Retry Policy

A retry policy attaches to any client<llm> and retries requests that fail due to network errors.

Basic Example

retry_policy MyPolicyName {
    max_retries 3
}

client<llm> MyClient {
    provider anthropic
    retry_policy MyPolicyName
    options {
        model "claude-sonnet-4-20250514"
        api_key env.ANTHROPIC_API_KEY
    }
}

Fields

Field	Type	Required	Description
`max_retries`	int	Yes	Additional retries after initial failure
`strategy`	Strategy	No	Retry timing strategy (default: `constant_delay`)

Strategies

constant_delay

Waits a fixed duration between retries.

retry_policy MyPolicyName {
    max_retries 3
    strategy {
        type constant_delay
        delay_ms 200
    }
}

Option	Type	Default	Description
`type`	`constant_delay`	-	Strategy type
`delay_ms`	int	200	Delay in milliseconds

exponential_backoff

Increases delay exponentially between retries.

retry_policy ExponentialPolicy {
    max_retries 5
    strategy {
        type exponential_backoff
        delay_ms 200
        multiplier 1.5
        max_delay_ms 10000
    }
}

Option	Type	Default	Description
`type`	`exponential_backoff`	-	Strategy type
`delay_ms`	int	200	Initial delay in milliseconds
`multiplier`	float	1.5	Multiplier applied after each retry
`max_delay_ms`	int	10000	Maximum delay in milliseconds

Interaction with Timeouts

When using both retry policies and timeouts:

Each retry attempt gets the full timeout duration
A timeout triggers the retry mechanism
Total time = (attempts × timeout) + retry delays

retry_policy Exponential {
    max_retries 3
    strategy {
        type exponential_backoff
    }
}

client<llm> MyClient {
    provider openai
    retry_policy Exponential
    options {
        model "gpt-4o"
        http {
            request_timeout_ms 30000
        }
    }
}

Basic Example​

Fields​

Strategies​

constant_delay​

exponential_backoff​

Interaction with Timeouts​

See Also​