feat(processor): make Ollama model + request timeout configurable

.env.example

@@ -6,6 +6,8 @@ LISTEN_ADDR=:8080
 KAFKA_BROKER=localhost:9092
 ES_URL=http://localhost:9200
 OLLAMA_HOST=http://localhost:11434
+OLLAMA_MODEL=llama3.2:1b
+OLLAMA_REQUEST_TIMEOUT=30
 MODEL_PATH=./isolation_forest.joblib
 
 # Dashboard

docker-compose.yml

@@ -90,6 +90,8 @@ services:
       KAFKA_BROKER: kafka:9092
       ES_URL: http://elasticsearch:9200
       OLLAMA_HOST: http://ollama:11434
+      OLLAMA_MODEL: llama3.2:1b
+      OLLAMA_REQUEST_TIMEOUT: 30
       MODEL_PATH: /app/data/isolation_forest.joblib
     depends_on:
       kafka:

k8s/06-processing-service.yml

@@ -52,6 +52,10 @@ spec:
               value: "http://elasticsearch:9200"
             - name: OLLAMA_HOST
               value: "http://ollama:11434"
+            - name: OLLAMA_MODEL
+              value: "llama3.2:1b"
+            - name: OLLAMA_REQUEST_TIMEOUT
+              value: "30"
           resources:
             requests:
               memory: "512Mi"

processing-service/processor.py

@@ -35,9 +35,14 @@
 KAFKA_RECONNECT_BACKOFF_MS = 1000
 KAFKA_RECONNECT_BACKOFF_MAX_MS = 60000
 
-# Ollama LLM client — host is configurable for K8s / Docker networking
+# Ollama LLM client — host, model, and request timeout are all configurable
+# for K8s / Docker networking and for swapping models without a code change.
 OLLAMA_HOST = os.getenv("OLLAMA_HOST", "http://localhost:11434")
-_ollama_client = ollama.Client(host=OLLAMA_HOST)
+OLLAMA_MODEL = os.getenv("OLLAMA_MODEL", "llama3.2:1b")
+# Bounds how long a single Ollama request may block, so a hung Ollama
+# instance can't stall anomaly processing indefinitely.
+OLLAMA_REQUEST_TIMEOUT = float(os.getenv("OLLAMA_REQUEST_TIMEOUT", "30"))
+_ollama_client = ollama.Client(host=OLLAMA_HOST, timeout=OLLAMA_REQUEST_TIMEOUT)
 
 
 # ---------------------------------------------------------------------------
@@ -123,7 +128,7 @@ def run_ai_analysis(log_text, log_data, last_failure_time):
         return True, "AI Analysis Unavailable (Ollama cooldown)", last_failure_time
 
     try:
-        response = _ollama_client.chat(model='llama3.2:1b', messages=[
+        response = _ollama_client.chat(model=OLLAMA_MODEL, messages=[
             {
                 'role': 'system',
                 'content': (

processing-service/test_processor.py

@@ -11,12 +11,14 @@
 import pytest
 from sklearn.ensemble import IsolationForest
 
+import processor
 from processor import (
     KAFKA_RECONNECT_BACKOFF_MAX_MS,
     KAFKA_RECONNECT_BACKOFF_MS,
     KAFKA_RECONNECT_INITIAL_DELAY_SEC,
     KAFKA_RETRY_BACKOFF_MS,
     OLLAMA_COOLDOWN_SEC,
+    OLLAMA_MODEL,
     TRAIN_INTERVAL,
     check_rate_anomaly,
     has_keyword_indicator,
@@ -179,6 +181,41 @@ def test_run_ai_analysis_ollama_exception():
     assert new_failure > past_failure
 
 
+def test_run_ai_analysis_uses_configured_model():
+    """The chat() call must use OLLAMA_MODEL, not a hardcoded model name."""
+    log_data = {"service": "test"}
+    past_failure = time.time() - OLLAMA_COOLDOWN_SEC - 10
+
+    with patch("processor._ollama_client.chat") as mock_chat:
+        mock_chat.return_value = {"message": {"content": "Normal"}}
+        run_ai_analysis("some info log", log_data, past_failure)
+
+    assert mock_chat.call_args.kwargs["model"] == OLLAMA_MODEL
+
+
+def test_ollama_model_env_var_overrides_default(monkeypatch):
+    """OLLAMA_MODEL is read from the environment at import time (with
+    "llama3.2:1b" as the default), and run_ai_analysis always uses the
+    current module-level value rather than a hardcoded literal - so
+    reconfiguring OLLAMA_MODEL changes the model used with no code change.
+    (monkeypatch.setattr on the already-imported module is used instead of
+    importlib.reload, because reloading processor.py re-executes its
+    module-level prometheus_client.Counter/Histogram registrations and
+    raises "Duplicated timeseries in CollectorRegistry".)
+    """
+    assert processor.OLLAMA_MODEL == "llama3.2:1b"  # unchanged default
+
+    monkeypatch.setattr(processor, "OLLAMA_MODEL", "custom-model:latest")
+
+    log_data = {"service": "test"}
+    past_failure = time.time() - OLLAMA_COOLDOWN_SEC - 10
+    with patch("processor._ollama_client.chat") as mock_chat:
+        mock_chat.return_value = {"message": {"content": "Normal"}}
+        processor.run_ai_analysis("some info log", log_data, past_failure)
+
+    assert mock_chat.call_args.kwargs["model"] == "custom-model:latest"
+
+
 def test_build_kafka_consumer_sets_backoff_config():
     with patch("processor.KafkaConsumer") as mock_consumer:
         class EmptyConsumer: