Refacto: Configuration management

.gitignore

@@ -2,7 +2,7 @@
 .version.cache
 
 # Local app config file
-config.toml
+config.yaml
 
 # Created by https://www.toptal.com/developers/gitignore/api/osx,linux,eclipse,windows,visualstudio,visualstudiocode,dotenv
 # Edit at https://www.toptal.com/developers/gitignore?templates=osx,linux,eclipse,windows,visualstudio,visualstudiocode,dotenv

README.md

@@ -113,96 +113,98 @@ graph
 
 ### Run locally
 
-Create a local configuration file, a file named `config.toml` at the root of the project. The TOML file can be placed anywhere in the folder or in any parent directory.
+Create a local configuration file, a file named `config.yaml` at the root of the project. The TOML file can be placed anywhere in the folder or in any parent directory.
 
-```toml
-# config.toml
+```yaml
+# config.yaml
 # /!\ All the file values are for example, you must change them
-[api]
-# root_path = "[api-path]"
-
-[oidc]
-algorithms = ["RS256"]
-api_audience = "[aad_app_id]"
-issuers = ["https://login.microsoftonline.com/[tenant_id]/v2.0"]
-jwks = "https://login.microsoftonline.com/common/discovery/v2.0/keys"
-
-[monitoring]
-
-[monitoring.logging]
-app_level = "DEBUG" # Enum: "NOSET", "DEBUG", "INFO", "WARN", "ERROR", "FATAL", "CRITICAL"
-sys_level = "WARN" # Enum: "NOSET", "DEBUG", "INFO", "WARN", "ERROR", "FATAL", "CRITICAL"
-
-[monitoring.azure_app_insights]
-connection_str = "InstrumentationKey=[key];[...]"
-
-[persistence]
-cache = "redis" # Enum: "redis"
-search = "qdrant" # Enum: "qdrant"
-store = "cosmos" # Enum: "cache", "cosmos"
-stream = "redis" # Enum: "redis"
-
-[persistence.qdrant]
-host = "[host]"
-
-[persistence.redis]
-db = 0
-host = "[host]"
-
-[persistence.cosmos]
-# Containers "conversation" (/user_id), "message" (/conversation_id), "user" (/dummy), "usage" (/user_id) must exist
-url = "https://[deployment].documents.azure.com:443"
-database = "[db_name]"
-
-[ai]
-
-[ai.openai]
-ada_deploy_id = "ada"
-ada_max_tokens = 2049
-api_base = "https://[deployment].openai.azure.com"
-gpt_deploy_id = "gpt"
-gpt_max_tokens = 4096
-
-[ai.azure_content_safety]
-api_base = "https://[deployment].cognitiveservices.azure.com"
-api_token = "[api_token]"
-max_length = 1000
-
-[tools]
-
-[tools.azure_form_recognizer]
-api_base = "https://[deployment].cognitiveservices.azure.com"
-api_token = "[api_token]"
-
-[tools.bing]
-search_url = "https://api.bing.microsoft.com/v7.0/search"
-subscription_key = "[api_token]"
-
-[tools.tmdb]
-bearer_token = "[jwt_token]"
-
-[tools.news]
-api_key = "[api_token]"
-
-[tools.listen_notes]
-api_key = "[api_token]"
-
-[tools.open_weather_map]
-api_key = "[api_token]"
-
-[tools.google_places]
-api_key = "[api_token]"
-
-[[tools.azure_cognitive_search]]
-api_key = "[api_token]"
-content_key = "content"
-displayed_name = "My search engine"
-index_name = "[index_name]"
-language = "en-US"
-semantic_configuration = "default"
-service_name = "[service_name]"
-top_k = 10
-usage = "Useful when you need legal content contained in one of the codes (incl. civil, commercial, labor, electoral) derived from French legislation. Use this in priority for legal content in France."
+api: {}
+    # root_path: API_PATH
+
+oidc:
+    algorithms: ["RS256"]
+    api_audience: e9d5f20f-7f14-4204-a9a2-0d91d6af5c82
+    issuers:
+        [
+            "https://login.microsoftonline.com/72f988bf-86f1-41af-91ab-2d7cd011db47/v2.0",
+            "https://login.microsoftonline.com/9188040d-6c67-4c5b-b112-36a304b66dad/v2.0",
+        ]
+    jwks: https://login.microsoftonline.com/common/discovery/v2.0/keys
+
+monitoring:
+    logging:
+        app_level: DEBUG
+        sys_level: WARN
+    azure_app_insights:
+        connection_str: InstrumentationKey=XXX;IngestionEndpoint=https://westeurope-5.in.applicationinsights.azure.com/;LiveEndpoint=https://westeurope.livediagnostics.monitor.azure.com
+
+persistence:
+    cache:
+        type: redis
+        config:
+            db: 0
+            host: localhost
+    search:
+        type: qdrant
+        config:
+            host: localhost
+    store:
+        type: cosmos
+        config:
+            url: https://private-gpt.documents.azure.com:443
+            database: private-gpt
+    stream:
+        type: redis
+        config:
+            db: 0
+            host: localhost
+
+ai:
+    openai:
+        ada_deploy_id: ada
+        ada_max_tokens: 2049
+        api_base: https://private-gpt-fr.openai.azure.com
+        gpt_deploy_id: gpt
+        gpt_max_tokens: 16384
+    azure_content_safety:
+        api_base: https://moaw-dev-clesne-moaw-search-acs.cognitiveservices.azure.com
+        api_token: XXX
+        max_length: 1000
+
+tools:
+    azure_form_recognizer:
+        api_base: https://private-gpt.cognitiveservices.azure.com
+        api_token: XXX
+    bing:
+        search_url: https://api.bing.microsoft.com/v7.0/search
+        subscription_key: XXX
+    tmdb:
+        bearer_token: XXX
+    news:
+        api_key: XXX
+    listen_notes:
+        api_key: XXX
+    open_weather_map:
+        api_key: XXX
+    google_places:
+        api_key: XXX
+    azure_cognitive_search:
+        - api_key: XXX
+            displayed_name: Légifrance
+            index_name: legifrance-xml
+            language: fr-FR
+            semantic_configuration: default
+            service_name: samples-v2
+            top_k: 10
+            usage: Useful when you need legal content contained in one of the codes (incl. civil, commercial, labor, electoral) derived from French legislation. Use this in priority for France legal related data.
+        - api_key: XXX
+            displayed_name: Open Food Facts
+            index_name: openfoodfacts-json-v2
+            language: en-US
+            semantic_configuration: default
+            service_name: samples-v2
+            top_k: 10
+            usage: Useful when you need details about a food product (incl. nutrient levels, allergens, additives, stores, generic name). Use this in priority food related data.
 ```
 
 Now, you can either run the application as container or with live reload. For development, it is recommended to use live reload. For demo, it is recommended to use the container.

cicd/helm/private-gpt/Chart.lock

@@ -6,4 +6,4 @@ dependencies:
   repository: oci://registry-1.docker.io/bitnamicharts
   version: 18.0.4
 digest: sha256:d2b04f80cb92a439086247637917406f435e287a10c81d4ff421a5aa66c4c44b
-generated: "2023-11-21T13:21:45.838475+01:00"
+generated: "2023-11-28T18:30:15.805929+01:00"

cicd/helm/private-gpt/templates/conversation-api-config.yaml

@@ -6,7 +6,7 @@ metadata:
     {{- include "private-gpt.labels" . | nindent 4 }}
     app.kubernetes.io/component: conversation-api
 data:
-  config.toml: |
+  config.yaml: |
     [oidc]
     algorithms = ["{{ join "\",\"" .Values.oidc.algorithms | required "A value for .Values.oidc.algorithms is required" }}"]
     api_audience = {{ .Values.oidc.api_audience | quote | required "A value for .Values.oidc.api_audience is required" }}
@@ -85,7 +85,6 @@ data:
     # Tool {{ .displayed_name | required "A value for .displayed_name is required" }}
     [[tools.azure_cognitive_search]]
     api_key = {{ .api_key | quote | required "A value for .api_key is required" }}
-    content_key = {{ .content_key | quote | required "A value for .content_key is required" }}
     displayed_name = {{ .displayed_name | quote | required "A value for .displayed_name is required" }}
     index_name = {{ .index_name | quote | required "A value for .index_name is required" }}
     service_name = {{ .service_name | quote | required "A value for .service_name is required" }}

cicd/helm/private-gpt/templates/conversation-api-deployment.yaml

@@ -49,8 +49,8 @@ spec:
               port: http
           volumeMounts:
             - name: config
-              mountPath: /app/config.toml
-              subPath: config.toml
+              mountPath: /app/config.yaml
+              subPath: config.yaml
             - name: tmp
               mountPath: /tmp
           env:

cicd/helm/private-gpt/values.yaml

@@ -67,7 +67,6 @@ tools:
   azure_cognitive_search:
     []
     # - api_key: xxx
-    #   content_key: content
     #   displayed_name: Légifrance
     #   index_name: legifrance-xml
     #   service_name: samples-v2

src/conversation-api/Makefile

@@ -34,7 +34,8 @@ start:
 		--no-server-header \
 		--port 8081 \
 		--proxy-headers \
-		--reload
+		--reload \
+
 
 build:
 	$(docker) build \

src/conversation-api/ai/contentsafety.py

@@ -1,38 +1,21 @@
-# Import utils
-from utils import build_logger, get_config
-
-# Import misc
 from azure.core.credentials import AzureKeyCredential
 from fastapi import HTTPException, status
+from helpers.config import CONFIG
+from helpers.logging import build_logger
 from tenacity import retry, stop_after_attempt, wait_random_exponential
 import azure.ai.contentsafety as azure_cs
 import azure.core.exceptions as azure_exceptions
 
 
-###
-# Init misc
-###
-
 _logger = build_logger(__name__)
 
-###
-# Init Azure Content Safety
-###
 
 # Score are following: 0 - Safe, 2 - Low, 4 - Medium, 6 - High
 # See: https://review.learn.microsoft.com/en-us/azure/cognitive-services/content-safety/concepts/harm-categories?branch=release-build-content-safety#severity-levels
 ACS_SEVERITY_THRESHOLD = 2
-ACS_API_BASE = get_config(
-    ["ai", "azure_content_safety"], "api_base", str, required=True
-)
-ACS_API_TOKEN = get_config(
-    ["ai", "azure_content_safety"], "api_token", str, required=True
-)
-ACS_MAX_LENGTH = get_config(
-    ["ai", "azure_content_safety"], "max_length", int, required=True
-)
 acs_client = azure_cs.ContentSafetyClient(
-    ACS_API_BASE, AzureKeyCredential(ACS_API_TOKEN)
+    credential=AzureKeyCredential(CONFIG.ai.azure_content_safety.api_token.get_secret_value()),
+    endpoint=CONFIG.ai.azure_content_safety.api_base,
 )
 
 
@@ -45,7 +28,7 @@ class ContentSafety:
     async def is_moderated(self, prompt: str) -> bool:
         _logger.debug(f"Checking moderation for text: {prompt}")
 
-        if len(prompt) > ACS_MAX_LENGTH:
+        if len(prompt) > CONFIG.ai.azure_content_safety.max_length:
             _logger.info(f"Message ({len(prompt)}) too long for moderation")
             raise HTTPException(
                 status_code=status.HTTP_400_BAD_REQUEST,