Skip to content

Switchyard lets LLM applications route traffic across models and providers while preserving native OpenAI and Anthropic API compatibility - enabling flexible model selection, benchmarking, and cost/performance optimization.

Rust

fix(relay): reject unsupported caller credential forwarding (#759)

Bryan Bednarskia7ba62d27b35
diff --git a/crates/switchyard-nemo-relay-plugin/README.md b/crates/switchyard-nemo-relay-plugin/README.md
index 8a7f485f4..981db02fc 100644
--- a/crates/switchyard-nemo-relay-plugin/README.md
+++ b/crates/switchyard-nemo-relay-plugin/README.md
@@ -199,10 +199,18 @@ credentials, one route targeting an `openai_chat` client can serve all three
caller formats. Separate targets and routes are not required solely for format
translation.
-Where `forward_auth` is supported, the caller and target must use the same
-credential family: OpenAI-compatible (Chat Completions and Responses) or
-Anthropic (Messages). Credential forwarding through Relay is also subject to
-the limitation tracked in [NVBug 6777302](https://nvbugs/6777302) (NVIDIA internal).
+The native Relay plugin rejects routes that use `forward_auth = true` during
+configuration validation and activation. This includes routing-model calls and
+alternate targets. Relay does not provide caller credentials to the plugin's
+provider calls. Secure forwarding support is tracked in
+[NeMo Relay #1108](https://github.com/NVIDIA/NeMo-Relay/issues/1108).
+
+For deployment-owned credentials, remove `forward_auth` or set it to `false`
+and configure `api_key_env` on each authenticated client. The two options cannot
+be enabled together. If each caller must use its own provider credential, use
+standalone `switchyard-server`. Standalone forwarding requires the caller and
+target to use the same credential family: OpenAI-compatible (Chat Completions
+and Responses) or Anthropic (Messages).
Support for provider-specific fields depends on the source and target formats.
Test any fields that your application relies on before deploying a translated
diff --git a/crates/switchyard-nemo-relay-plugin/src/config.rs b/crates/switchyard-nemo-relay-plugin/src/config.rs
index 2df6386b3..2d8db081f 100644
--- a/crates/switchyard-nemo-relay-plugin/src/config.rs
+++ b/crates/switchyard-nemo-relay-plugin/src/config.rs
@@ -30,7 +30,7 @@ pub(crate) struct SwitchyardConfig {
impl SwitchyardConfig {
pub(crate) fn load_runner(&self) -> Result<Runner, String> {
- match (&self.switchyard_config_path, &self.switchyard_config) {
+ let runner = match (&self.switchyard_config_path, &self.switchyard_config) {
(Some(path), None) => Runner::load(path).map_err(|error| error.to_string()),
(None, Some(config)) => toml::to_string(config)
.map_err(|error| format!("failed to serialize Switchyard configuration: {error}"))
@@ -41,7 +41,23 @@ impl SwitchyardConfig {
(None, None) => {
Err("configure one of switchyard_config_path or switchyard_config".to_string())
}
+ }?;
+ // Relay keeps caller credentials outside the request passed to execution plugins.
+ for model in runner.models() {
+ if runner
+ .route(model.id.as_str())
+ .and_then(|route| route.caller_auth())
+ .is_some()
+ {
+ return Err(format!(
+ "route {} uses forward_auth = true, which the native Relay plugin does not support; \
+ disable forward_auth and configure api_key_env for deployment-owned credentials, \
+ or use standalone switchyard-server for caller-owned credentials",
+ model.id
+ ));
+ }
}
+ Ok(runner)
}
}
@@ -49,6 +65,7 @@ impl SwitchyardConfig {
mod tests {
use std::io::Write;
+ use nemo_relay_plugin::{DiagnosticLevel, NativePlugin};
use serde_json::json;
use tempfile::NamedTempFile;
@@ -127,6 +144,103 @@ mod tests {
assert!(runner.route("switchyard/default").is_some());
}
+ // Both plugin validation and activation must reject every configuration source.
+ fn assert_forward_auth_rejected(deployment: Map<String, Value>) {
+ let source = toml::to_string(&deployment).unwrap();
+ // The same deployment remains supported by the standalone runner.
+ let runner = Runner::from_toml(&source).unwrap();
+ assert!(
+ runner
+ .route("switchyard/default")
+ .unwrap()
+ .caller_auth()
+ .is_some()
+ );
+ let mut file = NamedTempFile::new().unwrap();
+ write!(file, "{source}").unwrap();
+
+ for plugin_config in [
+ json!({"switchyard_config": deployment}),
+ json!({"switchyard_config_path": file.path()}),
+ ] {
+ let plugin_config = plugin_config.as_object().unwrap();
+ let diagnostics = crate::SwitchyardPlugin.validate(plugin_config);
+ assert_eq!(diagnostics.len(), 1);
+ assert!(matches!(diagnostics[0].level, DiagnosticLevel::Error));
+ assert_eq!(diagnostics[0].code, "switchyard.invalid_config");
+ for detail in [
+ "switchyard/default",
+ "forward_auth",
+ "api_key_env",
+ "switchyard-server",
+ ] {
+ assert!(diagnostics[0].message.contains(detail), "{diagnostics:?}");
+ }
+ assert!(matches!(
+ crate::parse_config(plugin_config).and_then(crate::runtime::SwitchyardRuntime::new),
+ Err(error) if error == diagnostics[0].message
+ ));
+ }
+ }
+
+ #[test]
+ fn rejects_forward_auth_for_all_provider_formats() {
+ for format in ["openai_chat", "openai_responses", "anthropic_messages"] {
+ let mut deployment = inline_deployment();
+ deployment["llm_clients"]["primary"]["format"] = json!(format);
+ deployment["llm_clients"]["primary"]["forward_auth"] = json!(true);
+ assert_forward_auth_rejected(deployment);
+ }
+ }
+
+ #[test]
+ fn rejects_forward_auth_on_routing_only_and_alternate_targets() {
+ for route in [
+ json!({
+ "id": "switchyard/default", "type": "advisor",
+ "executor_target": "default", "advisor_target": "forwarded"
+ }),
+ json!({
+ "id": "switchyard/default", "type": "random",
+ "targets": ["default", "forwarded"]
+ }),
+ ] {
+ let mut deployment = inline_deployment();
+ deployment["llm_clients"]["forwarded"] = json!({
+ "format": "openai_chat", "base_url": "https://example.test/v1",
+ "forward_auth": true
+ });
+ deployment["targets"]["forwarded"] = json!({
+ "id": "example/forwarded", "llm_client": "forwarded"
+ });
+ deployment["routes"]["default"] = route;
+ assert_forward_auth_rejected(deployment);
+ }
+ }
+
+ #[test]
+ fn accepts_deployment_credentials_and_unused_forward_auth_clients() {
+ let mut deployment = inline_deployment();
+ deployment["llm_clients"]["primary"]["forward_auth"] = json!(false);
+ // PATH supplies a non-secret, existing value without mutating process environment.
+ deployment["llm_clients"]["primary"]["api_key_env"] = json!("PATH");
+ deployment["llm_clients"]["unused"] = json!({
+ "format": "openai_chat", "base_url": "https://example.test/v1",
+ "forward_auth": true
+ });
+ deployment["targets"]["unused"] = json!({
+ "id": "example/unused", "llm_client": "unused"
+ });
+ let plugin_config = json!({"switchyard_config": deployment});
+ let plugin_config = plugin_config.as_object().unwrap();
+ assert!(crate::SwitchyardPlugin.validate(plugin_config).is_empty());
+ assert!(
+ crate::parse_config(plugin_config)
+ .and_then(crate::runtime::SwitchyardRuntime::new)
+ .is_ok()
+ );
+ }
+
#[test]
fn requires_exactly_one_configuration_source() {
let neither = SwitchyardConfig {
diff --git a/docs/integrations/nemo_relay.md b/docs/integrations/nemo_relay.md
index 66f38118b..e451466b7 100644
--- a/docs/integrations/nemo_relay.md
+++ b/docs/integrations/nemo_relay.md
@@ -190,10 +190,18 @@ credentials, one route targeting an `openai_chat` client can serve all three
caller formats. Separate targets and routes are not required solely for format
translation.
-Where `forward_auth` is supported, the caller and target must use the same
-credential family: OpenAI-compatible (Chat Completions and Responses) or
-Anthropic (Messages). Credential forwarding through Relay is also subject to
-the limitation tracked in [NVBug 6777302](https://nvbugs/6777302) (NVIDIA internal).
+The native Relay plugin rejects routes that use `forward_auth = true` during
+configuration validation and activation. This includes routing-model calls and
+alternate targets. Relay does not provide caller credentials to the plugin's
+provider calls. Secure forwarding support is tracked in
+[NeMo Relay #1108](https://github.com/NVIDIA/NeMo-Relay/issues/1108).
+
+For deployment-owned credentials, remove `forward_auth` or set it to `false`
+and configure `api_key_env` on each authenticated client. The two options cannot
+be enabled together. If each caller must use its own provider credential, use
+standalone `switchyard-server`. Standalone forwarding requires the caller and
+target to use the same credential family: OpenAI-compatible (Chat Completions
+and Responses) or Anthropic (Messages).
Support for provider-specific fields depends on the source and target formats.
Test any fields that your application relies on before deploying a translated