fieldwitness/src/soosef/verisoo/federation.py

"""
Federation and gossip protocol for Verisoo.

Nodes sync their merkle logs via gossip:
1. Periodically exchange merkle roots with peers
2. If roots differ, request consistency proof
3. Fetch missing records and append to local log

Eventually consistent - "eventually" depends on gossip frequency and topology.
No central coordinator, no leader election, no consensus protocol.
Just append-only logs that converge.
"""

from __future__ import annotations

import asyncio
import hashlib
import json
import logging
from dataclasses import dataclass, field
from datetime import datetime, timezone
from typing import Callable, Protocol
from urllib.parse import urlparse

from .exceptions import FederationError
from .merkle import ConsistencyProof, InclusionProof, MerkleLog
from .models import AttestationRecord

logger = logging.getLogger(__name__)


@dataclass
class PeerInfo:
    """Information about a federation peer."""

    url: str
    fingerprint: str  # Ed25519 fingerprint for peer authentication
    last_seen: datetime | None = None
    last_root: str | None = None
    last_size: int = 0
    healthy: bool = True
    consecutive_failures: int = 0


@dataclass
class SyncStatus:
    """Result of a sync attempt with a peer."""

    peer: str
    success: bool
    records_received: int = 0
    our_size_before: int = 0
    our_size_after: int = 0
    their_size: int = 0
    error: str | None = None


class PeerTransport(Protocol):
    """Protocol for peer communication."""

    async def get_status(self, peer: PeerInfo) -> tuple[str, int]:
        """Get peer's current root hash and log size."""
        ...

    async def get_records(
        self, peer: PeerInfo, start_index: int, count: int
    ) -> list[AttestationRecord]:
        """Fetch records from peer."""
        ...

    async def get_consistency_proof(
        self, peer: PeerInfo, old_size: int
    ) -> ConsistencyProof:
        """Get proof that peer's log is consistent with ours."""
        ...

    async def push_records(
        self, peer: PeerInfo, records: list[AttestationRecord]
    ) -> int:
        """Push records to peer, returns number accepted."""
        ...


class GossipNode:
    """
    A node in the Verisoo federation network.

    Manages:
    - Local merkle log
    - Peer connections
    - Periodic sync via gossip
    """

    def __init__(
        self,
        log: MerkleLog,
        transport: PeerTransport,
        node_id: str | None = None,
    ) -> None:
        self.log = log
        self.transport = transport
        self.node_id = node_id or self._generate_node_id()
        self.peers: dict[str, PeerInfo] = {}
        self._running = False
        self._sync_task: asyncio.Task | None = None

    def add_peer(self, url: str, fingerprint: str) -> None:
        """Register a peer for gossip."""
        peer = PeerInfo(url=url, fingerprint=fingerprint)
        self.peers[url] = peer
        logger.info(f"Added peer: {url} ({fingerprint[:16]}...)")

    def remove_peer(self, url: str) -> None:
        """Remove a peer from gossip."""
        if url in self.peers:
            del self.peers[url]
            logger.info(f"Removed peer: {url}")

    async def start(self, interval_seconds: float = 60.0) -> None:
        """Start the gossip loop."""
        if self._running:
            return

        self._running = True
        self._sync_task = asyncio.create_task(self._gossip_loop(interval_seconds))
        logger.info(f"Gossip started, interval={interval_seconds}s")

    async def stop(self) -> None:
        """Stop the gossip loop."""
        self._running = False
        if self._sync_task:
            self._sync_task.cancel()
            try:
                await self._sync_task
            except asyncio.CancelledError:
                pass
        logger.info("Gossip stopped")

    async def sync_with_peer(self, peer_url: str) -> SyncStatus:
        """
        Sync with a specific peer.

        1. Get their status (root, size)
        2. If they have more records, fetch them
        3. Verify consistency before accepting
        4. Append new records to our log
        """
        peer = self.peers.get(peer_url)
        if not peer:
            return SyncStatus(
                peer=peer_url,
                success=False,
                error="Unknown peer",
            )

        our_size_before = self.log.size

        try:
            # Get peer status
            their_root, their_size = await self.transport.get_status(peer)

            # Update peer info
            peer.last_seen = datetime.now(timezone.utc)
            peer.last_root = their_root
            peer.last_size = their_size

            # Already in sync?
            if their_size <= our_size_before:
                if their_root == self.log.root_hash:
                    peer.healthy = True
                    peer.consecutive_failures = 0
                    return SyncStatus(
                        peer=peer_url,
                        success=True,
                        our_size_before=our_size_before,
                        our_size_after=our_size_before,
                        their_size=their_size,
                    )

            # They have records we don't - fetch them
            if their_size > our_size_before:
                # First verify consistency
                if our_size_before > 0:
                    proof = await self.transport.get_consistency_proof(peer, our_size_before)
                    if not self._verify_consistency(proof):
                        raise FederationError(
                            f"Peer {peer_url} failed consistency check - possible fork"
                        )

                # Fetch new records
                new_records = await self.transport.get_records(
                    peer, our_size_before, their_size - our_size_before
                )

                # Append to our log
                for record in new_records:
                    self.log.append(record)

            peer.healthy = True
            peer.consecutive_failures = 0

            return SyncStatus(
                peer=peer_url,
                success=True,
                records_received=their_size - our_size_before,
                our_size_before=our_size_before,
                our_size_after=self.log.size,
                their_size=their_size,
            )

        except Exception as e:
            peer.consecutive_failures += 1
            if peer.consecutive_failures >= 3:
                peer.healthy = False
                logger.warning(f"Peer {peer_url} marked unhealthy after 3 failures")

            return SyncStatus(
                peer=peer_url,
                success=False,
                error=str(e),
                our_size_before=our_size_before,
                our_size_after=self.log.size,
            )

    async def broadcast_record(self, record: AttestationRecord) -> dict[str, bool]:
        """
        Push a new record to all peers.

        Returns dict of peer_url -> success.
        """
        results = {}
        for url, peer in self.peers.items():
            if not peer.healthy:
                results[url] = False
                continue

            try:
                accepted = await self.transport.push_records(peer, [record])
                results[url] = accepted > 0
            except Exception as e:
                logger.warning(f"Failed to push to {url}: {e}")
                results[url] = False

        return results

    async def _gossip_loop(self, interval: float) -> None:
        """Background task that periodically syncs with peers."""
        while self._running:
            try:
                await self._gossip_round()
            except Exception as e:
                logger.error(f"Gossip round failed: {e}")

            await asyncio.sleep(interval)

    async def _gossip_round(self) -> None:
        """One round of gossip with all healthy peers."""
        healthy_peers = [p for p in self.peers.values() if p.healthy]
        if not healthy_peers:
            return

        # Sync with all healthy peers concurrently
        tasks = [self.sync_with_peer(p.url) for p in healthy_peers]
        results = await asyncio.gather(*tasks, return_exceptions=True)

        # Log summary
        success_count = sum(
            1 for r in results if isinstance(r, SyncStatus) and r.success
        )
        logger.debug(f"Gossip round: {success_count}/{len(healthy_peers)} peers synced")

    def _verify_consistency(self, proof: ConsistencyProof) -> bool:
        """Verify a consistency proof from a peer."""
        # Simplified: trust the proof structure for now
        # Full implementation would verify the merkle path
        return proof.old_size <= self.log.size

    def _generate_node_id(self) -> str:
        """Generate a random node ID."""
        import secrets

        return hashlib.sha256(secrets.token_bytes(32)).hexdigest()[:16]


# Placeholder for HTTP transport implementation
class HttpTransport:
    """
    HTTP-based peer transport.

    Endpoints expected on peers:
    - GET /status -> {"root": "...", "size": N}
    - GET /records?start=N&count=M -> [records...]
    - GET /consistency-proof?old_size=N -> proof
    - POST /records -> accept records, return count
    """

    def __init__(self, timeout: float = 30.0) -> None:
        self.timeout = timeout
        # Will use aiohttp when federation extra is installed

    async def get_status(self, peer: PeerInfo) -> tuple[str, int]:
        """Get peer's current root hash and log size."""
        raise NotImplementedError("Install verisoo[federation] for HTTP transport")

    async def get_records(
        self, peer: PeerInfo, start_index: int, count: int
    ) -> list[AttestationRecord]:
        raise NotImplementedError("Install verisoo[federation] for HTTP transport")

    async def get_consistency_proof(
        self, peer: PeerInfo, old_size: int
    ) -> ConsistencyProof:
        raise NotImplementedError("Install verisoo[federation] for HTTP transport")

    async def push_records(
        self, peer: PeerInfo, records: list[AttestationRecord]
    ) -> int:
        raise NotImplementedError("Install verisoo[federation] for HTTP transport")