Basic Streaming and In-Memory Guard Support

.github/workflows/pr_qa.yml

@@ -15,27 +15,18 @@ jobs:
     steps:
       - uses: actions/checkout@v2
       - uses: actions/setup-python@v4
-      - name: Quality Checks
+      - name: Install Dependencies
         run: |
-          python -m pip install --upgrade pip;
+          python -m venv ./.venv
+          source ./.venv/bin/activate
           make install-lock;
-          opentelemetry-bootstrap -a install;
-          make qa;
-
-          ##
-          ## NOTE:
-          ## The below checks for a file named .version-change-type that is used to automatically increment the version number during publish flows.
-          ## It also checks for changes in a file called RELEASENOTES.md that are used to auto-maintain a rolling CHANGELOG.
-          ## This is commented out for now until we finalize what the publish/release flow will look like.
-          ##
+          make install-dev;
 
-          # changeType=$(<.version-change-type)
-          # if [ -z "$changeType" ];
-          #   then
-          #     echo "missing file .version-change-type!"
-          #     exit 1
-          # fi
-          # echo "Checking for release notes..."
-          # git fetch origin main ${{ github.event.pull_request.base.sha }};
-          # diff=$(git diff -U0 ${{ github.event.pull_request.base.sha }} ${{ github.sha }} RELEASENOTES.md);
-          # if [ -z "$diff" ]; then echo "Missing release notes! exiting..."; exit 1; fi
+          curl https://raw.githubusercontent.com/guardrails-ai/guardrails-api-client/main/service-specs/guardrails-service-spec.yml -o ./open-api-spec.yml
+          npx @redocly/cli bundle --dereferenced --output ./open-api-spec.json --ext json ./open-api-spec.yml
+
+      - name: Run Quality Checks
+        run: |
+          source ./.venv/bin/activate
+
+          make qa;

.gitignore

@@ -23,4 +23,5 @@ opentelemetry-lambda-layer
 open-api-spec.json
 open-api-spec.yml
 .python-version
-requirements-lock-old.txt
+requirements-lock-old.txt
+models

Dockerfile.dev

@@ -48,4 +48,4 @@ COPY . .
 
 EXPOSE 8000
 
-CMD opentelemetry-instrument gunicorn --bind 0.0.0.0:8000 --timeout=90 --threads=10 "app:create_app()"
+CMD gunicorn --bind 0.0.0.0:8000 --timeout=90 --threads=10 "app:create_app()"

Dockerfile.heavy

@@ -44,5 +44,5 @@ COPY . .
 
 EXPOSE 8000
 
-CMD opentelemetry-instrument gunicorn --bind 0.0.0.0:8000 --timeout=90 --threads=10 "app:create_app()"
+CMD gunicorn --bind 0.0.0.0:8000 --timeout=90 --threads=10 "app:create_app()"
 # CMD gunicorn --forwarded-allow-ips="*" --bind 0.0.0.0:8000 --timeout=60 --threads=10 "app:create_app()"

Dockerfile.lite

@@ -0,0 +1,55 @@
+FROM public.ecr.aws/docker/library/python:3.11.6-slim
+
+ARG GITHUB_TOKEN
+ARG HF_TOKEN
+
+COPY .guardrailsrc /root/.guardrailsrc
+
+# COPY --from=public.ecr.aws/awsguru/aws-lambda-adapter:0.7.1 /lambda-adapter /opt/extensions/lambda-adapter
+# COPY ./opentelemetry-lambda-layer /opt
+
+# Create app directory
+WORKDIR /app
+
+# check the version
+RUN python3 --version
+# start the virtual environment
+RUN python3 -m venv /opt/venv
+
+# Enable venv
+ENV PATH="/opt/venv/bin:$PATH"
+
+# Install git and curl
+RUN apt-get update
+RUN apt-get install -y git curl gcc jq
+
+# Copy the requirements file
+COPY requirements*.txt .
+
+RUN curl https://truststore.pki.rds.amazonaws.com/global/global-bundle.pem -o ./global-bundle.pem
+
+# Install app dependencies
+RUN pip install -r requirements-lock.txt
+
+# Download punkt data
+RUN python -m nltk.downloader -d /opt/nltk_data punkt
+
+# RUN guardrails hub install hub://guardrails/profanity_free
+RUN guardrails hub install hub://guardrails/valid_length
+RUN guardrails hub install hub://guardrails/lowercase
+RUN guardrails hub install hub://guardrails/regex_match
+
+COPY ./custom-install ./custom-install
+
+RUN python ./custom-install/install.py
+
+# Freeze dependencies
+RUN pip freeze > requirements-lock.txt
+
+# Copy the whole folder inside the Image filesystem
+COPY . .
+
+EXPOSE 8000
+
+CMD gunicorn --bind 0.0.0.0:8000 --timeout=90 --threads=10 --limit-request-line=0 --limit-request-fields=1000 --limit-request-field_size=0 "app:create_app()"
+# CMD gunicorn --forwarded-allow-ips="*" --bind 0.0.0.0:8000 --timeout=60 --threads=10 "app:create_app()"

Dockerfile.prod

@@ -37,5 +37,5 @@ COPY . .
 
 EXPOSE 8000
 
-CMD opentelemetry-instrument gunicorn --bind 0.0.0.0:8000 --timeout=5 --threads=10 "app:create_app()"
+CMD gunicorn --bind 0.0.0.0:8000 --timeout=5 --threads=10 "app:create_app()"
 # CMD gunicorn --forwarded-allow-ips="*" --bind 0.0.0.0:8000 --timeout=60 --threads=10 "app:create_app()"

Makefile

@@ -1,7 +1,11 @@
 # Installs production dependencies
 install:
 	pip install -r requirements.txt;
+	# This is a workaround because of this issue: https://github.com/open-telemetry/opentelemetry-python-contrib/issues/2053
+	pip uninstall aiohttp -y
+	pip install opentelemetry-distro
 	opentelemetry-bootstrap -a install
+	pip install aiohttp
 
 # Installs development dependencies
 install-dev:
@@ -53,7 +57,7 @@ source:
 	source ./.venv/bin/activate
 
 test:
-	python3 -m pytest ./tests
+	pytest ./tests
 
 test-cov:
 	coverage run --source=./src -m pytest ./tests

app.py

@@ -4,7 +4,9 @@
 from werkzeug.middleware.proxy_fix import ProxyFix
 from urllib.parse import urlparse
 from guardrails import configure_logging
-# from opentelemetry.instrumentation.flask import FlaskInstrumentor
+from opentelemetry.instrumentation.flask import FlaskInstrumentor
+from src.clients.postgres_client import postgres_is_enabled
+from src.otel import otel_is_enabled, initialize
 
 
 class ReverseProxied(object):
@@ -19,6 +21,11 @@ def __call__(self, environ, start_response):
 
 
 def create_app():
+    if os.environ.get("APP_ENVIRONMENT") != "production":
+        from dotenv import load_dotenv
+
+        load_dotenv()
+
     app = Flask(__name__)
 
     app.config["APPLICATION_ROOT"] = "/"
@@ -31,12 +38,16 @@ def create_app():
     guardrails_log_level = os.environ.get("GUARDRAILS_LOG_LEVEL", "INFO")
     configure_logging(log_level=guardrails_log_level)
 
-    # FlaskInstrumentor().instrument_app(app)
+    if otel_is_enabled():
+        FlaskInstrumentor().instrument_app(app)
+        initialize()
 
-    from src.clients.postgres_client import PostgresClient
+    # if no pg_host is set, don't set up postgres
+    if postgres_is_enabled():
+        from src.clients.postgres_client import PostgresClient
 
-    pg_client = PostgresClient()
-    pg_client.initialize(app)
+        pg_client = PostgresClient()
+        pg_client.initialize(app)
 
     from src.blueprints.root import root_bp
     from src.blueprints.guards import guards_bp

config.py

@@ -0,0 +1,11 @@
+'''
+All guards defined here will be initialized, if and only if
+the application is using in memory guards.
+
+The application will use in memory guards if pg_host is left
+undefined. Otherwise, a postgres instance will be started 
+and guards will be persisted into postgres. In that case,
+these guards will not be initialized.
+'''
+
+from guardrails import Guard

custom-install/install.py

@@ -0,0 +1,113 @@
+import os
+import sys
+import logging
+import json
+from typing import Any, Dict
+from rich.console import Console
+from guardrails.cli.hub.install import (
+    get_site_packages_location,
+    install_hub_module,
+    run_post_install,
+    add_to_hub_inits
+)
+from guardrails.cli.server.module_manifest import ModuleManifest
+from string import Template
+
+console = Console()
+
+os.environ[
+    "COLOREDLOGS_LEVEL_STYLES"
+] = "spam=white,faint;success=green,bold;debug=magenta;verbose=blue;notice=cyan,bold;warning=yellow;error=red;critical=background=red"  # noqa
+LEVELS = {
+    "SPAM": 5,
+    "VERBOSE": 15,
+    "NOTICE": 25,
+    "SUCCESS": 35,
+}
+for key in LEVELS:
+    logging.addLevelName(LEVELS.get(key), key)  # type: ignore
+logger = logging.getLogger("custom-install")
+
+
+def load_manifest(fileName: str) -> Dict[str, Any]:
+    with open(f"custom-install/manifests/{fileName}") as manifest_file:
+        content = manifest_file.read()
+        return json.loads(content)
+
+custom_manifests = {
+    "guardrails/provenance_llm": load_manifest("provenance-llm.json"),
+    "guardrails/detect_pii": load_manifest("detect-pii.json"),
+    "guardrails/competitor_check": load_manifest("competitor-check.json"),
+   "guardrails/many_shot_jailbreak": load_manifest("jailbreak.json"),
+   "tryolabs/restricttotopic": load_manifest("restrict-to-topic.json"),
+}
+
+def get_validator_manifest(module_name) -> ModuleManifest:
+    manifest = custom_manifests.get(module_name, {})
+    return ModuleManifest.from_dict(manifest)
+
+def custom_install(package_uri: str):
+    """Install a validator from the Hub."""
+    if not package_uri.startswith("hub://"):
+        logger.error("Invalid URI!")
+        sys.exit(1)
+
+    console.print(f"\nInstalling {package_uri}...\n")
+    logger.log(
+        level=LEVELS.get("SPAM"), msg=f"Installing {package_uri}..."  # type: ignore
+    )
+
+    # Validation
+    module_name = package_uri.replace("hub://", "")
+
+    # Prep
+    with console.status("Fetching manifest", spinner="bouncingBar"):
+        module_manifest = get_validator_manifest(module_name)
+        site_packages = get_site_packages_location()
+
+    # Install
+    with console.status("Downloading dependencies", spinner="bouncingBar"):
+        install_hub_module(module_manifest, site_packages)
+
+    # Post-install
+    with console.status("Running post-install setup", spinner="bouncingBar"):
+        run_post_install(module_manifest, site_packages)
+        add_to_hub_inits(module_manifest, site_packages)
+
+    success_message_cli = Template(
+        """✅Successfully installed ${module_name}!
+
+[bold]Import validator:[/bold]
+from guardrails.hub import ${export}
+
+[bold]Get more info:[/bold]
+https://hub.guardrailsai.com/validator/${id}
+"""
+    ).safe_substitute(
+        module_name=package_uri,
+        id=module_manifest.id,
+        export=module_manifest.exports[0],
+    )
+    success_message_logger = Template(
+        """✅Successfully installed ${module_name}!
+
+Import validator:
+from guardrails.hub import ${export}
+
+Get more info:
+https://hub.guardrailsai.com/validator/${id}
+"""
+    ).safe_substitute(
+        module_name=package_uri,
+        id=module_manifest.id,
+        export=module_manifest.exports[0],
+    )
+    console.print(success_message_cli)  # type: ignore
+    logger.log(level=LEVELS.get("SPAM"), msg=success_message_logger)  # type: ignore
+
+
+# custom_install("hub://guardrails/provenance_llm")
+custom_install("hub://tryolabs/restricttotopic")
+# custom_install("hub://guardrails/detect_pii")
+# custom_install("hub://guardrails/competitor_check")
+# custom_install("hub://guardrails/many_shot_jailbreak")

custom-install/manifests/competitor-check.json

@@ -0,0 +1,46 @@
+{
+  "name": "Competitor Check",
+  "author": {
+      "name": "Guardrails AI",
+      "email": "contact@guardrailsai.com"
+  },
+  "maintainers": [{
+      "name": "Karan Acharya",
+      "email": "karan@guardrailsai.com"
+  }],
+  "repository": {
+      "url": "https://github.com/guardrails-ai/competitor_check.git",
+      "branch": "frontend_demo"
+  },
+  "index": "./__init__.py",
+  "exports": [
+      "CompetitorCheck"
+  ],
+  "tags": {
+      "language": [
+          "en"
+      ],
+      "certification": [
+          "Guardrails Certified"
+      ],
+      "contentType": [
+          "string"
+      ],
+      "infrastructureRequirements": [
+          "ML"
+      ],
+      "riskCategory": [
+          "Brand risk"
+      ],
+      "useCases": [
+          "Chatbots",
+          "Customer Support"
+      ]
+  },
+  "id": "guardrails/competitor_check",
+  "namespace": "guardrails",
+  "packageName": "competitor_check",
+  "moduleName": "validator",
+  "requiresAuth": false,
+  "postInstall": "post-install.py"
+}