Body limitCopy Anchor Link

Route.connectBodyLimit(maxBytes) caps the entire decoded HTTP request body for unary RPCs. Oversized messages become a Connect resource_exhausted error (HTTP 429), not Ktor's default 413.

Decoded size, not wire size aloneCopy Anchor Link

The cap is evaluated against the decoded message:

• Uncompressed bodies are bounded as bytes arrive (including chunked requests without Content-Length), with a Content-Length fast path that rejects early.
• Compressed bodies are bounded after decompression, so a small gzip bomb cannot inflate past the limit unnoticed.

routing {
    route("/com.example.v1.MyService") {
        connectBodyLimit(maxBytes = 4 * 1024 * 1024)
        install(ContentNegotiation) {
            connectJson()
        }
        myService(MyServiceHandler)
    }
}

Routes outside the connectBodyLimit scope keep ordinary Ktor body-limit behaviour. That is intentional: REST and Connect can disagree on overflow policy in the same application.

Pair with compressionCopy Anchor Link

If you accept Content-Encoding: gzip, install connectBodyLimit as well. UnaryCompressionGuard does not bound expansion by itself.

Streaming receive limitsCopy Anchor Link

Use ConnectStreaming for client-, server-, and bidirectional-streaming RPCs:

import io.github.ichizero.connect.ktor.streaming.ConnectStreaming

routing {
    install(ConnectStreaming) {
        maxMessageSize = 8 * 1024 * 1024
    }
    myService(MyServiceHandler)
}

The plugin limits each request envelope payload, excluding its 5-byte prefix, without buffering the stream. The value must be positive; without configuration, handlers use 4 MiB. It does not affect unary requests or response sizes. A child route can install its own setting, and a manually bound handler's explicit maxMessageSize overrides the plugin. Generated source stays unchanged; recompile it against the updated library to pick up the inline handlers' configuration lookup.

Keep full-duplex routes outside connectBodyLimit / Ktor RequestBodyLimit, which can buffer small chunks until EOF and prevent request/response interleaving. Whole-body limits also count envelope prefixes and all messages together, so they are not a substitute for streaming receive limits.