Implements the core retry decision logic for the Algolia client. It selects tryable hosts based on call kind (read/write), decides whether to retry based on HTTP status codes and error types (network, timeout, rate-limit), and tracks host state (up/down/expiry). The strategy is configurable via transport.Configuration and is designed to be extracted into any Go HTTP client that needs robust retry with host failover.

Manages individual host state with up/down/retry-after/expiry semantics. Tracks last update, retry count, and provides methods to mark hosts up/down. Includes logic to parse Retry-After headers and apply backoff. A self-contained component that could be reused in any multi-host HTTP client to implement health-based load balancing.

A stateful, WHATWG-compliant Server-Sent Events decoder. Parses SSE streams line-by-line, handles event fields (id, event, data, retry), and accumulates data across multiple lines. Includes a scanLines function and retry parsing. This is a hand-written component that is fully reusable for any Go application needing to consume SSE streams, independent of Algolia-specific types.

Defines a custom error type that wraps multiple underlying errors encountered when all hosts have been exhausted. Implements the error interface, supports Is() for sentinel matching, and provides access to the list of intermediate errors and a correlation ID. Demonstrates a clean pattern for aggregating errors in retry scenarios and can be extracted as a general-purpose 'all attempts failed' error.