hindsight-api 0.2.1__py3-none-any.whl → 0.3.0__py3-none-any.whl

hindsight_api/engine/db_budget.py

@@ -0,0 +1,284 @@
+"""
+Database connection budget management.
+
+Limits concurrent database connections per operation to prevent
+a single operation (e.g., recall with parallel queries) from
+exhausting the connection pool.
+"""
+
+import asyncio
+import logging
+import uuid
+from contextlib import asynccontextmanager
+from dataclasses import dataclass, field
+from typing import TYPE_CHECKING, AsyncIterator
+
+if TYPE_CHECKING:
+    import asyncpg
+
+logger = logging.getLogger(__name__)
+
+
+@dataclass
+class OperationBudget:
+    """
+    Tracks connection budget for a single operation.
+
+    Each operation gets a semaphore limiting its concurrent connections.
+    """
+
+    operation_id: str
+    max_connections: int
+    semaphore: asyncio.Semaphore = field(init=False)
+    active_count: int = field(default=0, init=False)
+
+    def __post_init__(self):
+        self.semaphore = asyncio.Semaphore(self.max_connections)
+
+
+class ConnectionBudgetManager:
+    """
+    Manages per-operation connection budgets.
+
+    Usage:
+        manager = ConnectionBudgetManager(default_budget=4)
+
+        # Start an operation
+        async with manager.operation(max_connections=2) as op:
+            # Acquire connections within the budget
+            async with op.acquire(pool) as conn:
+                await conn.fetch(...)
+
+            # Multiple connections respect the budget
+            async with op.acquire(pool) as conn1, op.acquire(pool) as conn2:
+                # At most 2 concurrent connections for this operation
+                ...
+    """
+
+    def __init__(self, default_budget: int = 4):
+        """
+        Initialize the budget manager.
+
+        Args:
+            default_budget: Default max connections per operation
+        """
+        self.default_budget = default_budget
+        self._operations: dict[str, OperationBudget] = {}
+        self._lock = asyncio.Lock()
+
+    @asynccontextmanager
+    async def operation(
+        self,
+        max_connections: int | None = None,
+        operation_id: str | None = None,
+    ) -> AsyncIterator["BudgetedOperation"]:
+        """
+        Create a budgeted operation context.
+
+        Args:
+            max_connections: Max concurrent connections for this operation.
+                           Defaults to manager's default_budget.
+            operation_id: Optional custom operation ID. Auto-generated if not provided.
+
+        Yields:
+            BudgetedOperation context for acquiring connections
+        """
+        op_id = operation_id or f"op-{uuid.uuid4().hex[:12]}"
+        budget = max_connections or self.default_budget
+
+        async with self._lock:
+            if op_id in self._operations:
+                raise ValueError(f"Operation {op_id} already exists")
+            self._operations[op_id] = OperationBudget(op_id, budget)
+
+        try:
+            yield BudgetedOperation(self, op_id)
+        finally:
+            async with self._lock:
+                self._operations.pop(op_id, None)
+
+    def _get_budget(self, operation_id: str) -> OperationBudget:
+        """Get budget for an operation (internal use)."""
+        budget = self._operations.get(operation_id)
+        if not budget:
+            raise ValueError(f"Operation {operation_id} not found")
+        return budget
+
+
+class BudgetedOperation:
+    """
+    A single operation with connection budget.
+
+    Provides methods to acquire connections within the budget.
+    """
+
+    def __init__(self, manager: ConnectionBudgetManager, operation_id: str):
+        self._manager = manager
+        self.operation_id = operation_id
+
+    @property
+    def budget(self) -> OperationBudget:
+        """Get the budget for this operation."""
+        return self._manager._get_budget(self.operation_id)
+
+    @asynccontextmanager
+    async def acquire(self, pool: "asyncpg.Pool") -> AsyncIterator["asyncpg.Connection"]:
+        """
+        Acquire a connection within the operation's budget.
+
+        Blocks if the operation has reached its connection limit.
+
+        Args:
+            pool: asyncpg connection pool
+
+        Yields:
+            Database connection
+        """
+        budget = self.budget
+        async with budget.semaphore:
+            budget.active_count += 1
+            conn = await pool.acquire()
+            try:
+                yield conn
+            finally:
+                budget.active_count -= 1
+                await pool.release(conn)
+
+    def wrap_pool(self, pool: "asyncpg.Pool") -> "BudgetedPool":
+        """
+        Wrap a pool with this operation's budget.
+
+        The returned BudgetedPool can be passed to functions expecting a pool,
+        and all acquire() calls will be limited by this operation's budget.
+
+        Args:
+            pool: asyncpg connection pool to wrap
+
+        Returns:
+            BudgetedPool that limits connections to this operation's budget
+        """
+        return BudgetedPool(pool, self)
+
+    async def acquire_many(
+        self,
+        pool: "asyncpg.Pool",
+        count: int,
+    ) -> AsyncIterator[list["asyncpg.Connection"]]:
+        """
+        Acquire multiple connections within the budget.
+
+        Note: This acquires connections sequentially to respect the budget.
+        For parallel acquisition, use multiple acquire() calls with asyncio.gather().
+
+        Args:
+            pool: asyncpg connection pool
+            count: Number of connections to acquire
+
+        Yields:
+            List of database connections
+        """
+        connections = []
+        try:
+            for _ in range(count):
+                conn = await pool.acquire()
+                connections.append(conn)
+            yield connections
+        finally:
+            for conn in connections:
+                await pool.release(conn)
+
+
+# Global default manager instance
+_default_manager: ConnectionBudgetManager | None = None
+
+
+def get_budget_manager(default_budget: int = 4) -> ConnectionBudgetManager:
+    """
+    Get or create the global budget manager.
+
+    Args:
+        default_budget: Default max connections per operation
+
+    Returns:
+        Global ConnectionBudgetManager instance
+    """
+    global _default_manager
+    if _default_manager is None:
+        _default_manager = ConnectionBudgetManager(default_budget=default_budget)
+    return _default_manager
+
+
+@asynccontextmanager
+async def budgeted_operation(
+    max_connections: int | None = None,
+    operation_id: str | None = None,
+    default_budget: int = 4,
+) -> AsyncIterator[BudgetedOperation]:
+    """
+    Convenience function to create a budgeted operation.
+
+    Args:
+        max_connections: Max concurrent connections for this operation
+        operation_id: Optional custom operation ID
+        default_budget: Default budget if manager not yet created
+
+    Yields:
+        BudgetedOperation context
+
+    Example:
+        async with budgeted_operation(max_connections=2) as op:
+            async with op.acquire(pool) as conn:
+                await conn.fetch(...)
+    """
+    manager = get_budget_manager(default_budget)
+    async with manager.operation(max_connections, operation_id) as op:
+        yield op
+
+
+class BudgetedPool:
+    """
+    A pool wrapper that limits concurrent connection acquisitions.
+
+    This can be passed to functions expecting a pool, and acquire()
+    calls will be limited by the budget semaphore.
+
+    Usage:
+        async with budgeted_operation(max_connections=4) as op:
+            budgeted_pool = op.wrap_pool(pool)
+            # Pass budgeted_pool to functions that expect a pool
+            await some_function(budgeted_pool, ...)
+    """
+
+    def __init__(self, pool: "asyncpg.Pool", operation: BudgetedOperation):
+        self._pool = pool
+        self._operation = operation
+
+    async def acquire(self) -> "asyncpg.Connection":
+        """
+        Acquire a connection within the budget.
+
+        Note: Caller must release the connection when done.
+        Prefer using as context manager via acquire_with_retry or op.acquire().
+        """
+        budget = self._operation.budget
+        await budget.semaphore.acquire()
+        budget.active_count += 1
+        try:
+            return await self._pool.acquire()
+        except Exception:
+            budget.active_count -= 1
+            budget.semaphore.release()
+            raise
+
+    async def release(self, conn: "asyncpg.Connection") -> None:
+        """Release a connection back to the pool."""
+        budget = self._operation.budget
+        try:
+            await self._pool.release(conn)
+        finally:
+            budget.active_count -= 1
+            budget.semaphore.release()
+
+    def __getattr__(self, name):
+        """Proxy other attributes to the underlying pool."""
+        return getattr(self._pool, name)

hindsight_api/engine/db_utils.py

@@ -83,11 +83,22 @@ async def acquire_with_retry(pool: asyncpg.Pool, max_retries: int = DEFAULT_MAX_
     Yields:
         An asyncpg connection
     """
+    import time
+
+    start = time.time()
 
     async def acquire():
         return await pool.acquire()
 
     conn = await retry_with_backoff(acquire, max_retries=max_retries)
+    acquire_time = time.time() - start
+
+    # Log slow connection acquisitions (indicates pool contention)
+    if acquire_time > 0.05:  # 50ms threshold
+        pool_size = pool.get_size()
+        pool_free = pool.get_idle_size()
+        logger.warning(f"[DB POOL] Slow acquire: {acquire_time:.3f}s | size={pool_size}, idle={pool_free}")
+
     try:
         yield conn
     finally: