Merge pull request #157 from NillionNetwork/feat/config_settings_improvements

feat: added configs and improved dependency checks

Author: jcabrero

README.md

@@ -140,19 +140,18 @@ For a complete production setup with custom images:
 ```shell
 # 1a. Generate the Production 1 image
 python3 ./scripts/docker-composer.py --prod \
-  -f docker/compose/docker-compose.nilai-prod-1.yml \
-  --image 'nillion/nilai-api:latest=public.ecr.aws/k5d9x2g2/nilai-api:v0.2.0-alpha2' \
-  --image 'nillion/nilai-vllm:latest=public.ecr.aws/k5d9x2g2/nilai-vllm:v0.2.0-alpha2' \
-  --image 'nillion/nilai-attestation:latest=public.ecr.aws/k5d9x2g2/nilai-attestation:v0.2.0-alpha2' \
-  --testnet \
+  -f docker/compose/docker-compose.nilai-prod-2.yml \
+  --image 'nillion/nilai-api:latest=public.ecr.aws/k5d9x2g2/nilai-api:v0.2.0-rc0' \
+  --image 'nillion/nilai-attestation:latest=public.ecr.aws/k5d9x2g2/nilai-attestation:v0.2.0-rc0' \
+  --image 'nillion/nilai-vllm:latest=public.ecr.aws/k5d9x2g2/nilai-vllm:v0.2.0-rc0' \
   -o production-compose.yml
 
 # 1b. Generate the Production 2 image
 python3 ./scripts/docker-composer.py --prod \
-  -f docker/compose/docker-compose.nilai-prod-2.yml \
-  --image 'nillion/nilai-api:latest=public.ecr.aws/k5d9x2g2/nilai-api:v0.2.0-alpha2' \
-  --image 'nillion/nilai-vllm:latest=public.ecr.aws/k5d9x2g2/nilai-vllm:v0.2.0-alpha2' \
-  --image 'nillion/nilai-attestation:latest=public.ecr.aws/k5d9x2g2/nilai-attestation:v0.2.0-alpha2' \
+  -f docker/compose/docker-compose.nilai-prod-1.yml \
+  --image 'nillion/nilai-api:latest=public.ecr.aws/k5d9x2g2/nilai-api:v0.2.0-rc0' \
+  --image 'nillion/nilai-attestation:latest=public.ecr.aws/k5d9x2g2/nilai-attestation:v0.2.0-rc0' \
+  --testnet \
   -o production-compose.yml
 
 

docker-compose.prod.yml

@@ -21,7 +21,7 @@ services:
     env_file:
       - .env.mainnet
     environment:
-      - POSTGRES_DB=nuc-nilai-mainnet2
+      - POSTGRES_DB=${POSTGRES_DB_NUC}
     volumes:
       - ${FILES}/nilai-api/config.yaml:/app/nilai-api/src/nilai_api/config/config.yaml
   caddy:

docker-compose.testnet.prod.yml

@@ -8,7 +8,6 @@ services:
       - .env.testnet
     environment:
       - AUTH_STRATEGY=nuc
-      - POSTGRES_DB=${POSTGRES_DB_TESTNET}
     volumes:
       - ${FILES}/testnet/nilai-api/config.yaml:/app/nilai-api/src/nilai_api/config/config.yaml
   grafana:

docker/compose/docker-compose.nilai-prod-1.yml

@@ -13,7 +13,7 @@ services:
       memlock: -1
       stack: 67108864
     env_file:
-      - .env
+      - .env.mainnet
     restart: unless-stopped
     depends_on:
       etcd:

nilai-api/alembic/versions/9ddf28cf6b6f_feat_new_user_config_and_web_rate_limits.py

@@ -22,7 +22,13 @@
 def upgrade() -> None:
     # ### commands auto generated by Alembic - please adjust! ###
     op.add_column(
-        "query_logs", sa.Column("web_search_calls", sa.Integer(), nullable=False)
+        "query_logs",
+        sa.Column(
+            "web_search_calls",
+            sa.Integer(),
+            server_default=sa.text("0"),
+            nullable=False,
+        ),
     )
     op.add_column("users", sa.Column("rate_limits", sa.JSON(), nullable=True))
     op.drop_column("users", "web_search_ratelimit_minute")

nilai-api/pyproject.toml

@@ -10,9 +10,7 @@ authors = [
 ]
 requires-python = ">=3.12"
 dependencies = [
-    "accelerate>=1.1.1",
     "alembic>=1.14.1",
-    "cryptography>=43.0.1",
     "fastapi[standard]>=0.115.5",
     "gunicorn>=23.0.0",
     "nilai-common",
@@ -22,20 +20,22 @@ dependencies = [
     "httpx>=0.27.2",
     "nilrag>=0.1.11",
     "openai>=1.99.2",
-    "pg8000>=1.31.2",
     "prometheus_fastapi_instrumentator>=7.0.2",
     "asyncpg>=0.30.0",
-    "greenlet>=3.1.1",
-    "redis>=5.2.1",
-    "authlib>=1.4.1",
-    "verifier",
+    "redis>=6.4.0",
     "web3>=7.8.0",
     "click>=8.1.8",
     "nuc-helpers",
     "nuc>=0.1.0",
     "pyyaml>=6.0.1",
     "trafilatura>=1.7.0",
     "secretvaults",
+    "pydantic>=2.0.0",
+    "ecdsa>=0.19.0",
+    "secp256k1>=0.14.0",
+    "hexbytes>=1.2.0",
+    "eth-account>=0.13.0",
+    "sentence-transformers>=5.1.1",
     "e2b-code-interpreter>=1.0.3",
 ]
 
@@ -47,6 +47,4 @@ build-backend = "hatchling.build"
 [tool.uv.sources]
 nilai-common = { workspace = true }
 nuc-helpers = { workspace = true }
-
-# TODO: Remove this once the secretvaults package is released with the fix
 secretvaults = { git = "https://github.com/jcabrero/secretvaults-py", rev = "main" }

nilai-api/src/nilai_api/config/config-a779.yaml

@@ -0,0 +1,31 @@
+# Web Search Configuration
+web_search:
+  api_key: null
+  api_path: "https://api.search.brave.com/res/v1/web/search"
+  count: 3
+  lang: "en"
+  country: "us"
+  timeout: 20.0
+  max_concurrent_requests: 20
+  rps: 20
+
+# Rate Limiting Configuration
+rate_limiting:
+  user_rate_limit: null # For-good rate limit
+  user_rate_limit_minute: 10
+  user_rate_limit_hour: 100
+  user_rate_limit_day: 500
+  web_search_rate_limit_minute: null
+  web_search_rate_limit_hour: null
+  web_search_rate_limit_day: null
+  web_search_rate_limit: 100 # For-good rate limit
+  model_concurrent_rate_limit:
+    meta-llama/Llama-3.2-1B-Instruct: 45
+    meta-llama/Llama-3.2-3B-Instruct: 50
+    meta-llama/Llama-3.1-8B-Instruct: 30
+    cognitivecomputations/Dolphin3.0-Llama3.1-8B: 30
+    deepseek-ai/DeepSeek-R1-Distill-Qwen-14B: 5
+    hugging-quants/Meta-Llama-3.1-70B-Instruct-AWQ-INT4: 5
+    openai/gpt-oss-20b: 50
+    google/gemma-3-27b-it: 50
+    default: 50

nilai-api/src/nilai_api/config/config-e176.yaml

@@ -0,0 +1,35 @@
+# In production, this file is automatically generated by the `ansible` playbook.
+# Configuration with structured sections and default values
+
+
+# Web Search Configuration
+web_search:
+  api_key: null
+  api_path: "https://api.search.brave.com/res/v1/web/search"
+  count: 3
+  lang: "en"
+  country: "us"
+  timeout: 20.0
+  max_concurrent_requests: 20
+  rps: 20
+
+# Rate Limiting Configuration
+rate_limiting:
+  user_rate_limit: null # For-good rate limit
+  user_rate_limit_minute: 100
+  user_rate_limit_hour: 1000
+  user_rate_limit_day: 10000
+  web_search_rate_limit_minute: null
+  web_search_rate_limit_hour: null
+  web_search_rate_limit_day: 500
+  web_search_rate_limit: null # For-good rate limit
+  model_concurrent_rate_limit:
+    meta-llama/Llama-3.2-1B-Instruct: 45
+    meta-llama/Llama-3.2-3B-Instruct: 50
+    meta-llama/Llama-3.1-8B-Instruct: 30
+    cognitivecomputations/Dolphin3.0-Llama3.1-8B: 30
+    deepseek-ai/DeepSeek-R1-Distill-Qwen-14B: 5
+    hugging-quants/Meta-Llama-3.1-70B-Instruct-AWQ-INT4: 5
+    openai/gpt-oss-20b: 50
+    google/gemma-3-27b-it: 50
+    default: 50

nilai-api/src/nilai_api/config/config-f910.yaml

@@ -0,0 +1,35 @@
+# In production, this file is automatically generated by the `ansible` playbook.
+# Configuration with structured sections and default values
+
+
+# Web Search Configuration
+web_search:
+  api_key: null
+  api_path: "https://api.search.brave.com/res/v1/web/search"
+  count: 3
+  lang: "en"
+  country: "us"
+  timeout: 20.0
+  max_concurrent_requests: 20
+  rps: 20
+
+# Rate Limiting Configuration
+rate_limiting:
+  user_rate_limit: null # For-good rate limit
+  user_rate_limit_minute: 100
+  user_rate_limit_hour: 1000
+  user_rate_limit_day: 10000
+  web_search_rate_limit_minute: 1
+  web_search_rate_limit_hour: 3
+  web_search_rate_limit_day: 72
+  web_search_rate_limit: null # For-good rate limit
+  model_concurrent_rate_limit:
+    meta-llama/Llama-3.2-1B-Instruct: 45
+    meta-llama/Llama-3.2-3B-Instruct: 50
+    meta-llama/Llama-3.1-8B-Instruct: 30
+    cognitivecomputations/Dolphin3.0-Llama3.1-8B: 30
+    deepseek-ai/DeepSeek-R1-Distill-Qwen-14B: 5
+    hugging-quants/Meta-Llama-3.1-70B-Instruct-AWQ-INT4: 5
+    openai/gpt-oss-20b: 50
+    google/gemma-3-27b-it: 50
+    default: 50

scripts/docker-composer.py

@@ -18,6 +18,29 @@
 import sys
 import shutil
 import yaml
+from dotenv import load_dotenv
+
+
+def load_environment_variables(is_dev):
+    """
+    Load environment variables from .env.mainnet file.
+
+    Returns:
+        dict: Environment variables dictionary with .env.mainnet loaded
+    """
+    env = os.environ.copy()
+
+    # Load .env.mainnet file if it exists
+    env_file = ".env.mainnet" if not is_dev else ".env.dev"
+    if os.path.isfile(env_file):
+        print(f"Loading environment variables from {env_file}")
+        load_dotenv(env_file, override=True)
+        # Update the env dict with the loaded variables
+        env.update(os.environ)
+    else:
+        print(f"Warning: {env_file} not found, using system environment variables only")
+
+    return env
 
 
 def find_docker_compose_command():
@@ -369,8 +392,8 @@ def main():
     if use_no_path_resolution:
         config_cmd.append("--no-path-resolution")
 
-    # Set FILES environment variable to a valid placeholder path that we'll replace later
-    env = os.environ.copy()
+    # Load environment variables from .env.mainnet and set FILES environment variable
+    env = load_environment_variables(args.dev)
     files_placeholder = "/tmp/files_placeholder"
     env["FILES"] = files_placeholder