Files
hermes-agent/tests/agent/test_compression_closed_adoption.py
ethernet 890bbbda1f Merge remote-tracking branch 'origin/main' into ethie/pm-clean
# Conflicts:
#	apps/desktop/e2e/archived-hidden-session-recoverable.spec.ts
#	apps/desktop/e2e/bot-chat-message-agent-friendly-name.spec.ts
#	apps/desktop/e2e/bot-mailbox-unreadable-ticket.spec.ts
#	apps/desktop/e2e/bot-mode-roster-localized.spec.ts
#	apps/desktop/e2e/bot-mode-row-click-mirrors-registry.spec.ts
#	apps/desktop/e2e/bot-mode-tab-shows-bot-name.spec.ts
#	apps/desktop/e2e/bot-roster-group-row-organisation.spec.ts
#	apps/desktop/e2e/bot-roster-ignores-infra-dirs.spec.ts
#	apps/desktop/e2e/bot-roster-timestamp-meta.spec.ts
#	apps/desktop/e2e/bot-roster-user-sections.spec.ts
#	apps/desktop/e2e/bot-routines-pane-narrow.spec.ts
#	apps/desktop/e2e/bot-row-open-recent-session.spec.ts
#	apps/desktop/e2e/bot-tile-ignores-ambient-composer-model.spec.ts
#	apps/desktop/e2e/group-composer-auto-grow.spec.ts
#	apps/desktop/e2e/group-create-gate-remote-roster.spec.ts
#	apps/desktop/e2e/group-prompt-renamed-primary-handle.spec.ts
#	apps/desktop/e2e/hosted-room-backend-continuity.spec.ts
#	apps/desktop/e2e/hosted-room-legacy-store-migration.spec.ts
#	apps/desktop/e2e/settings-scope-chips-bot-title.spec.ts
#	apps/desktop/e2e/worktree-branch-status.spec.ts
#	apps/desktop/electron/backend-probes.test.ts
#	apps/desktop/electron/connection-apply.test.ts
#	apps/desktop/electron/desktop-electron-pin.test.ts
#	apps/desktop/electron/desktop-uninstall.test.ts
#	apps/desktop/electron/gateway-file-download-transport.test.ts
#	apps/desktop/electron/gateway-stop-before-update.test.ts
#	apps/desktop/electron/github-api-auth.test.ts
#	apps/desktop/electron/registry-primary-profile-scope.test.ts
#	apps/desktop/electron/update-api-check.test.ts
#	apps/desktop/electron/update-handoff-marker.test.ts
#	apps/desktop/electron/venv-blocker-scan.test.ts
#	apps/desktop/scripts/after-extract.test.mjs
#	apps/desktop/scripts/local-pack-publish.test.mjs
#	apps/desktop/scripts/tasks-scroll.test.mjs
#	apps/desktop/src/app/settings/model-settings.test.tsx
#	apps/desktop/src/app/updates-overlay.blockers.test.tsx
#	apps/desktop/src/components/desktop-install-overlay.test.tsx
#	apps/desktop/src/lib/update-copy.test.ts
#	scripts/ci/check_os_marker_fakes.py
#	tests-js/desktop-mac-usage-descriptions.test.ts
#	tests-js/node-engine-alignment.test.ts
#	tests/agent/lsp/test_install_and_lint_fixes.py
#	tests/agent/test_command_token_source.py
#	tests/agent/test_compression_boundary_hook.py
#	tests/agent/test_create_openai_client_ssl_verify.py
#	tests/agent/test_custom_provider_ca_probes.py
#	tests/agent/test_endpoint_blackhole.py
#	tests/agent/test_estimator_parity.py
#	tests/agent/test_in_place_compaction.py
#	tests/agent/test_moa_loop_mode.py
#	tests/agent/test_model_metadata.py
#	tests/agent/test_skill_session_platform_gate.py
#	tests/agent/test_skill_utils.py
#	tests/agent/test_ssl_ca_guard.py
#	tests/computer_use/test_doctor.py
#	tests/cron/test_codex_execution_paths.py
#	tests/cron/test_cron_bot_chat_delivery.py
#	tests/cron/test_cron_script.py
#	tests/cron/test_media_delivery_parity.py
#	tests/cron/test_misfire_catchup.py
#	tests/cron/test_parallel_pool.py
#	tests/cron/test_recurring_eagain_redispatch.py
#	tests/gateway/test_choice_picker.py
#	tests/gateway/test_control_socket_windows_live.py
#	tests/gateway/test_dingtalk.py
#	tests/gateway/test_feishu.py
#	tests/gateway/test_feishu_onboard.py
#	tests/gateway/test_gateway_shutdown.py
#	tests/gateway/test_matrix.py
#	tests/gateway/test_model_command_custom_providers.py
#	tests/gateway/test_reasoning_command.py
#	tests/gateway/test_runtime_footer.py
#	tests/gateway/test_session.py
#	tests/gateway/test_session_hygiene.py
#	tests/gateway/test_status.py
#	tests/gateway/test_teams.py
#	tests/gateway/test_turn_lease.py
#	tests/gateway/test_whatsapp_connect.py
#	tests/hermes_cli/test_approvals_command.py
#	tests/hermes_cli/test_auth_store_lock_concurrent.py
#	tests/hermes_cli/test_backup.py
#	tests/hermes_cli/test_banner_git_state.py
#	tests/hermes_cli/test_certifi_repair.py
#	tests/hermes_cli/test_cmd_update.py
#	tests/hermes_cli/test_compat_manifest_targets.py
#	tests/hermes_cli/test_computer_use_cli.py
#	tests/hermes_cli/test_cpr_local_leak.py
#	tests/hermes_cli/test_dashboard_auth_gate.py
#	tests/hermes_cli/test_dashboard_procs_kill_grace.py
#	tests/hermes_cli/test_desktop_lifecycle_windows_live.py
#	tests/hermes_cli/test_doctor.py
#	tests/hermes_cli/test_doctor_command_install.py
#	tests/hermes_cli/test_fleet_config_migration_windows_live.py
#	tests/hermes_cli/test_gateway.py
#	tests/hermes_cli/test_gateway_platform_gating.py
#	tests/hermes_cli/test_gateway_restart_loop.py
#	tests/hermes_cli/test_gateway_task_probe.py
#	tests/hermes_cli/test_gateway_wsl.py
#	tests/hermes_cli/test_gui_command.py
#	tests/hermes_cli/test_install_cua_driver.py
#	tests/hermes_cli/test_kanban_db.py
#	tests/hermes_cli/test_lazy_command_exports.py
#	tests/hermes_cli/test_lazy_refresh_venv_repair.py
#	tests/hermes_cli/test_linux_desktop_entry.py
#	tests/hermes_cli/test_local_runtime.py
#	tests/hermes_cli/test_local_runtime_updates.py
#	tests/hermes_cli/test_managed_uv.py
#	tests/hermes_cli/test_mcp_reload_confirm_gate.py
#	tests/hermes_cli/test_nous_subscription.py
#	tests/hermes_cli/test_npm_engine.py
#	tests/hermes_cli/test_personality_none.py
#	tests/hermes_cli/test_pet_toggle.py
#	tests/hermes_cli/test_plan_reconciliation_windows_live.py
#	tests/hermes_cli/test_plugin_event_bus.py
#	tests/hermes_cli/test_plugin_manifest_v2.py
#	tests/hermes_cli/test_plugin_packs.py
#	tests/hermes_cli/test_plugins_cmd.py
#	tests/hermes_cli/test_plugins_cmd_enable_disable_nested.py
#	tests/hermes_cli/test_process_identity.py
#	tests/hermes_cli/test_profiles.py
#	tests/hermes_cli/test_profiles_sidebar_cache.py
#	tests/hermes_cli/test_pty_bridge.py
#	tests/hermes_cli/test_resolve_turn_limit.py
#	tests/hermes_cli/test_serve_runtime_inventory.py
#	tests/hermes_cli/test_session_vacuum_config.py
#	tests/hermes_cli/test_set_config_value.py
#	tests/hermes_cli/test_signal_handler_kanban_worker.py
#	tests/hermes_cli/test_slash_confirm_windows.py
#	tests/hermes_cli/test_stale_pid_guard.py
#	tests/hermes_cli/test_startup_fast_guards.py
#	tests/hermes_cli/test_status.py
#	tests/hermes_cli/test_telegram_managed_bot.py
#	tests/hermes_cli/test_tools_config.py
#	tests/hermes_cli/test_update_apply_shallow_count.py
#	tests/hermes_cli/test_update_autostash.py
#	tests/hermes_cli/test_update_concurrent_quarantine.py
#	tests/hermes_cli/test_update_fetch_failure_classifier.py
#	tests/hermes_cli/test_update_fleet_probe_resume_token.py
#	tests/hermes_cli/test_update_handoff_backend_reap.py
#	tests/hermes_cli/test_update_handoff_desktop_rebuild.py
#	tests/hermes_cli/test_update_head_moved_gate.py
#	tests/hermes_cli/test_update_host_obligation.py
#	tests/hermes_cli/test_update_import_guard.py
#	tests/hermes_cli/test_update_interrupted_recovery.py
#	tests/hermes_cli/test_update_inventory.py
#	tests/hermes_cli/test_update_launchd_unloaded_gateway.py
#	tests/hermes_cli/test_update_missing_configured_deps.py
#	tests/hermes_cli/test_update_modified_notice.py
#	tests/hermes_cli/test_update_multiplex_migration_hook.py
#	tests/hermes_cli/test_update_no_gateway_restart.py
#	tests/hermes_cli/test_update_orphan_backend_reap.py
#	tests/hermes_cli/test_update_parked_branch_guard.py
#	tests/hermes_cli/test_update_post_pull_syntax_guard.py
#	tests/hermes_cli/test_update_receipt.py
#	tests/hermes_cli/test_update_self_lock.py
#	tests/hermes_cli/test_update_shim_fail_closed.py
#	tests/hermes_cli/test_update_shim_self_lock.py
#	tests/hermes_cli/test_update_sqlite_remediation.py
#	tests/hermes_cli/test_update_stale_dashboard.py
#	tests/hermes_cli/test_update_stale_virtualenv.py
#	tests/hermes_cli/test_update_venv_health.py
#	tests/hermes_cli/test_update_venv_ownership_preflight.py
#	tests/hermes_cli/test_update_wedged_gateway.py
#	tests/hermes_cli/test_update_yes_flag.py
#	tests/hermes_cli/test_update_zip_two_phase.py
#	tests/hermes_cli/test_urllib_security.py
#	tests/hermes_cli/test_ux_messages_auth_config.py
#	tests/hermes_cli/test_ux_messages_startup.py
#	tests/hermes_cli/test_venv_holder_classifier.py
#	tests/hermes_cli/test_verify_console_scripts.py
#	tests/hermes_cli/test_verify_core_dependencies.py
#	tests/hermes_cli/test_web_server.py
#	tests/hermes_cli/test_web_server_console_ws.py
#	tests/hermes_cli/test_web_server_ws_ping.py
#	tests/hermes_cli/test_web_ui_build.py
#	tests/hermes_state/test_fts_rebuild_admission.py
#	tests/hermes_state/test_hermes_state.py
#	tests/plugins/memory/test_memory_lazy_install.py
#	tests/plugins/test_google_meet_plugin.py
#	tests/plugins/test_langfuse_plugin.py
#	tests/plugins/test_security_guidance_plugin.py
#	tests/plugins/test_transform_llm_output_hook.py
#	tests/scripts/desktop_update/test_desktop_update_windows_gateway_flag.py
#	tests/scripts/desktop_update/test_desktop_update_windows_python_handoff.py
#	tests/scripts/desktop_update/test_desktop_update_windows_timestamp.py
#	tests/scripts/install/test_install_clone_throttle_fallback.py
#	tests/scripts/install/test_install_lockfile_churn.py
#	tests/scripts/install/test_install_no_initial_commit.py
#	tests/scripts/install/test_install_sh_browser_install.py
#	tests/scripts/install/test_install_sh_node_prerelease.py
#	tests/scripts/install/test_install_sh_symlink_stomp.py
#	tests/scripts/install/test_install_sh_uv_lock_config.py
#	tests/scripts/install/test_install_unmerged_index.py
#	tests/scripts/test_contributor_map.py
#	tests/scripts/test_run_tests_parallel.py
#	tests/skills/test_competitor_news_monitor_skill.py
#	tests/skills/test_document_to_action_items_skill.py
#	tests/skills/test_google_workspace_setup.py
#	tests/skills/test_google_workspace_setup_deps.py
#	tests/skills/test_grounded_citations_skill.py
#	tests/skills/test_ip_as_logo_skill.py
#	tests/skills/test_live_dashboard_skill.py
#	tests/skills/test_mcp_oauth_remote_gateway_skill.py
#	tests/skills/test_office_document_skills.py
#	tests/skills/test_openclaw_migration.py
#	tests/skills/test_product_price_monitor_skill.py
#	tests/skills/test_scrollcraft_skill.py
#	tests/skills/test_setup_wizard_generator_skill.py
#	tests/skills/test_weekly_review_planning_skill.py
#	tests/test_engines_satisfiable.py
#	tests/test_fast_safe_load.py
#	tests/test_hermes_bootstrap.py
#	tests/test_hermes_constants.py
#	tests/test_hermes_logging.py
#	tests/test_managed_runtime_resolution.py
#	tests/test_model_tools_async_bridge.py
#	tests/test_packaging_build_guard.py
#	tests/test_packaging_metadata.py
#	tests/test_yaml_indent_consistency.py
#	tests/tools/test_approval_timeout_overflow.py
#	tests/tools/test_base_environment.py
#	tests/tools/test_bot_mode_dm.py
#	tests/tools/test_browser_chromium_check.py
#	tests/tools/test_browser_hardening.py
#	tests/tools/test_browser_homebrew_paths.py
#	tests/tools/test_browser_npx_warmup.py
#	tests/tools/test_browser_orphan_reaper.py
#	tests/tools/test_browser_real_profile.py
#	tests/tools/test_browser_use_cli.py
#	tests/tools/test_clipboard.py
#	tests/tools/test_code_execution.py
#	tests/tools/test_code_execution_modes.py
#	tests/tools/test_code_execution_windows_env.py
#	tests/tools/test_computer_use.py
#	tests/tools/test_delegate_liveness_timeout.py
#	tests/tools/test_execute_code_approval_cluster.py
#	tests/tools/test_execution_flag_detection.py
#	tests/tools/test_fal_common.py
#	tests/tools/test_file_operations.py
#	tests/tools/test_file_tools.py
#	tests/tools/test_file_tools_cwd_resolution.py
#	tests/tools/test_file_tools_live.py
#	tests/tools/test_lazy_deps.py
#	tests/tools/test_lazy_deps_durable_target.py
#	tests/tools/test_lazy_deps_managed.py
#	tests/tools/test_local_env_blocklist.py
#	tests/tools/test_local_tempdir.py
#	tests/tools/test_macos_protected_search.py
#	tests/tools/test_mcp_npx_cached_bin.py
#	tests/tools/test_oneshot_completion_linger.py
#	tests/tools/test_process_registry.py
#	tests/tools/test_read_file_schema_gating.py
#	tests/tools/test_skill_improvements.py
#	tests/tools/test_skills_sync.py
#	tests/tools/test_termux_api_detection.py
#	tests/tools/test_tirith_security.py
#	tests/tools/test_transcription_tools.py
#	tests/tools/test_tts_streaming.py
#	tests/tools/test_wake_word.py
#	tests/tui_gateway/test_compute_host_borrowed_lease.py
#	tests/tui_gateway/test_compute_host_turn_protocol.py
#	tests/tui_gateway/test_isolated_orphan_activity.py
#	tests/tui_gateway/test_protocol.py
#	tests/tui_gateway/test_slash_worker_profile_home.py
#	tests/tui_gateway/test_subprocess_encoding.py
#	tests/tui_gateway/test_tui_gateway_server.py
#	ui-tui/src/__tests__/terminalParity.test.ts
#	ui-tui/src/__tests__/termuxComposerLayout.test.ts
#	ui-tui/src/__tests__/textInputFastEcho.test.ts
2026-09-23 07:02:44 -04:00

210 lines
9.0 KiB
Python

"""Compression race at the flush chokepoint: a turn writing against a session
already closed by compression must adopt the LIVE continuation tip instead of
dying with ``session_persistence_failed`` and a misleading "full disk" dialog.
The store resolves the continuation chain transitively via the canonical API
``SessionDB.get_compression_tip`` (bounded walk, excludes branch/delegate/tool
children, prefers live children over stale closed siblings). This suite proves
the agent flush path:
* adopts a unique live child (depth-1 case),
* follows a chain of >=2 compressions to the live head — THE regression the
depth-1 ``find_live_compression_child`` API missed (#82001),
* fails closed when no continuation exists (no retry loop),
* fails closed when the resolved tip is itself closed (``ws_orphan_reap``),
* performs the tip lookup exactly once per flush (adoption budget), and
* never renders the failure with the historical full-disk misdiagnosis.
"""
from __future__ import annotations
from pathlib import Path
from types import SimpleNamespace
from hermes_state import SessionDB
from run_agent import AIAgent
def _flush_agent(db, session_id):
"""Bind the real flush methods onto a stand-in over a live SessionDB."""
agent = SimpleNamespace(
_session_db=db,
_session_db_created=True,
_persist_disabled=False,
session_id=session_id,
_session_persist_lock=None,
_flushed_db_message_ids=set(),
_flushed_db_message_session_id=None,
_last_flushed_db_idx=0,
_db_flush_scan_prefix=None,
_persist_user_message_idx=None,
_persist_user_message_override=None,
_persist_user_message_timestamp=None,
_pending_cli_user_message=None,
_active_session_turn_lease_holder=None,
_last_persistence_error_cause=None,
_compression_adoption_failed=False,
)
agent._ensure_db_session = lambda: None
agent._flush_messages_to_session_db = (
AIAgent._flush_messages_to_session_db.__get__(agent, AIAgent)
)
agent._flush_messages_to_session_db_unlocked = (
AIAgent._flush_messages_to_session_db_unlocked.__get__(agent, AIAgent)
)
return agent
def _build_compression_chain(db: SessionDB, chain: list[str]) -> tuple[str, str]:
"""Create ``chain[0] -> ... -> chain[-1]`` where every session except the
last is compression-ended and the last is live. Returns (root, live_head).
"""
for i, sid in enumerate(chain):
parent = chain[i - 1] if i > 0 else None
db.create_session(sid, source="tui", parent_session_id=parent)
if i < len(chain) - 1:
db.end_session(sid, "compression")
return chain[0], chain[-1]
def test_flush_adopts_unique_live_continuation(tmp_path: Path) -> None:
db = SessionDB(db_path=tmp_path / "state.db")
try:
db.create_session("parent", source="tui")
db.append_message("parent", "user", "before split")
db.end_session("parent", "compression")
db.create_session("child", source="tui", parent_session_id="parent")
agent = _flush_agent(db, "parent")
messages = [{"role": "user", "content": "steered after compression"}]
result = agent._flush_messages_to_session_db(messages, [])
assert result is True, "flush must succeed after adopting the continuation"
assert agent.session_id == "child"
durable = db.get_messages_as_conversation("child")
assert any(
m.get("content") == "steered after compression" for m in durable
), "the user message must land in the child session, not be lost"
# The compression-closed parent stays immutable.
parent_rows = db.get_messages_as_conversation("parent")
assert not any(
m.get("content") == "steered after compression" for m in parent_rows
)
assert agent._compression_adoption_failed is False
finally:
db.close()
def test_flush_adopts_live_head_across_compression_chain(tmp_path: Path) -> None:
"""A stale writer behind a chain of >=2 compressions adopts the live head.
This is the exact lineage from #82001 (`root(compressed) -> mid(compressed)
-> tip(live)`) that a depth-1 live-child lookup cannot resolve, because the
direct child is itself already compression-ended.
"""
db = SessionDB(db_path=tmp_path / "state.db")
try:
root, head = _build_compression_chain(db, ["root", "mid", "tip"])
agent = _flush_agent(db, root)
messages = [{"role": "user", "content": "steered after double rotation"}]
result = agent._flush_messages_to_session_db(messages, [])
assert result is True, "flush must succeed by adopting the chain head"
assert agent.session_id == head, "agent must move to the live chain head"
durable = db.get_messages_as_conversation(head)
assert any(
m.get("content") == "steered after double rotation" for m in durable
), "the user message must land in the chain head, not be lost"
finally:
db.close()
def test_flush_fails_closed_when_no_continuation(tmp_path: Path) -> None:
db = SessionDB(db_path=tmp_path / "state.db")
try:
db.create_session("parent", source="tui")
db.append_message("parent", "user", "before split")
db.end_session("parent", "compression")
agent = _flush_agent(db, "parent")
messages = [{"role": "user", "content": "steered after compression"}]
result = agent._flush_messages_to_session_db(messages, [])
assert result is False, "no continuation -> fail closed (never guess)"
assert agent.session_id == "parent", "session id must not change"
assert agent._compression_adoption_failed is True
assert agent._last_persistence_error_cause == "compression_closed"
finally:
db.close()
def test_flush_fails_closed_when_tip_is_stale_closed(tmp_path: Path) -> None:
"""The canonical tip walk may land on a stale closed sibling (e.g.
``ws_orphan_reap``) — a non-live tip must NOT be adopted; fail closed."""
db = SessionDB(db_path=tmp_path / "state.db")
try:
db.create_session("parent", source="tui")
db.append_message("parent", "user", "before split")
db.end_session("parent", "compression")
db.create_session("stale", source="tui", parent_session_id="parent")
db.end_session("stale", "ws_orphan_reap")
agent = _flush_agent(db, "parent")
messages = [{"role": "user", "content": "steered after compression"}]
result = agent._flush_messages_to_session_db(messages, [])
assert result is False, "non-live tip must fail closed (never adopt stale)"
assert agent.session_id == "parent"
assert agent._compression_adoption_failed is True
finally:
db.close()
def test_flush_adopts_exactly_once_no_retry_loop(tmp_path: Path, monkeypatch) -> None:
"""Adoption budget: the tip lookup runs at most once per flush, and a
second closed-parent write after adoption fails closed instead of looping.
"""
from hermes_state_errors import CompressionSessionClosedError
db = SessionDB(db_path=tmp_path / "state.db")
try:
_build_compression_chain(db, ["root", "tip"])
agent = _flush_agent(db, "root")
tip_calls = {"count": 0}
orig_tip = SessionDB.get_compression_tip
def _counting_tip(self, session_id):
tip_calls["count"] += 1
return orig_tip(self, session_id)
monkeypatch.setattr(SessionDB, "get_compression_tip", _counting_tip)
# Every batch write raises closed — including the post-adoption retry
# against the live tip (simulating the tip rotating again mid-flush).
def _always_closed(self, *, session_id, messages, **kwargs):
raise CompressionSessionClosedError(session_id)
monkeypatch.setattr(SessionDB, "append_messages_batch", _always_closed)
messages = [{"role": "user", "content": "steered after compression"}]
result = agent._flush_messages_to_session_db(messages, [])
assert result is False, "second closed-parent write must fail closed"
assert tip_calls["count"] == 1, "tip lookup must happen exactly once"
assert agent._compression_adoption_failed is True
finally:
db.close()
# ---------------------------------------------------------------------------
# Diagnostics: the failure must never read like a disk problem.
# ---------------------------------------------------------------------------
def test_compression_closed_error_classifies_as_compression_closed() -> None:
from hermes_state import classify_persistence_error
from hermes_state_errors import CompressionSessionClosedError, PERSISTENCE_ERROR_CAUSES
cause = classify_persistence_error(CompressionSessionClosedError("session-abc"))
assert cause == "compression_closed"
assert cause in PERSISTENCE_ERROR_CAUSES
# String form (post-RPC wrapping) classifies identically.
assert (
classify_persistence_error(str(CompressionSessionClosedError("session-abc")))
== "compression_closed"
)