server: concurrency fix + monitoring - add /metrics prometheus compatible endpoint (#5708)

* server: monitoring - add /metrics prometheus compatible endpoint

* server: concurrency issue, when 2 task are waiting for results, only one call thread is notified

* server: metrics - move to a dedicated struct

4 files changed, 32 insertions, 0 deletions

diff --git a/examples/server/tests/features/environment.py b/examples/server/tests/features/environment.py
index 13cc8410..09e82674 100644
--- a/examples/server/tests/features/environment.py
+++ b/examples/server/tests/features/environment.py
@@ -16,6 +16,8 @@ def before_scenario(context, scenario):
 
 
 def after_scenario(context, scenario):
+    if context.server_process is None:
+        return
     if scenario.status == "failed":
         if 'GITHUB_ACTIONS' in os.environ:
             print(f"\x1b[33;101mSCENARIO FAILED: {scenario.name} server logs:\x1b[0m\n\n")
diff --git a/examples/server/tests/features/server.feature b/examples/server/tests/features/server.feature
index 5f81d256..0139f89d 100644
--- a/examples/server/tests/features/server.feature
+++ b/examples/server/tests/features/server.feature
@@ -13,6 +13,7 @@ Feature: llama.cpp server
     And   1 slots
     And   embeddings extraction
     And   32 server max tokens to predict
+    And   prometheus compatible metrics exposed
     Then  the server is starting
     Then  the server is healthy
 
@@ -25,6 +26,7 @@ Feature: llama.cpp server
     And   <n_predict> max tokens to predict
     And   a completion request with no api error
     Then  <n_predicted> tokens are predicted matching <re_content>
+    And   prometheus metrics are exposed
 
     Examples: Prompts
       | prompt                           | n_predict | re_content                   | n_predicted |
diff --git a/examples/server/tests/features/steps/steps.py b/examples/server/tests/features/steps/steps.py
index 9c825fdb..051fd440 100644
--- a/examples/server/tests/features/steps/steps.py
+++ b/examples/server/tests/features/steps/steps.py
@@ -13,6 +13,7 @@ import aiohttp
 import openai
 from behave import step
 from behave.api.async_step import async_run_until_complete
+from prometheus_client import parser
 
 
 @step(u"a server listening on {server_fqdn}:{server_port}")
@@ -34,6 +35,8 @@ def step_server_config(context, server_fqdn, server_port):
     context.server_api_key = None
     context.server_continuous_batching = False
     context.server_embeddings = False
+    context.server_metrics = False
+    context.server_process = None
     context.server_seed = None
     context.user_api_key = None
 
@@ -82,6 +85,11 @@ def step_server_embeddings(context):
     context.server_embeddings = True
 
 
+@step(u'prometheus compatible metrics exposed')
+def step_server_metrics(context):
+    context.server_metrics = True
+
+
 @step(u"the server is starting")
 def step_start_server(context):
     start_server_background(context)
@@ -424,6 +432,23 @@ def step_check_options_header_value(context, cors_header, cors_header_value):
     assert context.options_response.headers[cors_header] == cors_header_value
 
 
+@step(u'prometheus metrics are exposed')
+@async_run_until_complete
+async def step_prometheus_metrics_exported(context):
+    async with aiohttp.ClientSession() as session:
+        async with await session.get(f'{context.base_url}/metrics') as metrics_response:
+            assert metrics_response.status == 200
+            assert metrics_response.headers['Content-Type'] == "text/plain; version=0.0.4"
+            metrics_raw = await metrics_response.text()
+            metric_exported = False
+            for metric in parser.text_string_to_metric_families(metrics_raw):
+                match metric.name:
+                    case "llamacpp:kv_cache_usage_ratio":
+                        assert len(metric.samples) > 0
+                        metric_exported = True
+            assert metric_exported, "No metrics exported"
+
+
 async def concurrent_requests(context, f_completion, *args, **kwargs):
     n_prompts = len(context.prompts)
     if context.debug:
@@ -753,6 +778,8 @@ def start_server_background(context):
         server_args.append('--cont-batching')
     if context.server_embeddings:
         server_args.append('--embedding')
+    if context.server_metrics:
+        server_args.append('--metrics')
     if context.model_alias is not None:
         server_args.extend(['--alias', context.model_alias])
     if context.n_ctx is not None:
diff --git a/examples/server/tests/requirements.txt b/examples/server/tests/requirements.txt
index 3e51b12d..334fa4a7 100644
--- a/examples/server/tests/requirements.txt
+++ b/examples/server/tests/requirements.txt
@@ -1,3 +1,4 @@
 aiohttp~=3.9.3
 behave~=1.2.6
 openai~=0.25.0
+prometheus-client~=0.20.0