rajvarun77 opened a new pull request, #3338: URL: https://github.com/apache/brpc/pull/3338
### What problem does this PR solve? Issue Number: #3109 Problem Summary: Inline redis protocol support (#3024) accepts any alpha-leading line as an inline command. The HTTP/2 client connection preface (`PRI * HTTP/2.0\r\n\r\nSM\r\n\r\n`) is alpha-leading, so `RedisCommandParser::ConsumeImpl` parsed it as an inline command (first token `PRI`) and returned `PARSE_OK` instead of `PARSE_ERROR_TRY_OTHERS`. As a result, protocol auto-detection in `InputMessenger` never fell through to the HTTP/2 handler, and a gRPC client calling a brpc server with `redis_service` enabled failed with: ``` code = Unavailable desc = connection closed before server preface received ``` ### What is changed and the side effects? Changed: - In the inline branch of `RedisCommandParser::ConsumeImpl` (`src/brpc/redis_command.cpp`), detect the HTTP/2 connection preface — both a full match and a not-yet-complete prefix — and return `PARSE_ERROR_TRY_OTHERS`, leaving the buffer untouched so detection falls through to the HTTP/2 protocol. - The check is gated on the first byte (`'P'`) to avoid overhead on normal inline commands. Why match the preface rather than blacklist command names: some HTTP methods are also valid redis commands (e.g. `GET`), so a name-based blacklist is inherently ambiguous. The HTTP/2 preface is a fixed magic string that no redis command begins with, so matching it is unambiguous and has no false positives (no redis command starts with `PR…`). - Added `RedisTest.inline_does_not_eat_h2_preface` (`test/brpc_redis_unittest.cpp`) covering: the full preface → `TRY_OTHERS`; a partial prefix → `TRY_OTHERS`; and a genuine inline command sharing the leading byte (`PING`) → `PARSE_OK`. Side effects: - Performance effects: negligible — one extra first-byte comparison per inline command, and a short `memcmp` only when it is `'P'`. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
