atlas-iac/testing/tests/test_hermes_publication_retry_fence.py

"""Transactional publication-retry fence coverage."""
from __future__ import annotations

import json
import sqlite3
from contextlib import contextmanager

from testing.tests.test_hermes_cli_support import _load


fence = _load("publication_retry_fence")


class Native:
    """Minimal existing-native transaction surface used by the adapter."""

    @staticmethod
    @contextmanager
    def write_txn(connection):
        connection.execute("BEGIN IMMEDIATE")
        try:
            yield
        except Exception:
            connection.rollback()
            raise
        else:
            connection.commit()

    @staticmethod
    def _append_event(connection, task_id, kind, payload):
        connection.execute(
            "INSERT INTO task_events(task_id,run_id,kind,payload) VALUES(?,?,?,?)",
            (task_id, None, kind, json.dumps(payload, sort_keys=True)),
        )


def _database():
    connection = sqlite3.connect(":memory:")
    connection.row_factory = sqlite3.Row
    connection.executescript("""
      CREATE TABLE tasks(id TEXT PRIMARY KEY,status TEXT,current_run_id INTEGER,
        consecutive_failures INTEGER,last_failure_error TEXT);
      CREATE TABLE task_links(child_id TEXT,parent_id TEXT);
      CREATE TABLE task_events(id INTEGER PRIMARY KEY AUTOINCREMENT,task_id TEXT,
        run_id INTEGER,kind TEXT,payload TEXT);
    """)
    connection.execute("INSERT INTO tasks VALUES('root','done',NULL,0,NULL)")
    connection.execute("INSERT INTO tasks VALUES('child','blocked',NULL,1,'broker')")
    connection.execute("INSERT INTO task_links VALUES('child','root')")
    return connection


def _blocked(connection, marker):
    event_id = connection.execute(
        "INSERT INTO task_events(task_id,run_id,kind,payload) VALUES(?,?,?,?)",
        ("child", 8, "blocked", json.dumps({"reason": "transport\\n" + marker})),
    ).lastrowid
    connection.commit()
    return event_id


def test_adapter_reopens_only_the_exact_native_fence():
    """The existing native transaction protects the event check and state change."""
    connection = _database()
    marker = "[hermes-publication-retry-fence:8:" + "a" * 64 + "]"
    event_id = _blocked(connection, marker)

    assert fence.guarded_unblock(Native, connection, "child", "root", "8", (event_id, marker))
    assert connection.execute("SELECT status FROM tasks WHERE id='child'").fetchone()[0] == "ready"
    assert connection.execute("SELECT kind FROM task_events ORDER BY id DESC LIMIT 1").fetchone()[0] == "unblocked"


def test_adapter_keeps_a_later_human_transient_block_closed():
    """A later event, even with the same block kind, invalidates coordinator ownership."""
    connection = _database()
    marker = "[hermes-publication-retry-fence:8:" + "b" * 64 + "]"
    event_id = _blocked(connection, marker)
    connection.execute(
        "INSERT INTO task_events(task_id,run_id,kind,payload) VALUES(?,?,?,?)",
        ("child", None, "blocked", json.dumps({"reason": "human decision"})),
    )
    connection.commit()

    assert not fence.guarded_unblock(Native, connection, "child", "root", "8", (event_id, marker))
    assert connection.execute("SELECT status FROM tasks WHERE id='child'").fetchone()[0] == "blocked"


def test_block_records_the_exact_native_event_as_coordinator_owned(tmp_path, monkeypatch):
    """Only the coordinator's successful fenced block creates a sidecar ownership row."""
    monkeypatch.setattr(fence.supervisor_state, "KANBAN_ROOT", tmp_path / "boards")
    connection = _database()
    connection.execute("UPDATE tasks SET status='running',current_run_id=8 WHERE id='child'")
    connection.commit()

    class BlockingNative(Native):
        @staticmethod
        def block_task(conn, task_id, *, reason, kind, expected_run_id):
            changed = conn.execute(
                "UPDATE tasks SET status='blocked',current_run_id=NULL WHERE id=? AND status='running' AND current_run_id=?",
                (task_id, expected_run_id),
            ).rowcount
            if changed:
                conn.execute(
                    "INSERT INTO task_events(task_id,run_id,kind,payload) VALUES(?,?,?,?)",
                    (task_id, expected_run_id, "blocked", json.dumps({"reason": reason, "kind": kind})),
                )
            return bool(changed)

    receipt = {"result_digest": "c" * 64}
    assert fence.block(BlockingNative, connection, "soteria", "child", "8", "transport", "transient", receipt)
    event_id = connection.execute("SELECT id FROM task_events ORDER BY id DESC LIMIT 1").fetchone()[0]
    marker = "[hermes-publication-retry-fence:8:" + "c" * 64 + "]"
    assert fence.read("soteria", "child", "8") == (event_id, marker)
    # Simulate a crash after native block but before its sidecar fence record.
    with fence.supervisor_state._connect("soteria") as state:
        state.execute("DELETE FROM publication_retry_fences")
        state.execute(
            "INSERT INTO publication_retries(board,child_task_id,source_run_id,source_ordinal,receipt_json) VALUES(?,?,?,?,?)",
            ("soteria", "child", "8", 0, json.dumps(receipt)),
        )
    assert fence.recover("soteria", "child", "8", connection, receipt) == (event_id, marker)