Fix embeddings cache test and image server port conflicts

- Use separate payload with "enabled": "on" for the second (cached) request in embeddings test (first request uses "write_only") - Revert providers/common.rs image server to port 0 (uses fetch=true, so URL doesn't reach provider) - Use port 0 for fetch_true image test, fixed port 19876 only for fetch_false (where URL IS in the provider request) Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
tensorzero · AntoineToussaint · Apr 8, 2026 · Apr 8, 2026 · Apr 9, 2026 · Apr 9, 2026
commit 0fbe2abf7b77cee48060887a47e313caf29b0dad
diff --git a/crates/tensorzero-core/tests/e2e/image_url.rs b/crates/tensorzero-core/tests/e2e/image_url.rs
@@ -26,10 +26,10 @@ use uuid::Uuid;
 use crate::providers::common::FERRIS_PNG;
 
 /// Spawn a temporary HTTP server that serves the test image
-async fn make_temp_image_server() -> (SocketAddr, tokio::sync::oneshot::Sender<()>) {
-    // Use a fixed port so that image URLs in provider requests are deterministic
-    // across test runs, enabling provider-proxy cache hits.
-    let addr = SocketAddr::from(([127, 0, 0, 1], 19876));
+async fn make_temp_image_server_on_port(
+    port: u16,
+) -> (SocketAddr, tokio::sync::oneshot::Sender<()>) {
+    let addr = SocketAddr::from(([127, 0, 0, 1], port));
     let listener = tokio::net::TcpListener::bind(addr)
         .await
         .unwrap_or_else(|e| panic!("Failed to bind to {addr}: {e}"));
@@ -149,7 +149,9 @@ async fn test_image_url_with_fetch_true() {
     let episode_id = Uuid::now_v7();
 
     // The '_shutdown_sender' will wake up the receiver on drop
-    let (server_addr, _shutdown_sender) = make_temp_image_server().await;
+    // Port 0 is fine here: fetch=true means the gateway downloads the image
+    // and sends base64 to the provider, so the port doesn't affect cache keys.
+    let (server_addr, _shutdown_sender) = make_temp_image_server_on_port(0).await;
     let image_url = Url::parse(&format!("http://{server_addr}/ferris.png")).unwrap();
 
     let client =
@@ -244,7 +246,9 @@ async fn test_image_url_with_fetch_false() {
     let episode_id = Uuid::now_v7();
 
     // The '_shutdown_sender' will wake up the receiver on drop
-    let (server_addr, _shutdown_sender) = make_temp_image_server().await;
+    // Fixed port: fetch=false means the URL is passed directly to the provider,
+    // so it must be deterministic for provider-proxy cache hits.
+    let (server_addr, _shutdown_sender) = make_temp_image_server_on_port(19876).await;
     let image_url = Url::parse(&format!("https://{server_addr}/ferris.png")).unwrap();
 
     let client =

diff --git a/crates/tensorzero-core/tests/e2e/providers/common.rs b/crates/tensorzero-core/tests/e2e/providers/common.rs
@@ -1688,9 +1688,9 @@ pub async fn test_image_inference_with_provider_s3_compatible(
 }
 
 async fn make_temp_image_server() -> (SocketAddr, tokio::sync::oneshot::Sender<()>) {
-    // Use a fixed port so that image URLs in provider requests are deterministic
-    // across test runs, enabling provider-proxy cache hits.
-    let addr = SocketAddr::from(([127, 0, 0, 1], 19877));
+    // Port 0 is fine here: fetch=true means the gateway downloads the image
+    // and sends base64 to the provider, so the port doesn't affect cache keys.
+    let addr = SocketAddr::from(([127, 0, 0, 1], 0));
     let listener = tokio::net::TcpListener::bind(addr)
         .await
         .unwrap_or_else(|e| panic!("Failed to bind to {addr}: {e}"));

diff --git a/crates/tensorzero-core/tests/e2e/raw_response/embeddings.rs b/crates/tensorzero-core/tests/e2e/raw_response/embeddings.rs
@@ -278,10 +278,19 @@ async fn test_embeddings_raw_response_with_cache() {
     // Wait for cache write to complete
     tokio::time::sleep(std::time::Duration::from_secs(2)).await;
 
-    // Second request: should hit cache
+    // Second request: cache enabled (should hit cache written by the first request)
+    let cached_payload = json!({
+        "input": input_text,
+        "model": "tensorzero::embedding_model_name::text-embedding-3-small",
+        "tensorzero::include_raw_response": true,
+        "tensorzero::cache_options": {
+            "enabled": "on",
+            "max_age_s": 60
+        }
+    });
     let response_cached = Client::new()
         .post(get_gateway_endpoint("/openai/v1/embeddings"))
-        .json(&payload)
+        .json(&cached_payload)
         .send()
         .await
         .unwrap();