pub fn new( config: &'a FallbackConfig, tried_upstreams: &'a [String], current_attempt: u32, ) -> Self

Create a new fallback evaluator.

Arguments

• config - The fallback configuration from the route
• tried_upstreams - List of upstreams that have already been tried
• current_attempt - Current fallback attempt number (0 = primary)

pub fn can_attempt_fallback(&self) -> bool

Check if more fallback attempts are allowed.

pub fn should_fallback_before_request( &self, upstream_id: &str, is_healthy: bool, is_budget_exhausted: bool, current_model: Option<&str>, ) -> Option<FallbackDecision>

Check pre-request conditions to determine if fallback should be used.

This checks conditions that can be evaluated before sending the request:

• Health check failures (if on_health_failure is enabled)
• Budget exhaustion (if on_budget_exhausted is enabled)

Arguments

• upstream_id - The upstream that would be used
• is_healthy - Whether the upstream is currently healthy
• is_budget_exhausted - Whether the token budget is exhausted
• current_model - The model being requested (for model mapping)

Returns

Some(FallbackDecision) if fallback should be used, None otherwise.

pub fn should_fallback_after_response( &self, upstream_id: &str, status_code: u16, latency_ms: u64, current_model: Option<&str>, ) -> Option<FallbackDecision>

Check post-response conditions to determine if fallback should be used.

This checks conditions that can only be evaluated after receiving a response:

• Error status codes (if on_error_codes is configured)
• Latency threshold exceeded (if on_latency_threshold_ms is configured)

Arguments

• upstream_id - The upstream that was used
• status_code - HTTP status code from the response
• latency_ms - Request latency in milliseconds
• current_model - The model being requested (for model mapping)

Returns

Some(FallbackDecision) if fallback should be used, None otherwise.

pub fn should_fallback_on_connection_error( &self, upstream_id: &str, error_message: &str, current_model: Option<&str>, ) -> Option<FallbackDecision>

Check if fallback should be triggered due to a connection error.

Arguments

• upstream_id - The upstream that failed to connect
• error_message - Description of the connection error
• current_model - The model being requested (for model mapping)

Returns

Some(FallbackDecision) if fallback should be used, None otherwise.

pub fn next_fallback(&self) -> Option<&FallbackUpstream>

Get the next untried fallback upstream.

This returns the first fallback upstream that:

1. Has not been tried yet
2. Is not marked as skip_if_unhealthy with unhealthy status

Note: Health status check is the caller’s responsibility since we don’t have access to the health check state here.

pub fn map_model(&self, upstream: &FallbackUpstream, model: &str) -> String

Apply model mapping for a fallback upstream.

Supports glob patterns like gpt-4* to match gpt-4, gpt-4-turbo, etc.

Arguments

• upstream - The fallback upstream configuration
• model - The original model name

Returns

The mapped model name, or the original if no mapping matches.