Add assertion after send_prefix_cache_requests in both prefix cache tests

threcc · threcc · commit 6046001b1d0c · 2026-03-10T17:38:41.000+01:00
diff --git a/tests/model_serving/model_server/llmd/test_llmd_singlenode_estimated_prefix_cache.py b/tests/model_serving/model_server/llmd/test_llmd_singlenode_estimated_prefix_cache.py
@@ -65,6 +65,7 @@ def test_singlenode_estimated_prefix_cache(
         successful = send_prefix_cache_requests(
             llmisvc=llmisvc, prompt=PREFIX_CACHE_PROMPT, token=llmisvc_token, count=NUM_REQUESTS
         )
+        assert successful == NUM_REQUESTS, f"Expected all {NUM_REQUESTS} requests to succeed, got {successful}"
 
         assert_prefix_cache_routing(
             prometheus=prometheus,
diff --git a/tests/model_serving/model_server/llmd/test_llmd_singlenode_precise_prefix_cache.py b/tests/model_serving/model_server/llmd/test_llmd_singlenode_precise_prefix_cache.py
@@ -67,6 +67,7 @@ def test_singlenode_precise_prefix_cache(
         successful = send_prefix_cache_requests(
             llmisvc=llmisvc, prompt=PREFIX_CACHE_PROMPT, token=llmisvc_token, count=NUM_REQUESTS
         )
+        assert successful == NUM_REQUESTS, f"Expected all {NUM_REQUESTS} requests to succeed, got {successful}"
 
         assert_prefix_cache_routing(
             prometheus=prometheus,

Original file line number	Diff line number	Diff line change
`@@ -65,6 +65,7 @@ def test_singlenode_estimated_prefix_cache(`
`65`	`65`	`successful = send_prefix_cache_requests(`
`66`	`66`	`llmisvc=llmisvc, prompt=PREFIX_CACHE_PROMPT, token=llmisvc_token, count=NUM_REQUESTS`
`67`	`67`	`)`
	`68`	`+ assert successful == NUM_REQUESTS, f"Expected all {NUM_REQUESTS} requests to succeed, got {successful}"`
`68`	`69`
`69`	`70`	`assert_prefix_cache_routing(`
`70`	`71`	`prometheus=prometheus,`
Original file line number	Diff line number	Diff line change
`@@ -67,6 +67,7 @@ def test_singlenode_precise_prefix_cache(`
`67`	`67`	`successful = send_prefix_cache_requests(`
`68`	`68`	`llmisvc=llmisvc, prompt=PREFIX_CACHE_PROMPT, token=llmisvc_token, count=NUM_REQUESTS`
`69`	`69`	`)`
	`70`	`+ assert successful == NUM_REQUESTS, f"Expected all {NUM_REQUESTS} requests to succeed, got {successful}"`
`70`	`71`
`71`	`72`	`assert_prefix_cache_routing(`
`72`	`73`	`prometheus=prometheus,`