BanGUI/backend/app/utils/scheduler_lock.py

"""Database-based scheduler lock for single-executor enforcement.

This module implements a database-backed lock mechanism that ensures only one
BanGUI instance runs the background scheduler, even in container orchestration
environments where multiple instances might start simultaneously.

The lock uses atomic database operations to prevent race conditions:
- Lock acquisition is atomic: INSERT fails if the singleton row already exists
- Lock release is atomic: DELETE with PID check ensures only the owner releases
- Stale lock detection uses heartbeat timestamps: a lock older than TTL is
  considered abandoned and eligible for cleanup on the next startup

This approach is more reliable than filesystem-based locking in containerized
environments because:
1. Database transactions are atomic (no TOCTOU race windows)
2. No NFS/network filesystem edge cases
3. Stale lock detection is timestamp-based, not PID-based (PID reuse is unreliable)
4. Works across container restarts and rolling deployments

The lock record stores:
    - id: Always 1 (singleton table)
    - pid: Process ID of the lock holder
    - hostname: Container/host name for debugging
    - created_at: When the lock was first acquired
    - heartbeat_at: When the lock was last confirmed alive (updated periodically)

On startup:
1. Cleanup any stale locks (where heartbeat_at > TTL)
2. Try to insert the lock for this instance
3. If INSERT succeeds, lock is acquired
4. If INSERT fails (IntegrityError), another instance holds the lock

On running (periodic):
    - Update heartbeat_at to keep the lock alive and prevent false positives

On shutdown:
    - Delete the lock (this instance is no longer running the scheduler)
"""

from __future__ import annotations

import os
import socket
import time
from typing import Any

import aiosqlite
import structlog

log: structlog.stdlib.BoundLogger = structlog.get_logger()

# Lock record expires if heartbeat hasn't been updated for this many seconds.
# This prevents stale locks from a crashed instance from blocking new startups.
# Set conservatively to allow temporary delays (e.g., high load) before considering
# the lock abandoned.
SCHEDULER_LOCK_TTL_SECONDS: int = 60

# Heartbeat interval: how often to update the lock's heartbeat_at timestamp.
# Must be significantly less than TTL (at least 3-4x smaller) to allow multiple
# consecutive missed heartbeats before the lock is considered stale.
# With TTL=60s and interval=5s, the lock survives ~12 missed heartbeats before
# expiring, providing robust protection against temporary delays.
SCHEDULER_LOCK_HEARTBEAT_INTERVAL_SECONDS: int = 5


async def init_scheduler_lock_table(db: aiosqlite.Connection) -> None:
    """Create the scheduler_lock table if it doesn't exist.

    This is called during database schema initialization and is safe to call
    multiple times (CREATE TABLE IF NOT EXISTS is idempotent).

    Args:
        db: The SQLite database connection.
    """
    await db.execute(
        """
        CREATE TABLE IF NOT EXISTS scheduler_lock (
            id INTEGER PRIMARY KEY CHECK (id = 1),
            pid INTEGER NOT NULL,
            hostname TEXT NOT NULL,
            created_at REAL NOT NULL,
            heartbeat_at REAL NOT NULL
        );
        """
    )
    await db.commit()


async def acquire_scheduler_lock(db: aiosqlite.Connection) -> bool:
    """Try to acquire the scheduler lock.

    This function performs two operations:
    1. Clean up any stale locks (where heartbeat_at + TTL < now)
    2. Try to insert a lock record for this instance

    If another instance already holds a valid lock, the INSERT will fail and
    this function returns False. The caller should reject startup with a clear
    error message.

    Args:
        db: The SQLite database connection.

    Returns:
        True if the lock was successfully acquired, False if held by another instance.

    Raises:
        RuntimeError: If database operations fail for reasons other than the lock
                      being held (e.g., database is corrupted or inaccessible).
    """
    now = time.time()
    pid = os.getpid()
    hostname = socket.gethostname()

    try:
        # Clean up stale locks first
        await db.execute(
            """
            DELETE FROM scheduler_lock
            WHERE (? - heartbeat_at) > ?
            """,
            (now, SCHEDULER_LOCK_TTL_SECONDS),
        )

        # Try to acquire the lock (atomic: INSERT fails if row exists)
        await db.execute(
            """
            INSERT INTO scheduler_lock (id, pid, hostname, created_at, heartbeat_at)
            VALUES (1, ?, ?, ?, ?)
            """,
            (pid, hostname, now, now),
        )
        await db.commit()

        log.info(
            "scheduler_lock_acquired",
            pid=pid,
            hostname=hostname,
        )
        return True

    except aiosqlite.IntegrityError:
        # Lock is already held by another instance (INSERT failed due to UNIQUE constraint)
        # Log details about who holds the lock to help with debugging
        try:
            cursor = await db.execute(
                "SELECT pid, hostname, created_at, heartbeat_at FROM scheduler_lock WHERE id = 1"
            )
            row = await cursor.fetchone()
            if row:
                lock_pid, lock_hostname, lock_created, lock_heartbeat = row
                age_seconds = now - lock_created
                heartbeat_age = now - lock_heartbeat
                log.warning(
                    "scheduler_lock_held_by_other_instance",
                    our_pid=pid,
                    lock_pid=lock_pid,
                    lock_hostname=lock_hostname,
                    lock_age_seconds=age_seconds,
                    heartbeat_age_seconds=heartbeat_age,
                )
        except Exception as e:
            log.warning("scheduler_lock_held_but_could_not_read_holder", error=str(e))

        return False

    except Exception as e:
        # Unexpected database error (not an IntegrityError)
        raise RuntimeError(
            f"Failed to acquire scheduler lock due to database error: {e}\n"
            "Check that the database is accessible and not corrupted."
        ) from e


async def release_scheduler_lock(db: aiosqlite.Connection) -> None:
    """Release the scheduler lock.

    This function should be called during application shutdown. It removes the
    lock record, allowing other instances to acquire it.

    Args:
        db: The SQLite database connection.

    Raises:
        RuntimeError: If database operations fail.
    """
    pid = os.getpid()

    try:
        cursor = await db.execute(
            "DELETE FROM scheduler_lock WHERE id = 1 AND pid = ?",
            (pid,),
        )
        await db.commit()

        if cursor.rowcount == 0:
            # This shouldn't happen in normal operation, but log it for visibility
            log.warning(
                "scheduler_lock_release_mismatch",
                our_pid=pid,
                message="Tried to release lock but we don't hold it. Another instance may have replaced us.",
            )
        else:
            log.info("scheduler_lock_released", pid=pid)

    except Exception as e:
        raise RuntimeError(f"Failed to release scheduler lock: {e}") from e


async def update_scheduler_lock_heartbeat(db: aiosqlite.Connection) -> bool:
    """Update the heartbeat timestamp to keep the lock alive.

    This function should be called periodically (every ~10 seconds) to prevent
    the lock from being considered stale. It only succeeds if this process
    still holds the lock.

    Args:
        db: The SQLite database connection.

    Returns:
        True if the heartbeat was updated (we still hold the lock), False if
        we no longer hold the lock (another instance has taken over).

    Raises:
        RuntimeError: If database operations fail.
    """
    now = time.time()
    pid = os.getpid()

    try:
        cursor = await db.execute(
            "UPDATE scheduler_lock SET heartbeat_at = ? WHERE id = 1 AND pid = ?",
            (now, pid),
        )
        await db.commit()

        if cursor.rowcount == 0:
            # We no longer hold the lock
            log.warning(
                "scheduler_lock_heartbeat_lost",
                our_pid=pid,
                message="Heartbeat failed; we no longer hold the lock.",
            )
            return False

        return True

    except Exception as e:
        raise RuntimeError(f"Failed to update scheduler lock heartbeat: {e}") from e


async def get_scheduler_lock_info(db: aiosqlite.Connection) -> dict[str, Any] | None:
    """Retrieve information about the current scheduler lock.

    This function is useful for debugging and monitoring. Returns None if no
    lock is currently held.

    Args:
        db: The SQLite database connection.

    Returns:
        A dict with keys: pid, hostname, created_at, heartbeat_at, or None
        if no lock exists.
    """
    try:
        cursor = await db.execute(
            "SELECT pid, hostname, created_at, heartbeat_at FROM scheduler_lock WHERE id = 1"
        )
        row = await cursor.fetchone()
        if row:
            pid, hostname, created_at, heartbeat_at = row
            return {
                "pid": pid,
                "hostname": hostname,
                "created_at": created_at,
                "heartbeat_at": heartbeat_at,
            }
        return None
    except Exception as e:
        log.warning("scheduler_lock_info_query_failed", error=str(e))
        return None