added fancier prompt that remembers state

README.md

@@ -6,3 +6,15 @@ You'd have to specify a bunch of docker containers that you'd like a Kubernetes
 
 So this is a managed Kubernetes service lol. But not really. Because you're still very much running Kubernetes - think of this tool as an assistant. 
 
+Problems I faced:
+1. lots of evicted pods
+2. failed readiness checks and it was unclear how to debug them and how to get started
+3. should have been a way to pass in a Dockerfile and generate a corresponding kubernetes YAML
+4. shouldn't have to learn the semantics of what it means to 
+
+
+Possible next steps:
+1. download the chat history of the kubernetes slack channel with user questions and expert answers, and use that to finetune gpt3 using openai api
+2. https://github.com/ht2/gpt_content_indexing/
+https://loft.sh/blog/kubernetes-monitoring-dashboards-5-best-open-source-tools/
+

app.py

@@ -9,21 +9,23 @@
 
 app = Flask(__name__)
 
+simple_in_memory_db = {}
 
 @app.route('/api', methods=['POST'])
-def api():
+def api(error_message, prompt, config):
+    # get user id
+    user_id = request.headers.get('user_id')
+
     # get the prompt and config objects from the request
     prompt = request.form['prompt']
     config = request.form['config']
-
-
-    # prompt = request.form.get('prompt')
-    # config = request.form.get('config')
 
     # make request to OpenAI API with prompt and config
     # return response from API to caller
-    final_prompt = handler(prompt, config)
-
+    suggestion = handler(prompt, config)
+    simple_in_memory_db[user_id].append((suggestion,))
+    # server remembers what we suggested to the caller so we can use it in the next request
+    return jsonify(suggestion)
 
 if __name__ == '__main__':
     app.run(debug=True, host='localhost', port=5000)

k8s.py

@@ -6,35 +6,12 @@
 openai.api_key = os.getenv("OPENAI_API_KEY", "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")
 
 def construct_prompt(query_str, config):
-    prompt = '''
-    You are a Kubernetes debugging assistant.
-    Given the following config object, and a question, suggest what actions can be taken with references ("SOURCES").
-    ALWAYS return a "SOURCES" part in your answer.
-
-    QUESTION: My cluster has 72 pods, a large chunk of which are the blenderbot pod. How do I prevent there from being so many replicas of the blenderbot?
-    =========
-    Config: apiVersion: apps/v1\nkind: Deployment\nmetadata:\n  name: blenderbot\n  namespace: chirpy\nspec:\n  selector:\n    matchLabels:\n      app: blenderbot\n  template:\n    metadata:\n      labels:\n        app: blenderbot\n    spec:\n      containers:\n      - image: 742352046111.dkr.ecr.us-east-1.amazonaws.com/chirpy/blenderbot:latest\n        name: main\n        ports:\n        - containerPort: 80\n          name: main\n        readinessProbe:\n          httpGet:\n            path: /\n            port: 80\n          initialDelaySeconds: 120\n      tolerations:\n      - effect: NoSchedule\n        key: node_type\n        operator: Equal\n        value: gpu\napiVersion: v1\nkind: Service\nmetadata:\n  name: blenderbot\n  namespace: chirpy\nspec:\n  ports:\n  - port: 80\n    protocol: TCP\n    targetPort: 80\n  selector:\n    app: blenderbot\n
-    =========
-    FINAL ANSWER: Check your node groups, its possible the pod isn't being able to get scheduled on any node and so is constantly respawning. 
-    SOURCES: https://kubernetes.io/docs/concepts/scheduling-eviction/assign-pod-node/
-
-    QUESTION: How to eat vegetables using kubectl?
-    =========
-    Config: apiVersion: apps/v1\nkind: Deployment\nmetadata:\n  name: blenderbot\n  namespace: chirpy\nspec:\n  selector:\n    matchLabels:\n      app: blenderbot\n  template:\n    metadata:\n      labels:\n        app: blenderbot\n    spec:\n      containers:\n      - image: 742352046111.dkr.ecr.us-east-1.amazonaws.com/chirpy/blenderbot:latest\n        name: main\n        ports:\n        - containerPort: 80\n          name: main\n        readinessProbe:\n          httpGet:\n            path: /\n            port: 80\n          initialDelaySeconds: 120\n      tolerations:\n      - effect: NoSchedule\n        key: node_type\n        operator: Equal\n        value: gpu\napiVersion: v1\nkind: Service\nmetadata:\n  name: blenderbot\n  namespace: chirpy\nspec:\n  ports:\n  - port: 80\n    protocol: TCP\n    targetPort: 80\n  selector:\n    app: blenderbot\n
-
-    =========
-    FINAL ANSWER: You can't eat vegetables using kubernetes. You can only eat them using your mouth.
-    SOURCES:
-
-    QUESTION: {query_str}
-    =========
-    {config}
-    =========
-    FINAL ANSWER:
-
+    prompt = f'''
+    You are a Kubernetes debugging assistant. I've tried the following commands on my cluster, and the outputs are below. 
+    Give me another command to try, with no English text, only kubectl commands.
+    {query_str}
     '''
-    return prompt.format(query_str=query_str, config=config)
-
+
 def handler(prompt, config):
     # TODO: only get the relevant parts of the config
     # TODO: decide what the relevant parts of the config are