Merge pull request #7 from Badrivishal/prometheus

Dockerfile

@@ -6,7 +6,15 @@ RUN pip install --no-cache-dir --upgrade pip \
     && if [ -f requirements.txt ]; then pip install --no-cache-dir -r requirements.txt; else pip install --no-cache-dir gradio; fi
 COPY . .
 
+ENV DEBIAN_FRONTEND noninteractive
+RUN apt-get update && \
+    apt-get upgrade -yq ca-certificates && \
+    apt-get install -yq --no-install-recommends \
+    prometheus-node-exporter
+
 EXPOSE 7860
+EXPOSE 8000
+EXPOSE 9100
 ENV GRADIO_SERVER_NAME=0.0.0.0
 
-CMD ["python", "app.py"]
+CMD bash -c "prometheus-node-exporter --web.listen-address=':9100' & python app.py"

app.py

@@ -7,7 +7,7 @@ import gradio as gr
 from datasets import load_dataset
 from sentence_transformers import SentenceTransformer
 from huggingface_hub import InferenceClient
-
+from prometheus_helper import PrometheusHelper
 # --- Credit ---
 # Most of this code was generated using AI (ChatGPT, GitHub Copilot). 
 # Please refer to the references of the report for concrete links to the respective AI interactions.
@@ -16,9 +16,11 @@ from huggingface_hub import InferenceClient
 INDEX_FILE = "xkcd.index"
 META_FILE = "meta.pkl"
 CHAT_MODEL = os.getenv("CHAT_MODEL", "meta-llama/Meta-Llama-3-8B-Instruct")
+prometheus_helper = PrometheusHelper()
 
 # --- Build / load index ---
 def build_index():
+    prometheus_helper.start_index_build_timer()
     print("Building FAISS index...")
     ds = load_dataset("olivierdehaene/xkcd", split="train")
     model = SentenceTransformer("all-MiniLM-L6-v2")
@@ -48,6 +50,7 @@ def build_index():
     with open(META_FILE, "wb") as f:
         pickle.dump(meta, f)
 
+    prometheus_helper.stop_index_build_timer()
     return index, meta
 
 def get_index():
@@ -77,11 +80,16 @@ def respond(
     else:
         return "⚠️ Please sign in with your Hugging Face account (top of the page) or set the HF_TOKEN environment variable"
 
+    prometheus_helper.start_request_timer()
     # Embed the query and search FAISS
+    prometheus_helper.start_faiss_index_search_timer()
     query_vec = embedder.encode([message], convert_to_numpy=True)
     D, I = index.search(query_vec, 5)
     candidates = [meta[int(i)] for i in I[0]]
 
+    prometheus_helper.stop_faiss_index_search_timer()
+    prometheus_helper.start_chat_model_call_timer()
+
     context = "\n".join(
         f"[{c['id']}] {c['title']}\nTranscript: {c['transcript']}\nExplanation: {c['explanation']}"
         for c in candidates
@@ -109,6 +117,8 @@ EXPLANATION
         temperature=0.0, # TODO
     )
 
+    prometheus_helper.stop_chat_model_call_timer()
+
     # Be tolerant to slight schema differences
     try:
         choice = resp.choices[0]
@@ -132,14 +142,17 @@ EXPLANATION
                 img_url = json.load(url)["img"]
                 print(f'Got image url: {img_url}')
             
+            prometheus_helper.record_frequency(int(id))
             return [out_text, gr.Image(value=img_url)]
         except ValueError:
             print("Couldn't parse xkcd ID or get image! That should not happen.")
-    
+    prometheus_helper.record_request(True)
+    prometheus_helper.stop_request_timer()
     return out_text
 
 if __name__ == "__main__":
     # --- UI ---
+    prometheus_helper.setup_prometheus()
     with gr.Blocks(theme='gstaff/xkcd') as demo:
         gr.Markdown("# xkcd Comic Finder")
         gr.Markdown(

prometheus.yml

@@ -0,0 +1,11 @@
+global:
+  scrape_interval: 15s
+
+scrape_configs:
+  - job_name: 'gradio_app'
+    static_configs:
+      - targets: ['xkcd_finder_app:8000']
+
+  - job_name: 'node_exporter'
+    static_configs:
+      - targets: ['xkcd_finder_app:9100']

prometheus_helper.py

@@ -0,0 +1,56 @@
+from prometheus_client import start_http_server, Counter, Summary, Histogram
+import time
+
+class PrometheusHelper:
+    def __init__(self):
+        self.request_counter = Counter('app_requests_total', 'Total number of requests')
+        self.successful_requests_counter = Counter('app_successful_requests_total', 'Total number of successful requests')
+        self.failed_requests_counter = Counter('app_failed_requests_total', 'Total number of failed requests')
+        self.request_duration_summary = Summary('app_request_duration_seconds', 'Time spent processing request')
+        self.comic_frequency = Counter('comic_frequency', 'Frequency of comics being selected', ['comic_id'])
+        self.index_build_start_time = None
+        self.index_build_duration = Summary('index_build_duration_seconds', 'Time spent building the index')
+        self.faiss_index_search_start_time = None
+        self.faiss_index_search_duration = Summary('faiss_index_search_duration_seconds', 'Time spent searching the index')
+        self.chat_model_call_start_time = None
+        self.chat_model_call_duration = Summary('chat_model_call_duration_seconds', 'Time spent calling the chat model')
+        # self.frequency_histogram = Histogram('comic_frequency', 'Frequency of comics being selected')
+
+
+    def start_faiss_index_search_timer(self):
+        self.faiss_index_search_start_time = time.time()
+    
+    def stop_faiss_index_search_timer(self):
+        self.faiss_index_search_duration.observe(time.time() - self.faiss_index_search_start_time)
+
+    def start_chat_model_call_timer(self):
+        self.chat_model_call_start_time = time.time()
+    
+    def stop_chat_model_call_timer(self):
+        self.chat_model_call_duration.observe(time.time() - self.chat_model_call_start_time)
+
+    def record_request(self, success: bool):
+        self.request_counter.inc()
+        if success:
+            self.successful_requests_counter.inc()
+        else:
+            self.failed_requests_counter.inc()
+
+    def start_index_build_timer(self):
+        self.index_build_start_time = time.time()
+
+    def stop_index_build_timer(self):
+        self.index_build_duration.observe(time.time() - self.index_build_start_time)
+
+    def setup_prometheus(self):
+        start_http_server(8000)
+
+    def start_request_timer(self):
+        self.start_time = time.time()
+    
+    def stop_request_timer(self):
+        self.request_duration_summary.observe(time.time() - self.start_time)
+
+    def record_frequency(self, comic_id: int):
+        self.comic_frequency.labels(comic_id=comic_id).inc()
+    

requirements.txt

@@ -3,4 +3,5 @@ faiss-cpu
 transformers
 sentence-transformers
 datasets
-gradio[oauth]
+gradio[oauth]
+prometheus_client==0.16.*