Module: Rubino::LLM::FailoverReason

Defined in:: lib/rubino/llm/error_classifier.rb

Overview

Why an API call failed — determines recovery strategy. A faithful subset of the reference FailoverReason: only the CORE reasons this gem can actually act on are ported. Provider-niche reasons (thinking_signature, llama_cpp_grammar, encrypted_content, long_context_tier, image_too_large, …) are intentionally dropped.

The load-bearing default is ‘unknown → retryable`: an unclassifiable provider blip backs off and retries rather than aborting.

Constant Summary collapse

AUTH = 401/403 — invalid credential, don’t retry as-is

:auth

BILLING = 402 / credit exhaustion — don’t retry

:billing

RATE_LIMIT = 429 — backoff then retry

:rate_limit

OVERLOADED = 503/529 — provider overloaded, backoff

:overloaded

SERVER_ERROR = 500/502 — internal server error, retry

:server_error

TIMEOUT = connection/read timeout / transport drop — retry

:timeout

CONTEXT_OVERFLOW = context too large — compress, not failover

:context_overflow

MODEL_NOT_FOUND = 404 / invalid model — fallback to another model

:model_not_found

FORMAT_ERROR = 400 bad request — abort + fallback

:format_error

UNKNOWN = unclassifiable — retry with backoff

:unknown

ALL =

[
  AUTH, BILLING, RATE_LIMIT, OVERLOADED, SERVER_ERROR, TIMEOUT,
  CONTEXT_OVERFLOW, MODEL_NOT_FOUND, FORMAT_ERROR, UNKNOWN
].freeze