feat(sandbox): adding docker sandbox

examples/docker_sandbox.py

@@ -0,0 +1,31 @@
+# Intall pandasai docker using
+# pip install pandasai-docker
+from pandasai_docker import DockerSandbox
+from pandasai_openai.openai import OpenAI
+
+import pandasai as pai
+
+llm = OpenAI(api_token="sk-*******")
+
+pai.config.set({"llm": llm})
+
+df = pai.read_csv("/Users/arslan/Documents/SinapTik/pandas-ai/artists.csv")
+
+
+sandbox = DockerSandbox()
+agent = Agent([df], memory_size=10, sandbox=sandbox)
+# Chat with the Agent
+response = agent.chat("plot top five artists streams")
+
+# destroy container after usage or let class destructor destroys it.
+sanbox.stop()
+
+
+# Use custom docker image
+sandbox = DockerSandbox("pandaai-sandbox", "/path/to/Dockerfile")
+agent = Agent([df], memory_size=10, sandbox=sandbox)
+# Chat with the Agent
+response = agent.chat("plot top five artists streams")
+
+# destroy container after usage or let class destructor destroys it.
+sandbox.stop()

extensions/sandbox/docker/README.md

@@ -0,0 +1,9 @@
+# Docker Sandbox Extension for PandasAI
+
+## Installation
+
+You can install this extension using poetry:
+
+```bash
+poetry install pandasai-docker
+```

extensions/sandbox/docker/pandasai_docker/Dockerfile

@@ -0,0 +1,12 @@
+FROM python:3.9
+
+LABEL image_name="pandaai-sandbox"
+
+# Install required Python packages
+RUN pip install pandas numpy matplotlib
+
+# Set the working directory inside the container
+WORKDIR /app
+
+# Default command keeps the container running (useful for testing or debugging)
+CMD ["sleep", "infinity"]

extensions/sandbox/docker/pandasai_docker/__init__.py

@@ -0,0 +1,3 @@
+from .docker_sandbox import DockerSandbox
+
+__all__ = ["DockerSandbox"]

extensions/sandbox/docker/pandasai_docker/docker_sandbox.py

@@ -0,0 +1,188 @@
+import io
+import logging
+import os
+import re
+import tarfile
+import uuid
+from typing import Optional
+
+import docker
+
+from pandasai.sandbox import Sandbox
+
+from .serializer import ResponseSerializer
+
+logger = logging.getLogger(__name__)
+
+
+class DockerSandbox(Sandbox):
+    def __init__(self, image_name="pandaai-sandbox", dockerfile_path=None):
+        super().__init__()
+        self._dockerfile_path: str = dockerfile_path or os.path.join(
+            os.path.dirname(__file__), "Dockerfile"
+        )
+        self._image_name: str = image_name
+        self._client: docker.DockerClient = docker.from_env()
+        self._container: Optional[docker.models.containers.Container] = None
+
+        # Build the image if it does not exist
+        if not self._image_exists():
+            self._build_image()
+
+        self._helper_code: str = self._read_start_code(
+            os.path.join(os.path.dirname(__file__), "serializer.py")
+        )
+
+    def _image_exists(self) -> bool:
+        try:
+            self._client.images.get(self._image_name)
+            return True
+        except docker.errors.ImageNotFound:
+            return False
+
+    def _build_image(self) -> None:
+        logger.info(
+            f"Building Docker image '{self._image_name}' from '{self._dockerfile_path}'..."
+        )
+        with open(self._dockerfile_path, "rb") as file:
+            self._client.images.build(fileobj=file, tag=self._image_name)
+
+    def start(self):
+        if not self._started:
+            logger.info(
+                f"Starting a Docker container from the image '{self._image_name}'"
+            )
+            self._container = self._client.containers.run(
+                self._image_name, command="sleep infinity", detach=True, tty=True
+            )
+            logger.info(
+                f"Started a Docker container with id '{self._container.id}' from the image '{self._image_name}'"
+            )
+            self._started = True
+
+    def stop(self) -> None:
+        if self._started and self._container:
+            logger.info(f"Stopping a Docker container with id '{self._container.id}''")
+            self._container.stop()
+            self._container.remove()
+            self._container = None
+            self._started = False
+
+    def _read_start_code(self, file_path: str) -> str:
+        """Read helper start code from a file as a string.
+
+        Args:
+            file_path (str): Path to the file.
+
+        Returns:
+            str: Code as a string.
+        """
+        with open(file_path, "r") as file:
+            return file.read()
+
+    def _exec_code(self, code: str, enviroment: dict) -> dict:
+        """Execute Python code in a Docker container.
+
+        Args:
+            code (str): Code to execute.
+            enviroment (dict): Environment variables to pass to the container.
+
+        Returns:
+            dict: Result of the code execution.
+        """
+        if not self._container:
+            raise RuntimeError("Container is not running.")
+
+        sql_queries = self._extract_sql_queries_from_code(code)
+
+        # Temporary chart storage path
+        chart_path = "/tmp/temp_chart.png"
+        # actual chart path
+        original_chart_path = None
+
+        if png_paths := re.findall(r"'([^']+\.png)'", code):
+            original_chart_path = png_paths[0]
+
+        # update chart path
+        code = re.sub(
+            r"""(['"])([^'"]*\.png)\1""",
+            lambda m: f"{m.group(1)}{chart_path}{m.group(1)}",
+            code,
+        )
+
+        # Execute SQL queries, save the query results to CSV files
+        datasets_map = {}
+        for sql_query in sql_queries:
+            execute_sql_query_func = enviroment.get("execute_sql_query")
+            if execute_sql_query_func is None:
+                raise RuntimeError(
+                    "execute_sql_query function is not defined in the environment."
+                )
+
+            query_df = execute_sql_query_func(sql_query)
+            filename = f"{uuid.uuid4().hex}.csv"
+            # Pass the files to the container for further processing
+            self.pass_csv(query_df, filename=filename)
+            datasets_map[sql_query] = filename
+
+        # Add the datasets_map variable to the code
+        dataset_map = f"""
+datasets_map = {datasets_map}
+
+def execute_sql_query(sql_query):
+    filename = datasets_map[sql_query]
+    filepath = os.path.join("/tmp", filename)
+    return pd.read_csv(filepath)
+
+"""
+        # serialization code to get output from docker
+        end_code = """
+print(parser.serialize(result))
+"""
+        # Concatenate code and helper code
+        code = self._helper_code + dataset_map + code + end_code
+
+        # Compile the code for errors
+        self._compile_code(code)
+
+        # Replace double quotes with escaped double quotes for command line code arguments
+        code = code.replace('"', '\\"')
+
+        logger.info(f"Submitting code to docker container {code}")
+
+        exit_code, output = self._container.exec_run(
+            cmd=f'python -c "{code}"', demux=True
+        )
+
+        if exit_code != 0:
+            raise RuntimeError(f"Error executing code: {output[1].decode()}")
+
+        response = output[0].decode()
+        return ResponseSerializer.deserialize(response, original_chart_path)
+
+    def pass_csv(self, csv_data, filename="file.csv") -> None:
+        if not self._container:
+            raise RuntimeError("Container is not running.")
+
+        # Convert the DataFrame to a CSV string
+        csv_string = csv_data.to_csv(index=False)
+
+        # Create a tar archive in memory
+        tar_stream = io.BytesIO()
+        with tarfile.open(fileobj=tar_stream, mode="w") as tar:
+            # Add the CSV string as a file in the tar archive
+            csv_bytes = csv_string.encode("utf-8")
+            tarinfo = tarfile.TarInfo(name=filename)
+            tarinfo.size = len(csv_bytes)
+            tar.addfile(tarinfo, io.BytesIO(csv_bytes))
+
+        # Seek to the beginning of the stream
+        tar_stream.seek(0)
+
+        # Transfer the tar archive to the container
+        self._container.put_archive("/tmp", tar_stream)
+
+    def __del__(self) -> None:
+        if self._container:
+            self._container.stop()
+            self._container.remove()

extensions/sandbox/docker/pandasai_docker/serializer.py

@@ -0,0 +1,73 @@
+import base64
+import datetime
+import json
+import os  # important to import
+import tarfile  # important to import
+from json import JSONEncoder
+
+import numpy as np
+import pandas as pd
+
+
+class ResponseSerializer:
+    @staticmethod
+    def serialize_dataframe(df: pd.DataFrame) -> dict:
+        if df.empty:
+            return {"columns": [], "data": [], "index": []}
+        return df.to_dict(orient="split")
+
+    @staticmethod
+    def serialize(result: dict) -> str:
+        if result["type"] == "dataframe":
+            if isinstance(result["value"], pd.Series):
+                result["value"] = result["value"].to_frame()
+            result["value"] = ResponseSerializer.serialize_dataframe(result["value"])
+
+        elif result["type"] == "plot" and isinstance(result["value"], str):
+            with open(result["value"], "rb") as image_file:
+                image_data = image_file.read()
+            result["value"] = base64.b64encode(image_data).decode()
+
+        return json.dumps(result, cls=CustomEncoder)
+
+    @staticmethod
+    def deserialize(response: str, chart_path: str = None) -> dict:
+        result = json.loads(response)
+        if result["type"] == "dataframe":
+            json_data = result["value"]
+            result["value"] = pd.DataFrame(
+                data=json_data["data"],
+                index=json_data["index"],
+                columns=json_data["columns"],
+            )
+
+        elif result["type"] == "plot" and chart_path:
+            image_data = base64.b64decode(result["value"])
+
+            # Write the binary data to a file
+            with open(chart_path, "wb") as image_file:
+                image_file.write(image_data)
+
+            result["value"] = chart_path
+
+        return result
+
+
+class CustomEncoder(JSONEncoder):
+    def default(self, obj):
+        if isinstance(obj, (np.integer, np.int64)):
+            return int(obj)
+
+        if isinstance(obj, (np.floating, np.float64)):
+            return float(obj)
+
+        if isinstance(obj, (pd.Timestamp, datetime.datetime, datetime.date)):
+            return obj.isoformat()
+
+        if isinstance(obj, pd.DataFrame):
+            return ResponseSerializer.serialize_dataframe(obj)
+
+        return super().default(obj)
+
+
+parser = ResponseSerializer()