Fei Deng

This commit significantly **improves the robustness of `Transfer-Encoding` header parsing** and **enhances HTTP/1 compliance** across various components within `pingora-core`. It introduces new utility functions like `find_last_te_token` and `is_chunked_encoding_from_headers` in `pingora-core/src/protocols/http/v1/common.rs` for more accurate detection. Specifically, the **HTTP/1 client** now correctly disables keepalive for HTTP/1.0 responses with `Transfer-Encoding`, while the **HTTP/1 server** disables keepalive when both `Transfer-Encoding` and `Content-Length` are present and rejects HTTP/1.0 requests using `Transfer-Encoding`. This **refactoring and bug fix** ensures better adherence to **RFC 9112** and prevents potential protocol ambiguities, impacting both client and server-side HTTP/1 communication and subrequest handling.

This commit primarily focuses on **refactoring** to enhance code clarity and maintainability across two key modules. It simplifies the error handling expression within the `poll_read_ready` function in `pingora-boringssl/src/boring_tokio.rs`, improving the readability of asynchronous I/O logic. Concurrently, an error message for size mismatches is streamlined within the `get_opt_sized` function in `pingora-core/src/protocols/l4/ext.rs`. This **maintenance** work contributes to better internal **code quality** and developer experience without introducing new features or altering external behavior.

This commit primarily **fixes a bug** in the **TLS connection establishment logic** within `pingora-core`'s `boringssl_openssl` connector. It corrects the conditional check for `second_keyshare` in the `connect` function, ensuring proper handling of peer options during TLS handshakes and resolving potential negotiation issues. Additionally, a minor **refactoring** in `pingora-boringssl` simplifies error handling within the `poll_read_ready` function by using a direct function reference. This change enhances the robustness of the **TLS connector** module and improves code clarity for async I/O operations.

This commit **bumps the Minimum Supported Rust Version (MSRV) to 1.84**, updating the `README.md` and `clippy.toml` configuration to reflect this new requirement. This **maintenance** change enables significant **refactoring** across core modules like `pingora-boringssl`, `pingora-cache`, `pingora-core`, and `pingora-proxy`. Specifically, it modernizes error handling by replacing `io::Error::new(io::ErrorKind::Other, ...)` with the more concise `io::Error::other` and improves code clarity by replacing `map_or` patterns with the `is_none_or` method. These changes leverage newer Rust language features, enhancing code readability and maintainability throughout the project.

This commit introduces a **bug fix** to the **`pingora-header-serde`** crate, specifically addressing **buffer capacity calculation** during header serialization and deserialization. It **simplifies** the internal logic of `serialize` and `deserialize` functions by replacing complex dynamic sizing with a single, maximum buffer size. This change enhances the **robustness and efficiency** of header processing, preventing potential allocation issues and streamlining the underlying code.

This commit introduces a **new feature** to the **`pingora-cache`** and **`pingora-lru`** libraries by **exposing internal LRU shard information**. The `Manager` struct within `pingora-cache` now provides public methods such as `shards`, `shard_weight`, `shard_len`, and `get_shard_for_key` to retrieve details about the cache's sharding, including shard count, total weight, item count, and a key's assigned shard. Additionally, the `Lru` struct in `pingora-lru` gains a `shard_weight` method to report the total weight of items within a specific shard. This enhancement provides greater introspection into the LRU cache's internal state, enabling more effective monitoring and debugging of cache distribution and performance.

This commit **enhances** the `pingora-header-serde` crate by modifying its header serialization and deserialization logic. It introduces **dynamic buffer allocation** to reserve additional space as needed, allowing the system to handle significantly **larger header sizes** up to a new upper limit. This change prevents potential buffer overflows and improves robustness when processing HTTP headers that exceed previous fixed-size limits. The update primarily affects the `serialize` and `deserialize` functions within the header handling subsystem, ensuring more flexible and resilient header processing.

This commit introduces a **refactoring** within the **`pingora-header-serde`** crate, specifically updating its **error context** and **frame content size reporting** mechanisms. The change involves migrating string handling in functions like `into_error` and `get_frame_content_size` to utilize the `ImmutStr` type. This transition aims to improve string management, potentially enhancing performance or memory efficiency for error messages and size reports that are often consumed by logging systems. The update ensures more consistent and optimized string representation across these critical serialization components.

This commit introduces a **new capability** to the **`pingora-header-serde`** crate by **enhancing error reporting** for header decompression. It now **logs the size of the decompressed header frame content** directly into error messages, providing critical context for debugging issues related to header processing. This change significantly improves the **observability** of header decompression failures. Additionally, it includes a minor **refactoring** of error context parameter types within functions like `into_error` and `decompress_to_buffer`.

This commit introduces a **new capability** to the **Pingora cache eviction** module, specifically within `pingora-cache/src/eviction/lru.rs`. It adds a new function, `cleanup_temp_files`, designed to identify and remove orphaned temporary cache files that might accumulate over time. This cleanup process is integrated into the `load_shards` function, ensuring that stale temporary files are automatically removed when cache shards are loaded. This **maintenance improvement** helps prevent disk space waste and enhances the overall reliability and health of the cache system.

This commit introduces **new capabilities** to the **`pingora-http`** crate by adding `case_header_iter` and `has_case` methods to the `RequestHeader` and `ResponseHeader` structs. These methods allow for more precise iteration and checking of **case-sensitive HTTP headers**. Additionally, it **refactors** the internal header serialization logic, utilizing a new helper function `header_to_h1_wire` for improved consistency and maintainability. This enhancement provides developers with finer-grained control over header manipulation and strengthens the core HTTP processing components.

This commit **enhances the reliability and diagnostic capabilities** of the **`pingora-header-serde`** library. It **fixes incorrect error messages** within Zstd compression and significantly **improves debug logging** for header parsing failures, particularly affecting functions like `compress` and `buf_to_http_header`. To further ensure correctness, **new test cases** have been introduced for header serialization and deserialization. This **maintenance and bug fix** effort provides clearer error reporting and more robust handling of HTTP headers, improving the overall stability of the system.

This commit introduces a **new feature** to the **`pingora-cache`** library, specifically enhancing the `ConcurrentHashTable` component. It adds a public function, `get_shard_at_idx`, which allows external callers to directly access a specific internal shard of the hash table by its index. This provides more granular control and introspection capabilities for the concurrent data structure, enabling more advanced usage patterns or debugging.

This commit introduces a **new feature** by adding a `for_each` method to the **`ConcurrentHashTable`** within the **`pingora-cache`** crate. The new `for_each` method allows callers to **iterate over all key-value pairs** stored in the concurrent hash table, providing a safe and efficient way to process its contents. This enhancement improves the usability of `ConcurrentHashTable` by offering a standard mechanism for inspecting its elements, which can be crucial for debugging, statistics gathering, or other aggregate operations.

This commit introduces a **new feature** to the **`pingora-load-balancing`** crate by adding a `BackendSummary` callback mechanism to the `HealthCheck` trait and `HttpHealthCheck` struct. This enhancement allows for **custom backend summary reporting**, enabling users to collect and provide more detailed information about the health status of individual backends. Consequently, **health check logging messages** are updated to leverage this new `backend_summary` callback, significantly improving the granularity and usefulness of reported backend health data. This change provides greater observability and flexibility in monitoring backend performance and availability.