kubiya-control-plane-api 0.1.0__py3-none-any.whl → 0.3.4__py3-none-any.whl

control_plane_api/app/lib/job_executor.py

@@ -0,0 +1,312 @@
+"""
+Job execution logic for routing and parameter substitution.
+
+This module handles:
+- Dynamic executor routing (auto/specific queue/environment)
+- Prompt template parameter substitution
+- Worker queue selection based on availability
+"""
+
+import re
+import structlog
+from typing import Dict, Any, Optional, Tuple
+
+logger = structlog.get_logger()
+
+
+def substitute_prompt_parameters(prompt_template: str, parameters: Dict[str, Any]) -> str:
+    """
+    Substitute parameters in prompt template.
+
+    Template variables use {{variable_name}} syntax.
+
+    Example:
+        prompt_template = "Run a backup of {{database}} at {{time}}"
+        parameters = {"database": "production", "time": "5pm"}
+        result = "Run a backup of production at 5pm"
+
+    Args:
+        prompt_template: Prompt template with {{variables}}
+        parameters: Dictionary of parameter values
+
+    Returns:
+        Prompt with substituted values
+
+    Raises:
+        ValueError: If required parameters are missing
+    """
+    # Find all variables in template
+    variables = re.findall(r"\{\{(\w+)\}\}", prompt_template)
+
+    # Check for missing parameters
+    missing = [var for var in variables if var not in parameters]
+    if missing:
+        raise ValueError(f"Missing required parameters: {', '.join(missing)}")
+
+    # Substitute variables
+    result = prompt_template
+    for var_name, var_value in parameters.items():
+        result = result.replace(f"{{{{{var_name}}}}}", str(var_value))
+
+    return result
+
+
+async def get_available_worker_queues(
+    organization_id: str,
+    environment_name: Optional[str] = None
+) -> list[Dict[str, Any]]:
+    """
+    Get list of worker queues with active workers.
+
+    Queries the worker_queues table and counts active workers from Redis heartbeats.
+
+    Args:
+        organization_id: Organization ID for multi-tenant filtering
+        environment_name: Optional environment name filter
+
+    Returns:
+        List of worker queue dictionaries with metadata
+    """
+    from control_plane_api.app.lib.supabase import get_supabase
+    from control_plane_api.app.routers.worker_queues import get_active_workers_from_redis
+
+    client = get_supabase()
+
+    try:
+        # Query worker_queues table for active queues
+        query = (
+            client.table("worker_queues")
+            .select("id, name, environment_id, status, environments(name)")
+            .eq("organization_id", organization_id)
+            .eq("status", "active")
+        )
+
+        # Filter by environment if specified
+        if environment_name:
+            # First get the environment ID
+            env_result = (
+                client.table("environments")
+                .select("id")
+                .eq("organization_id", organization_id)
+                .eq("name", environment_name)
+                .maybe_single()
+                .execute()
+            )
+            if env_result.data:
+                query = query.eq("environment_id", env_result.data["id"])
+            else:
+                logger.warning(
+                    "environment_not_found",
+                    organization_id=organization_id,
+                    environment_name=environment_name
+                )
+                return []
+
+        result = query.execute()
+
+        if not result.data:
+            logger.info("no_active_worker_queues_found", organization_id=organization_id)
+            return []
+
+        # Get active workers from Redis heartbeats
+        active_workers_data = await get_active_workers_from_redis(organization_id)
+
+        logger.info(
+            "checking_active_workers",
+            organization_id=organization_id,
+            total_queues_in_db=len(result.data),
+            active_workers_count=len(active_workers_data)
+        )
+
+        # Count workers per queue
+        worker_counts = {}
+        for worker_id, worker_data in active_workers_data.items():
+            queue_id = worker_data.get("worker_queue_id")
+            if queue_id:
+                worker_counts[queue_id] = worker_counts.get(queue_id, 0) + 1
+
+        # Transform to expected format
+        worker_queues = []
+        for queue in result.data:
+            active_worker_count = worker_counts.get(queue["id"], 0)
+
+            logger.debug(
+                "checking_queue_for_active_workers",
+                queue_id=queue["id"],
+                queue_name=queue.get("name"),
+                active_worker_count=active_worker_count
+            )
+
+            # Only include queues with active workers
+            if active_worker_count > 0:
+                env_data = queue.get("environments", {})
+                worker_queues.append({
+                    "queue_name": queue["id"],  # Use queue ID as the task queue name
+                    "environment_name": env_data.get("name") if env_data else None,
+                    "active_workers": active_worker_count,
+                    "idle_workers": 0,  # Not tracked separately
+                    "total_workers": active_worker_count,
+                })
+
+        logger.info(
+            "found_available_worker_queues",
+            organization_id=organization_id,
+            count=len(worker_queues),
+            worker_counts=worker_counts
+        )
+
+        return worker_queues
+
+    except Exception as e:
+        logger.error("failed_to_get_available_worker_queues", error=str(e), organization_id=organization_id)
+        return []
+
+
+async def select_worker_queue(
+    organization_id: str,
+    executor_type: str,
+    worker_queue_name: Optional[str] = None,
+    environment_name: Optional[str] = None,
+) -> Tuple[Optional[str], Optional[str]]:
+    """
+    Select appropriate worker queue for job execution.
+
+    Routing logic:
+    - AUTO: Select first available queue with idle workers (prefer idle over active)
+    - SPECIFIC_QUEUE: Use provided worker_queue_name
+    - ENVIRONMENT: Select first available queue in specified environment
+
+    Args:
+        organization_id: Organization ID
+        executor_type: Routing type ("auto", "specific_queue", "environment")
+        worker_queue_name: Explicit queue name (for SPECIFIC_QUEUE)
+        environment_name: Environment name (for ENVIRONMENT routing)
+
+    Returns:
+        Tuple of (worker_queue_name, environment_name) or (None, None) if no workers available
+    """
+    if executor_type == "specific_queue":
+        if not worker_queue_name:
+            raise ValueError("worker_queue_name is required for 'specific_queue' executor type")
+        return worker_queue_name, environment_name
+
+    # AUTO or ENVIRONMENT routing - need to find available workers
+    available_queues = await get_available_worker_queues(organization_id, environment_name)
+
+    if not available_queues:
+        logger.warning(
+            "no_available_worker_queues",
+            organization_id=organization_id,
+            executor_type=executor_type,
+            environment_name=environment_name,
+        )
+        return None, None
+
+    # Sort by idle workers first, then by total workers
+    # This ensures we prefer queues with capacity
+    available_queues.sort(
+        key=lambda q: (q["idle_workers"], q["total_workers"]),
+        reverse=True
+    )
+
+    selected = available_queues[0]
+    logger.info(
+        "selected_worker_queue",
+        organization_id=organization_id,
+        queue_name=selected["queue_name"],
+        environment_name=selected["environment_name"],
+        idle_workers=selected["idle_workers"],
+        total_workers=selected["total_workers"],
+    )
+
+    return selected["queue_name"], selected["environment_name"]
+
+
+async def resolve_job_entity(
+    supabase_client,
+    organization_id: str,
+    planning_mode: str,
+    entity_type: Optional[str],
+    entity_id: Optional[str],
+) -> Tuple[str, str, str]:
+    """
+    Resolve job entity (agent/team/workflow) and return execution details.
+
+    For predefined modes, validates that the entity exists and returns its details.
+    For on_the_fly mode, returns None values (planner will determine execution).
+
+    Args:
+        supabase_client: Supabase client instance
+        organization_id: Organization ID
+        planning_mode: Planning mode (on_the_fly, predefined_agent, etc.)
+        entity_type: Entity type (agent/team/workflow)
+        entity_id: Entity ID
+
+    Returns:
+        Tuple of (execution_type, entity_id, entity_name)
+
+    Raises:
+        ValueError: If entity doesn't exist or validation fails
+    """
+    if planning_mode == "on_the_fly":
+        # Planner will determine execution
+        return "agent", None, None
+
+    # Validate entity exists
+    if not entity_type or not entity_id:
+        raise ValueError(f"entity_type and entity_id are required for planning_mode '{planning_mode}'")
+
+    table_name = f"{entity_type}s"  # agents, teams, workflows
+
+    try:
+        result = (
+            supabase_client.table(table_name)
+            .select("id, name")
+            .eq("id", entity_id)
+            .eq("organization_id", organization_id)
+            .execute()
+        )
+
+        if not result.data:
+            raise ValueError(f"{entity_type} with ID {entity_id} not found")
+
+        entity = result.data[0]
+        return entity_type, entity["id"], entity["name"]
+
+    except Exception as e:
+        logger.error(
+            "failed_to_resolve_job_entity",
+            error=str(e),
+            planning_mode=planning_mode,
+            entity_type=entity_type,
+            entity_id=entity_id,
+        )
+        raise ValueError(f"Failed to resolve {entity_type}: {str(e)}")
+
+
+def merge_execution_config(
+    base_config: Dict[str, Any],
+    override_config: Optional[Dict[str, Any]] = None
+) -> Dict[str, Any]:
+    """
+    Merge base job config with execution-specific overrides.
+
+    Args:
+        base_config: Base configuration from job definition
+        override_config: Optional overrides for this execution
+
+    Returns:
+        Merged configuration dictionary
+    """
+    if not override_config:
+        return base_config.copy()
+
+    # Deep merge
+    merged = base_config.copy()
+    for key, value in override_config.items():
+        if key in merged and isinstance(merged[key], dict) and isinstance(value, dict):
+            merged[key] = {**merged[key], **value}
+        else:
+            merged[key] = value
+
+    return merged

control_plane_api/app/lib/kubiya_client.py

@@ -0,0 +1,235 @@
+"""Kubiya API client for authentication and runner management"""
+
+import httpx
+import os
+from typing import Optional, Dict, List
+import structlog
+
+logger = structlog.get_logger()
+
+KUBIYA_API_BASE = os.environ.get("KUBIYA_API_BASE", "https://api.kubiya.ai")
+
+
+class KubiyaClient:
+    """Client for Kubiya API"""
+
+    def __init__(self, api_base: str = KUBIYA_API_BASE):
+        self.api_base = api_base.rstrip("/")
+        self.client = httpx.AsyncClient(timeout=30.0)
+
+    async def validate_token_and_get_org(self, token: str) -> Optional[Dict]:
+        """
+        Validate token with Kubiya API and get organization details.
+        Automatically tries both Bearer (Auth0 idToken) and UserKey (API key) authentication.
+
+        Args:
+            token: Authentication token (Bearer/idToken or UserKey/API key)
+
+        Returns:
+            Dict with organization details:
+            {
+                "id": "org-uuid",
+                "name": "Organization Name",
+                "slug": "org-slug"
+            }
+            None if invalid token
+        """
+        try:
+            # Try Bearer authentication first (Auth0 idToken)
+            response = await self.client.get(
+                f"{self.api_base}/api/v1/users/me",
+                headers={"Authorization": f"Bearer {token}"},
+            )
+
+            # If Bearer fails with 401, try UserKey (API key)
+            if response.status_code == 401:
+                logger.debug("kubiya_bearer_auth_failed_trying_userkey")
+                response = await self.client.get(
+                    f"{self.api_base}/api/v1/users/me",
+                    headers={"Authorization": f"UserKey {token}"},
+                )
+
+            if response.status_code == 200:
+                data = response.json()
+
+                # Log full response for debugging
+                logger.info(
+                    "kubiya_api_response",
+                    response_keys=list(data.keys()),
+                    has_org=bool(data.get("org")),
+                    has_org_id=bool(data.get("org_id")),
+                )
+
+                # Extract organization from response
+                # Kubiya API returns org/org_id at root level, not nested
+                org_id = data.get("org") or data.get("org_id") or data.get("organization", {}).get("uuid")
+                org_name = data.get("org_name") or data.get("organization_name") or data.get("organization", {}).get("name")
+                org_slug = data.get("org_slug") or data.get("organization_slug") or data.get("organization", {}).get("slug")
+
+                org_data = {
+                    "id": org_id,
+                    "name": org_name,
+                    "slug": org_slug,
+                    "user_id": data.get("uuid") or data.get("id"),
+                    "user_email": data.get("email"),
+                    "user_name": data.get("name"),
+                }
+
+                logger.info(
+                    "kubiya_token_validated",
+                    org_id=org_data["id"],
+                    org_name=org_data["name"],
+                    user_email=org_data.get("user_email"),
+                )
+
+                return org_data
+
+            else:
+                logger.warning(
+                    "kubiya_token_invalid",
+                    status_code=response.status_code,
+                )
+                return None
+
+        except Exception as e:
+            logger.error("kubiya_api_error", error=str(e))
+            return None
+
+    async def get_runners(self, token: str, org_id: str) -> List[Dict]:
+        """
+        Get available runners for organization from Kubiya API.
+        Automatically tries both Bearer and UserKey authentication methods.
+
+        Args:
+            token: Authentication token (Bearer/idToken or UserKey/API key)
+            org_id: Organization UUID
+
+        Returns:
+            List of runner dicts:
+            [
+                {
+                    "name": "runner-name",
+                    "wss_url": "...",
+                    "task_id": "...",
+                    ...
+                }
+            ]
+        """
+        try:
+            # Try Bearer authentication first (Auth0 idToken)
+            response = await self.client.get(
+                f"{self.api_base}/api/v3/runners",
+                headers={"Authorization": f"Bearer {token}"},
+            )
+
+            # If Bearer fails with 401, try UserKey (API key)
+            if response.status_code == 401:
+                logger.info(
+                    "kubiya_runners_bearer_failed_trying_userkey",
+                    org_id=org_id,
+                )
+                response = await self.client.get(
+                    f"{self.api_base}/api/v3/runners",
+                    headers={"Authorization": f"UserKey {token}"},
+                )
+
+            if response.status_code == 200:
+                runners = response.json()
+
+                # Handle both array response and object response
+                if isinstance(runners, dict):
+                    # If it's a dict, extract the array from common keys
+                    runners = runners.get('runners', runners.get('data', []))
+
+                # Ensure it's a list
+                if not isinstance(runners, list):
+                    logger.warning(
+                        "kubiya_runners_unexpected_format",
+                        type=type(runners).__name__,
+                    )
+                    runners = []
+
+                logger.info(
+                    "kubiya_runners_fetched",
+                    org_id=org_id,
+                    runner_count=len(runners),
+                )
+
+                return runners
+
+            else:
+                logger.warning(
+                    "kubiya_runners_fetch_failed",
+                    status_code=response.status_code,
+                )
+                return []
+
+        except Exception as e:
+            logger.error("kubiya_runners_error", error=str(e))
+            return []
+
+    async def register_runner_heartbeat(
+        self, token: str, org_id: str, runner_name: str, metadata: Dict = None
+    ) -> bool:
+        """
+        Register runner heartbeat with Kubiya API.
+
+        Called by workers to report they're alive and polling.
+
+        Args:
+            token: Service token for worker
+            org_id: Organization UUID
+            runner_name: Runner name
+            metadata: Additional metadata (capabilities, version, etc.)
+
+        Returns:
+            True if successful, False otherwise
+        """
+        try:
+            response = await self.client.post(
+                f"{self.api_base}/api/v1/runners/heartbeat",
+                headers={"Authorization": f"UserKey {token}"},
+                json={
+                    "organization_id": org_id,
+                    "runner_name": runner_name,
+                    "status": "active",
+                    "metadata": metadata or {},
+                    "task_queue": f"{org_id}.{runner_name}",
+                },
+            )
+
+            if response.status_code in [200, 201, 204]:
+                logger.info(
+                    "kubiya_heartbeat_sent",
+                    org_id=org_id,
+                    runner_name=runner_name,
+                )
+                return True
+            else:
+                logger.warning(
+                    "kubiya_heartbeat_failed",
+                    status_code=response.status_code,
+                )
+                return False
+
+        except Exception as e:
+            logger.error("kubiya_heartbeat_error", error=str(e))
+            return False
+
+    async def close(self):
+        """Close the HTTP client"""
+        await self.client.aclose()
+
+
+# Singleton instance
+_kubiya_client: Optional[KubiyaClient] = None
+
+
+def get_kubiya_client() -> KubiyaClient:
+    """Get or create Kubiya client singleton"""
+    global _kubiya_client
+
+    if _kubiya_client is None:
+        _kubiya_client = KubiyaClient()
+
+    return _kubiya_client