feat(webui): add read-only console restart status and impact controls (Closes #667)

This commit is contained in:
2026-07-25 17:23:52 -04:00
parent 76f293eb28
commit 9a01543477
5 changed files with 1368 additions and 0 deletions
+452
View File
@@ -0,0 +1,452 @@
"""Read-only restart console: views, gates, and honesty rules (#667).
The console consumes the #655 substrate. These tests hold it to the three
properties that make a status surface trustworthy:
* an unreadable source is reported unavailable, never rendered as green;
* authorization is probed the way execution would probe it, so an allow is
never shown for something that could not run;
* the surface performs no mutation, including no write to the control-plane DB.
"""
from __future__ import annotations
import os
import sqlite3
import sys
import tempfile
import unittest
from datetime import datetime, timedelta, timezone
from pathlib import Path
sys.path.insert(0, str(Path(__file__).resolve().parent.parent))
from starlette.testclient import TestClient # noqa: E402
import restart_coordinator # noqa: E402
from webui import console_authz, restart_console, restart_views # noqa: E402
from webui.app import create_app # noqa: E402
NOW = datetime(2026, 7, 25, 21, 0, 0, tzinfo=timezone.utc)
def _principal(role: str) -> console_authz.Principal:
return console_authz.Principal(
subject="[email protected]",
role=role,
identity_source=console_authz.IDENTITY_LOCAL_DEV,
authenticated=True,
)
def _inventory(*, complete: bool = True, sessions=(), leases=()):
def _read(**_kwargs):
return {
"sessions": list(sessions),
"leases": list(leases),
"terminal_lock": None,
"prior_recovery_attempts": [],
"inventory_complete": complete,
"incomplete_reasons": (
[] if complete else ["fixture: inventory withheld"]
),
}
return _read
def _live_session(session_id: str = "prgs-author-1234-abcd") -> dict:
return {
"session_id": session_id,
"role": "author",
"profile": "prgs-author",
"pid": os.getpid(),
"status": "active",
"last_heartbeat_at": (NOW - timedelta(seconds=30)).isoformat(),
}
def drain_proof_fixture() -> dict:
"""A structurally complete but unsigned drain proof."""
return {
"version": "drain-proof/v1",
"proof_id": "deadbeef" * 8,
"clean": True,
"issued_at": (NOW - timedelta(minutes=1)).isoformat(),
"expires_at": (NOW + timedelta(minutes=5)).isoformat(),
"requesting_session_id": "s-live",
"impact_fingerprint": "f" * 64,
"checks": [],
"failed_checks": [],
}
class RestartClassMatrixTest(unittest.TestCase):
def test_every_policy_class_is_rendered(self) -> None:
views = restart_console.build_restart_class_views("operator")
self.assertEqual(len(views), len(restart_coordinator.RESTART_CLASS_POLICIES))
def test_viewer_capability_is_role_scoped_not_generic(self) -> None:
"""A worker role must not be shown as able to request a full restart."""
author = {
v.restart_class: v
for v in restart_console.build_restart_class_views("author")
}
operator = {
v.restart_class: v
for v in restart_console.build_restart_class_views("operator")
}
full = restart_coordinator.RestartClass.FULL_MCP_RESTART.value
self.assertFalse(author[full].viewer_may_request)
self.assertFalse(author[full].viewer_may_execute)
self.assertTrue(operator[full].viewer_may_request)
self.assertTrue(operator[full].viewer_may_execute)
def test_unknown_role_may_do_nothing(self) -> None:
views = restart_console.build_restart_class_views("not-a-role")
self.assertTrue(all(not v.viewer_may_request for v in views))
self.assertTrue(all(not v.viewer_may_execute for v in views))
class AuthorizationProbeTest(unittest.TestCase):
def test_probe_asks_for_execution_so_phase_gate_is_reported(self) -> None:
"""An admin clears the role bar and still cannot execute in Phase 1.
This is the case that distinguishes the two probes. Asked without
``for_execution`` an admin is *allowed* for ``system.restart_namespace``,
which on a control surface reads as a live button. Asked the way
execution asks, the same principal is refused ``phase_not_active``. The
console must report the second answer.
"""
by_id = {
a.action_id: a
for a in restart_console.build_action_authorizations(
_principal(console_authz.ADMIN)
)
}
restart = by_id["system.restart_namespace"]
self.assertFalse(restart.execution_enabled)
self.assertEqual(restart.reason_code, console_authz.DENY_PHASE_NOT_ACTIVE)
permissive = console_authz.authorize(
"system.restart_namespace", _principal(console_authz.ADMIN)
)
self.assertTrue(
permissive.allowed,
"guard precondition: without for_execution an admin is allowed, "
"which is exactly why the console must not probe that way",
)
def test_operator_is_refused_the_admin_only_restart_action(self) -> None:
"""Role refusal precedes the phase gate and is reported as such."""
by_id = {
a.action_id: a
for a in restart_console.build_action_authorizations(
_principal(console_authz.OPERATOR)
)
}
self.assertEqual(
by_id["system.restart_namespace"].reason_code,
console_authz.DENY_INSUFFICIENT_ROLE,
)
def test_anonymous_is_denied_unauthenticated(self) -> None:
by_id = {
a.action_id: a for a in restart_console.build_action_authorizations(None)
}
self.assertEqual(
by_id["system.restart_namespace"].reason_code,
console_authz.DENY_UNAUTHENTICATED,
)
def test_no_authorization_ever_reports_execution_enabled(self) -> None:
for role in (
console_authz.VIEWER,
console_authz.OPERATOR,
console_authz.CONTROLLER,
console_authz.ADMIN,
):
for auth in restart_console.build_action_authorizations(_principal(role)):
self.assertFalse(
auth.execution_enabled,
f"{role} reported execution_enabled for {auth.action_id}",
)
class ImpactPreviewTest(unittest.TestCase):
def test_impact_renders_from_coordinator_dto(self) -> None:
impact, source = restart_console.load_impact_report(
principal=_principal(console_authz.OPERATOR),
read_inventory=_inventory(sessions=[_live_session()]),
now=NOW,
)
self.assertTrue(source.available)
self.assertIsNotNone(impact)
self.assertEqual(
impact["restart_class"],
restart_coordinator.RestartClass.FULL_MCP_RESTART.value,
)
self.assertIn("verdict", impact)
self.assertFalse(impact["restart_performed"])
self.assertTrue(impact["dry_run"])
def test_incomplete_inventory_is_surfaced_and_denies(self) -> None:
impact, source = restart_console.load_impact_report(
principal=_principal(console_authz.OPERATOR),
read_inventory=_inventory(complete=False),
now=NOW,
)
self.assertFalse(impact["inventory_complete"])
self.assertFalse(impact["allow_restart"])
self.assertTrue(source.detail, "incomplete inventory must explain itself")
def test_inventory_reader_failure_is_unavailable_not_empty(self) -> None:
"""A reader that raises must not be rendered as 'no sessions affected'."""
def _boom(**_kwargs):
raise RuntimeError("control-plane unreachable")
impact, source = restart_console.load_impact_report(
principal=_principal(console_authz.OPERATOR),
read_inventory=_boom,
now=NOW,
)
self.assertIsNone(impact)
self.assertFalse(source.available)
self.assertIn("control-plane unreachable", source.detail)
class ControlPlaneReadTest(unittest.TestCase):
def test_missing_database_is_incomplete_not_empty(self) -> None:
inventory = restart_console.read_control_plane_inventory(
db_path="/nonexistent/control-plane.sqlite3"
)
self.assertFalse(inventory["inventory_complete"])
self.assertEqual(inventory["sessions"], [])
self.assertTrue(inventory["incomplete_reasons"])
def test_reader_never_creates_the_database(self) -> None:
"""Reading status must not bring a control-plane DB into existence.
The path deliberately sits in a directory that already exists: a
read-write ``sqlite3.connect`` would happily create the file there, so
this fails if the reader ever stops opening the database ``mode=ro``.
A nested-missing-directory path would pass for the wrong reason,
because sqlite cannot create the parent directory either way.
"""
with tempfile.TemporaryDirectory() as tmp:
path = os.path.join(tmp, "control_plane.sqlite3")
self.assertTrue(os.path.isdir(os.path.dirname(path)))
inventory = restart_console.read_control_plane_inventory(db_path=path)
self.assertFalse(
os.path.exists(path),
"reading restart status created a control-plane database",
)
self.assertFalse(inventory["inventory_complete"])
def test_reads_active_sessions_from_a_real_database(self) -> None:
with tempfile.TemporaryDirectory() as tmp:
path = os.path.join(tmp, "cp.sqlite3")
conn = sqlite3.connect(path)
conn.execute(
"CREATE TABLE sessions (session_id TEXT, role TEXT, profile TEXT,"
" pid INTEGER, status TEXT, last_heartbeat_at TEXT)"
)
conn.execute(
"CREATE TABLE work_items (work_item_id INTEGER, kind TEXT,"
" number INTEGER)"
)
conn.execute(
"CREATE TABLE leases (lease_id TEXT, session_id TEXT, role TEXT,"
" phase TEXT, status TEXT, worktree_path TEXT,"
" work_item_id INTEGER, expires_at TEXT)"
)
conn.execute(
"INSERT INTO sessions VALUES (?,?,?,?,?,?)",
("s-live", "author", "prgs-author", 4242, "active", NOW.isoformat()),
)
conn.execute(
"INSERT INTO sessions VALUES (?,?,?,?,?,?)",
("s-done", "author", "prgs-author", 11, "closed", NOW.isoformat()),
)
conn.execute("INSERT INTO work_items VALUES (1, 'issue', 667)")
conn.execute(
"INSERT INTO leases VALUES (?,?,?,?,?,?,?,?)",
(
"l-1",
"s-live",
"author",
"allocated",
"active",
None,
1,
NOW.isoformat(),
),
)
conn.commit()
conn.close()
inventory = restart_console.read_control_plane_inventory(db_path=path)
self.assertTrue(inventory["inventory_complete"])
self.assertEqual([s["session_id"] for s in inventory["sessions"]], ["s-live"])
self.assertEqual(inventory["leases"][0]["work_number"], 667)
class DrainAndReconcileTest(unittest.TestCase):
def test_absent_drain_proof_is_not_a_pass(self) -> None:
drain, source = restart_console.load_drain_status(proof=None, now=NOW)
self.assertIsNone(drain)
self.assertFalse(source.available)
self.assertIn("denies", source.detail)
def test_tampered_drain_proof_is_reported_invalid(self) -> None:
proof = drain_proof_fixture()
proof["clean"] = True
proof["proof_id"] = "0" * 64
drain, source = restart_console.load_drain_status(proof=proof, now=NOW)
self.assertTrue(source.available)
self.assertFalse(drain["valid"])
def test_absent_reconcile_proof_is_unavailable(self) -> None:
reconcile, source = restart_console.load_reconcile_status(load_proof=None)
self.assertIsNone(reconcile)
self.assertFalse(source.available)
def test_reconcile_proof_is_rendered_when_supplied(self) -> None:
payload = {
"overall_status": "degraded",
"mode": "log_only",
"resolved_count": 3,
"unresolved_count": 2,
"items": [
{
"dimension": "leases",
"status": "unresolved",
"summary": "2 orphaned leases",
"follow_up_required": True,
}
],
}
reconcile, source = restart_console.load_reconcile_status(
load_proof=lambda: payload
)
self.assertTrue(source.available)
self.assertEqual(reconcile["unresolved_count"], 2)
class RenderingTest(unittest.TestCase):
def _snapshot(self, **kwargs):
params = {
"principal": _principal(console_authz.OPERATOR),
"read_inventory": _inventory(sessions=[_live_session()]),
"now": NOW,
}
params.update(kwargs)
return restart_console.load_restart_console_snapshot(**params)
def test_page_renders_every_section(self) -> None:
html = restart_views.render_restart_console_page(self._snapshot())
for heading in (
"Impact preview",
"Drain proof",
"Post-restart reconcile",
"Restart classes",
"Approval controls",
"Break-glass",
):
self.assertIn(heading, html)
def test_hostile_session_id_is_escaped(self) -> None:
hostile = "<script>alert('x')</script>"
html = restart_views.render_restart_console_page(
self._snapshot(read_inventory=_inventory(sessions=[_live_session(hostile)]))
)
self.assertNotIn("<script>alert", html)
self.assertIn("&lt;script&gt;", html)
def test_unavailable_impact_says_unsafe_rather_than_clean(self) -> None:
def _boom(**_kwargs):
raise RuntimeError("nope")
snapshot = self._snapshot(read_inventory=_boom)
html = restart_views.render_restart_console_page(snapshot)
self.assertIn("blast radius of a restart is unknown", html)
self.assertIn("unavailable", html)
def test_break_glass_is_hidden_from_unprivileged_viewers(self) -> None:
viewer_html = restart_views.render_restart_console_page(
self._snapshot(principal=_principal(console_authz.VIEWER))
)
self.assertIn("visible to operator-class", viewer_html)
self.assertNotIn(
f"#{restart_console.BREAK_GLASS_ISSUE}", viewer_html
)
def test_break_glass_shown_to_operator_is_marked_unavailable(self) -> None:
html = restart_views.render_restart_console_page(self._snapshot())
self.assertIn("unavailable", html)
self.assertIn(f"#{restart_console.BREAK_GLASS_ISSUE}", html)
def test_snapshot_always_declares_itself_read_only(self) -> None:
self.assertTrue(self._snapshot().read_only)
class RestartConsoleRouteTest(unittest.TestCase):
def setUp(self) -> None:
self.client = TestClient(create_app())
def test_page_route_renders(self) -> None:
res = self.client.get("/runtime/restart")
self.assertEqual(res.status_code, 200)
self.assertIn("Restart status and impact", res.text)
def test_api_route_exports_snapshot(self) -> None:
res = self.client.get("/api/v1/system/restart/status")
self.assertEqual(res.status_code, 200)
payload = res.json()
self.assertTrue(payload["read_only"])
self.assertEqual(payload["links"]["issue"], 667)
self.assertEqual(
len(payload["restart_classes"]),
len(restart_coordinator.RESTART_CLASS_POLICIES),
)
def test_restart_class_is_selectable(self) -> None:
res = self.client.get(
"/api/v1/system/restart/status?restart_class=client_reconnect"
)
self.assertEqual(res.status_code, 200)
self.assertEqual(res.json()["impact"]["restart_class"], "client_reconnect")
def test_unknown_restart_class_fails_closed(self) -> None:
res = self.client.get(
"/api/v1/system/restart/status?restart_class=obliterate-everything"
)
self.assertEqual(res.status_code, 200)
impact = res.json()["impact"]
self.assertFalse(impact["allow_restart"])
def test_anonymous_api_reader_gets_no_execution_grant(self) -> None:
payload = self.client.get("/api/v1/system/restart/status").json()
self.assertFalse(payload["break_glass"]["available"])
for auth in payload["authorizations"]:
self.assertFalse(auth["execution_enabled"])
def test_route_is_registered_in_nav(self) -> None:
from webui.nav import nav_hrefs
self.assertIn("/runtime/restart", nav_hrefs())
def test_no_write_method_is_exposed(self) -> None:
"""The surface is read-only: nothing accepts a POST."""
for path in ("/runtime/restart", "/api/v1/system/restart/status"):
self.assertEqual(self.client.post(path).status_code, 405, path)
if __name__ == "__main__":
unittest.main()
+37
View File
@@ -47,6 +47,9 @@ from webui.traffic_views import render_traffic_page
from webui.worktree_scanner import load_hygiene_snapshot, snapshot_to_dict as worktree_snapshot_to_dict from webui.worktree_scanner import load_hygiene_snapshot, snapshot_to_dict as worktree_snapshot_to_dict
from webui.worktree_views import render_worktrees_page from webui.worktree_views import render_worktrees_page
from webui.runtime_health import load_runtime_snapshot, snapshot_to_dict as runtime_snapshot_to_dict from webui.runtime_health import load_runtime_snapshot, snapshot_to_dict as runtime_snapshot_to_dict
import restart_coordinator
from webui.restart_console import load_restart_console_snapshot
from webui.restart_views import render_restart_console_page
from webui.runtime_views import render_runtime_page from webui.runtime_views import render_runtime_page
from webui.session_loader import ( from webui.session_loader import (
load_session_view_snapshot, load_session_view_snapshot,
@@ -331,6 +334,33 @@ async def api_runtime(_request: Request) -> JSONResponse:
return JSONResponse(runtime_snapshot_to_dict(load_runtime_snapshot())) return JSONResponse(runtime_snapshot_to_dict(load_runtime_snapshot()))
def _restart_console_snapshot(request: Request):
"""Build the read-only restart snapshot for the requesting principal (#667)."""
principal = resolve_principal(request.headers)
restart_class = (
request.query_params.get("restart_class")
or restart_coordinator.RestartClass.FULL_MCP_RESTART.value
)
return load_restart_console_snapshot(
principal=principal, restart_class=restart_class
)
async def restart_console_page(request: Request) -> HTMLResponse:
"""Restart status, impact preview, and approval state (#667). Read-only."""
snapshot = _restart_console_snapshot(request)
return HTMLResponse(
render_page(
title="Restart", body_html=render_restart_console_page(snapshot)
)
)
async def api_restart_status(request: Request) -> JSONResponse:
"""JSON export of the read-only restart console snapshot (#667)."""
return JSONResponse(_restart_console_snapshot(request).as_dict())
async def sessions(_request: Request) -> HTMLResponse: async def sessions(_request: Request) -> HTMLResponse:
"""Runtime and session view (#641) — read-only composition of health + inventory.""" """Runtime and session view (#641) — read-only composition of health + inventory."""
snapshot = load_session_view_snapshot() snapshot = load_session_view_snapshot()
@@ -781,6 +811,13 @@ def create_app(*, bind_host: str | None = None) -> Starlette:
Route("/api/prompts", api_prompts, methods=["GET"]), Route("/api/prompts", api_prompts, methods=["GET"]),
Route("/runtime", runtime, methods=["GET"]), Route("/runtime", runtime, methods=["GET"]),
Route("/api/runtime", api_runtime, methods=["GET"]), Route("/api/runtime", api_runtime, methods=["GET"]),
# #667 read-only restart status / impact preview / approval state.
Route("/runtime/restart", restart_console_page, methods=["GET"]),
Route(
"/api/v1/system/restart/status",
api_restart_status,
methods=["GET"],
),
Route("/sessions", sessions, methods=["GET"]), Route("/sessions", sessions, methods=["GET"]),
Route("/api/sessions", api_sessions, methods=["GET"]), Route("/api/sessions", api_sessions, methods=["GET"]),
Route("/api/v1/sessions", api_sessions, methods=["GET"]), Route("/api/v1/sessions", api_sessions, methods=["GET"]),
+1
View File
@@ -48,6 +48,7 @@ NAV_GROUPS: tuple[NavGroup, ...] = (
)), )),
NavGroup("Runtime/Sessions", ( NavGroup("Runtime/Sessions", (
NavItem("/runtime", "Runtime health"), NavItem("/runtime", "Runtime health"),
NavItem("/runtime/restart", "Restart status"),
NavItem("/sessions", "Sessions"), NavItem("/sessions", "Sessions"),
)), )),
NavGroup("Projects", ( NavGroup("Projects", (
+579
View File
@@ -0,0 +1,579 @@
"""Read-only restart status, impact preview, and approval state (#667).
Phase 1 of the console restart surface. It *consumes* the #655 coordinator
substrate and renders it; it never restarts, reloads, drains, approves, or kills
anything. There is no apply path in this module, so there is no execution gate
here to arm incorrectly — the only writes the console could perform are the ones
it does not implement.
Sources, each independently fail-soft and each reported with its own
:class:`SourceStatus`:
* :mod:`restart_coordinator` — restart-class policy matrix (#663) and the
blast-radius impact report (#658).
* :mod:`drain_proof` — drain checklist and gate verdict (#661), verified
read-only against a caller-supplied proof.
* :mod:`post_restart_reconcile` — post-restart completion proof (#662).
* :mod:`webui.console_authz` — role authorization for the approval controls
(#633).
Three rules this module holds itself to, because a status surface that lies is
worse than one that is absent:
**A source that could not be read is reported unavailable, never green.** No
default, placeholder, or self-comparison is substituted for a reading that
failed. An unreadable control-plane DB yields ``inventory_complete=False``,
which the coordinator itself turns into a fail-closed verdict.
**Authorization is asked the way execution would ask it.** Every authorization
probe passes ``for_execution=True``, so the console reports whether the action
could actually run rather than the weaker "this principal is the right role".
While the console is in Phase 1 that answer is ``phase_not_active`` for every
phase-2 action, and the surface says so plainly instead of showing an allow.
**The database is opened read-only.** ``ControlPlaneDB()`` creates directories
and runs migrations on construction, which is a write; this module opens the
sqlite file with ``mode=ro`` exactly as :mod:`webui.inventory` does, and treats
a missing file as missing authority rather than an empty inventory.
"""
from __future__ import annotations
import os
import sqlite3
from dataclasses import dataclass, field
from datetime import datetime, timezone
from typing import Any, Callable, Mapping
import control_plane_db
import drain_proof
import restart_coordinator
from webui import console_authz
from webui.inventory import redact_path, scrub
# --- Source status ----------------------------------------------------------
STATUS_OK = "ok"
STATUS_UNAVAILABLE = "unavailable"
#: Console actions whose authorization state this surface reports. Both are
#: pre-existing #642 actions; this module adds no new console action because it
#: performs no console action.
REPORTED_ACTIONS: tuple[str, ...] = (
"system.restart_namespace",
"system.reload_namespace",
)
#: The break-glass workflow (#664) is not consumed here. It is declared so the
#: surface is honest about the gap rather than silently omitting a governance
#: path the operator has been told exists.
BREAK_GLASS_ISSUE = 664
BREAK_GLASS_PENDING_REASON = (
"The break-glass workflow (#664) is not yet available on this branch's "
"base; no break-glass control is offered and none is implied."
)
@dataclass(frozen=True)
class SourceStatus:
"""Whether one backing source could be read, and why not when it could not."""
name: str
status: str
detail: str = ""
@property
def available(self) -> bool:
return self.status == STATUS_OK
def as_dict(self) -> dict[str, Any]:
return {
"name": self.name,
"status": self.status,
"available": self.available,
"detail": self.detail,
}
@dataclass(frozen=True)
class RestartClassView:
"""One row of the #663 restart-class matrix, scoped to the viewer's role."""
restart_class: str
required_permission: str
expected_blast_radius: str
drain_requirement: str
full_drain_required: bool
approval_requirement: str
request_roles: tuple[str, ...]
execution_roles: tuple[str, ...]
viewer_may_request: bool
viewer_may_execute: bool
def as_dict(self) -> dict[str, Any]:
return {
"restart_class": self.restart_class,
"required_permission": self.required_permission,
"expected_blast_radius": self.expected_blast_radius,
"drain_requirement": self.drain_requirement,
"full_drain_required": self.full_drain_required,
"approval_requirement": self.approval_requirement,
"request_roles": list(self.request_roles),
"execution_roles": list(self.execution_roles),
"viewer_may_request": self.viewer_may_request,
"viewer_may_execute": self.viewer_may_execute,
}
@dataclass(frozen=True)
class ActionAuthorization:
"""Authorization state for one console action, asked as execution would."""
action_id: str
summary: str
required_role: str
allowed: bool
execution_enabled: bool
reason_code: str
detail: str
def as_dict(self) -> dict[str, Any]:
return {
"action_id": self.action_id,
"summary": self.summary,
"required_role": self.required_role,
"allowed": self.allowed,
"execution_enabled": self.execution_enabled,
"reason_code": self.reason_code,
"detail": self.detail,
}
@dataclass(frozen=True)
class BreakGlassSurface:
"""Declared-but-unavailable break-glass panel (#664 is not on this base)."""
available: bool
issue: int
reason: str
viewer_is_privileged: bool
def as_dict(self) -> dict[str, Any]:
return {
"available": self.available,
"issue": self.issue,
"reason": self.reason,
"viewer_is_privileged": self.viewer_is_privileged,
}
@dataclass(frozen=True)
class RestartConsoleSnapshot:
"""Everything the read-only restart console renders."""
generated_at: str
viewer_role: str
viewer_authenticated: bool
read_only: bool
impact: dict[str, Any] | None
impact_source: SourceStatus
drain: dict[str, Any] | None
drain_source: SourceStatus
reconcile: dict[str, Any] | None
reconcile_source: SourceStatus
restart_classes: tuple[RestartClassView, ...]
authorizations: tuple[ActionAuthorization, ...]
break_glass: BreakGlassSurface
notes: tuple[str, ...] = field(default_factory=tuple)
def as_dict(self) -> dict[str, Any]:
return {
"generated_at": self.generated_at,
"viewer_role": self.viewer_role,
"viewer_authenticated": self.viewer_authenticated,
"read_only": self.read_only,
"impact": self.impact,
"impact_source": self.impact_source.as_dict(),
"drain": self.drain,
"drain_source": self.drain_source.as_dict(),
"reconcile": self.reconcile,
"reconcile_source": self.reconcile_source.as_dict(),
"restart_classes": [c.as_dict() for c in self.restart_classes],
"authorizations": [a.as_dict() for a in self.authorizations],
"break_glass": self.break_glass.as_dict(),
"notes": list(self.notes),
"links": {
"issue": 667,
"extends": 642,
"umbrella": 655,
"coordinator": 658,
"drain_proof": 661,
"reconcile": 662,
"restart_classes": 663,
"break_glass": BREAK_GLASS_ISSUE,
"vision": 652,
"roadmap": 653,
},
}
def _utc_now() -> datetime:
return datetime.now(timezone.utc)
# --- Control-plane inventory (read-only) ------------------------------------
def read_control_plane_inventory(
*,
db_path: str | None = None,
limit: int = 200,
) -> dict[str, Any]:
"""Read sessions and leases for an impact evaluation, read-only.
Returns the inventory mapping
:func:`restart_coordinator.evaluate_restart_impact` expects.
``inventory_complete`` is True only when every read succeeded, so a partial
read denies rather than under-reporting the blast radius.
The database is never created, migrated, or written: a missing file means
the console has no session authority, which is not the same as there being
no sessions.
"""
path = (db_path or control_plane_db.default_db_path() or "").strip()
incomplete: list[str] = []
def _incomplete(reason: str) -> dict[str, Any]:
return {
"sessions": [],
"leases": [],
"terminal_lock": None,
"prior_recovery_attempts": [],
"inventory_complete": False,
"incomplete_reasons": [reason],
}
if not path:
return _incomplete("control-plane database path is not configured")
if not os.path.exists(path):
return _incomplete(
f"control-plane database not present at {redact_path(path)}; "
"no session or lease authority available"
)
try:
conn = sqlite3.connect(f"file:{path}?mode=ro", uri=True, timeout=5)
conn.row_factory = sqlite3.Row
except sqlite3.Error as exc:
return _incomplete(f"control-plane database could not be opened: {exc}")
sessions: list[dict[str, Any]] = []
leases: list[dict[str, Any]] = []
capped = max(1, int(limit))
try:
tables = {
str(row[0])
for row in conn.execute(
"SELECT name FROM sqlite_master WHERE type = 'table'"
).fetchall()
}
if "sessions" not in tables:
incomplete.append("control-plane database has no sessions table")
else:
sessions = [
dict(row)
for row in conn.execute(
"SELECT session_id, role, profile, pid, status,"
" last_heartbeat_at FROM sessions"
" WHERE status = 'active'"
" ORDER BY last_heartbeat_at DESC LIMIT ?",
(capped,),
).fetchall()
]
if "leases" not in tables:
incomplete.append("control-plane database has no leases table")
elif "work_items" not in tables:
incomplete.append(
"control-plane database has no work_items table; lease work "
"identity cannot be resolved"
)
else:
leases = [
dict(row)
for row in conn.execute(
"SELECT l.lease_id, l.session_id, l.role, l.phase,"
" l.status AS freshness, l.worktree_path,"
" w.kind AS work_kind, w.number AS work_number"
" FROM leases l"
" JOIN work_items w ON w.work_item_id = l.work_item_id"
" WHERE l.status = 'active'"
" ORDER BY l.expires_at DESC LIMIT ?",
(capped,),
).fetchall()
]
except sqlite3.Error as exc:
return _incomplete(f"control-plane database read failed: {exc}")
finally:
conn.close()
return {
"sessions": sessions,
"leases": leases,
"terminal_lock": None,
"prior_recovery_attempts": [],
"inventory_complete": not incomplete,
"incomplete_reasons": incomplete,
}
# --- Composition ------------------------------------------------------------
def build_restart_class_views(viewer_role: str | None) -> tuple[RestartClassView, ...]:
"""Render the #663 class matrix, marking what this viewer may request."""
normalized = str(viewer_role or "").strip().lower()
views: list[RestartClassView] = []
for policy in restart_coordinator.RESTART_CLASS_POLICIES.values():
views.append(
RestartClassView(
restart_class=policy.restart_class.value,
required_permission=policy.required_permission,
expected_blast_radius=policy.expected_blast_radius,
drain_requirement=policy.drain_requirement,
full_drain_required=policy.full_drain_required,
approval_requirement=policy.approval_requirement,
request_roles=tuple(policy.request_roles),
execution_roles=tuple(policy.execution_roles),
viewer_may_request=normalized in policy.request_roles,
viewer_may_execute=normalized in policy.execution_roles,
)
)
return tuple(views)
def build_action_authorizations(
principal: console_authz.Principal | None,
) -> tuple[ActionAuthorization, ...]:
"""Authorization state for the approval controls, asked as execution.
``for_execution=True`` is deliberate. Asking without it answers "is this
principal senior enough", which is not the question an operator looking at a
control needs answered; asking with it answers "would this run", and while
the console is in Phase 1 the honest answer is no.
"""
results: list[ActionAuthorization] = []
for action_id in REPORTED_ACTIONS:
action = console_authz.get_action(action_id)
decision = console_authz.authorize(action_id, principal, for_execution=True)
results.append(
ActionAuthorization(
action_id=action_id,
summary=action.summary if action else "",
required_role=(
action.minimum_role if action else console_authz.OPERATOR
),
allowed=bool(decision.allowed),
execution_enabled=bool(decision.execution_enabled),
reason_code=str(decision.reason_code or ""),
detail=str(decision.detail or ""),
)
)
return tuple(results)
def viewer_is_privileged(principal: console_authz.Principal | None) -> bool:
"""True when the viewer holds at least the operator role."""
who = principal if principal is not None else console_authz.ANONYMOUS
if not who.authenticated:
return False
return who.rank >= console_authz.ROLE_ORDER.index(console_authz.OPERATOR)
def load_impact_report(
*,
principal: console_authz.Principal | None = None,
restart_class: str = restart_coordinator.RestartClass.FULL_MCP_RESTART.value,
db_path: str | None = None,
limit: int = 200,
read_inventory: Callable[..., Mapping[str, Any]] | None = None,
now: datetime | None = None,
) -> tuple[dict[str, Any] | None, SourceStatus]:
"""Evaluate the blast radius for *restart_class*, always dry-run."""
reader = read_inventory or read_control_plane_inventory
try:
inventory = dict(reader(db_path=db_path, limit=limit))
except Exception as exc: # noqa: BLE001
return None, SourceStatus(
"impact",
STATUS_UNAVAILABLE,
f"control-plane inventory failed: {type(exc).__name__}: {exc}",
)
who = principal if principal is not None else console_authz.ANONYMOUS
viewer_role = str(who.role or "").strip().lower()
try:
report = restart_coordinator.evaluate_restart_impact(
inventory,
now=now,
dry_run=True,
restart_class=restart_class,
requester_role=viewer_role,
requester_permissions=restart_coordinator.permissions_for_role(
viewer_role
),
)
except Exception as exc: # noqa: BLE001
return None, SourceStatus(
"impact",
STATUS_UNAVAILABLE,
f"impact evaluation failed: {type(exc).__name__}: {exc}",
)
payload = scrub(report.as_dict())
detail = ""
if not report.inventory_complete:
detail = "; ".join(report.incomplete_reasons) or "inventory incomplete"
return payload, SourceStatus("impact", STATUS_OK, detail)
def load_drain_status(
*,
proof: Mapping[str, Any] | None = None,
now: datetime | None = None,
expected_impact_fingerprint: str | None = None,
) -> tuple[dict[str, Any] | None, SourceStatus]:
"""Verify a supplied drain proof read-only and report the verdict.
No proof supplied is not a failure and not a pass: it is reported as the
absence of a proof, which is exactly what the #661 gate would deny on.
"""
if proof is None:
return None, SourceStatus(
"drain",
STATUS_UNAVAILABLE,
"no drain proof supplied; the #661 gate denies a restart without a "
"valid unexpired clean proof",
)
try:
verified = drain_proof.verify_drain_proof(
proof,
now=now,
expected_impact_fingerprint=expected_impact_fingerprint,
)
except Exception as exc: # noqa: BLE001
return None, SourceStatus(
"drain",
STATUS_UNAVAILABLE,
f"drain proof verification failed: {type(exc).__name__}: {exc}",
)
return scrub(verified.as_dict()), SourceStatus("drain", STATUS_OK)
def load_reconcile_status(
*,
load_proof: Callable[[], Any] | None = None,
) -> tuple[dict[str, Any] | None, SourceStatus]:
"""Report the most recent post-restart completion proof (#662)."""
if load_proof is None:
return None, SourceStatus(
"reconcile",
STATUS_UNAVAILABLE,
"no post-restart completion proof source is wired into this view",
)
try:
proof = load_proof()
except Exception as exc: # noqa: BLE001
return None, SourceStatus(
"reconcile",
STATUS_UNAVAILABLE,
f"reconcile proof unavailable: {type(exc).__name__}: {exc}",
)
if proof is None:
return None, SourceStatus(
"reconcile",
STATUS_UNAVAILABLE,
"no post-restart reconcile has been recorded",
)
payload = proof.as_dict() if hasattr(proof, "as_dict") else dict(proof)
return scrub(payload), SourceStatus("reconcile", STATUS_OK)
def load_restart_console_snapshot(
*,
principal: console_authz.Principal | None = None,
restart_class: str = restart_coordinator.RestartClass.FULL_MCP_RESTART.value,
db_path: str | None = None,
limit: int = 200,
drain_proof_payload: Mapping[str, Any] | None = None,
read_inventory: Callable[..., Mapping[str, Any]] | None = None,
load_reconcile_proof: Callable[[], Any] | None = None,
now: datetime | None = None,
) -> RestartConsoleSnapshot:
"""Compose the read-only restart console snapshot."""
who = principal if principal is not None else console_authz.ANONYMOUS
moment = now or _utc_now()
impact, impact_source = load_impact_report(
principal=who,
restart_class=restart_class,
db_path=db_path,
limit=limit,
read_inventory=read_inventory,
now=moment,
)
fingerprint = None
if impact is not None:
try:
fingerprint = drain_proof.impact_fingerprint(impact)
except Exception: # noqa: BLE001
fingerprint = None
drain, drain_source = load_drain_status(
proof=drain_proof_payload,
now=moment,
expected_impact_fingerprint=fingerprint,
)
reconcile, reconcile_source = load_reconcile_status(
load_proof=load_reconcile_proof
)
notes: list[str] = [
"This surface is read-only: it evaluates and displays, and performs no "
"restart, reload, drain, approval, or process action.",
]
if not impact_source.available:
notes.append(
"Impact preview unavailable — a restart decision must not be made "
"from this page while the blast radius is unknown."
)
return RestartConsoleSnapshot(
generated_at=moment.isoformat(),
viewer_role=str(who.role or "anonymous"),
viewer_authenticated=bool(who.authenticated),
read_only=True,
impact=impact,
impact_source=impact_source,
drain=drain,
drain_source=drain_source,
reconcile=reconcile,
reconcile_source=reconcile_source,
restart_classes=build_restart_class_views(who.role),
authorizations=build_action_authorizations(who),
break_glass=BreakGlassSurface(
available=False,
issue=BREAK_GLASS_ISSUE,
reason=BREAK_GLASS_PENDING_REASON,
viewer_is_privileged=viewer_is_privileged(who),
),
notes=tuple(notes),
)
+299
View File
@@ -0,0 +1,299 @@
"""HTML views for the read-only restart console (#667).
Every interpolated value passes through :func:`_esc`. Values that can carry a
filesystem path or free-form operator text additionally pass through
:func:`webui.inventory.scrub_text`, which redacts credential-shaped tokens
*inside* a string rather than only at its start.
The page renders state and never offers a control that would mutate anything:
the approval and break-glass panels report authorization and availability, and
there is no form, button, or endpoint behind them.
"""
from __future__ import annotations
import html
from webui.inventory import scrub_text
from webui.restart_console import RestartConsoleSnapshot, SourceStatus
def _esc(value: object) -> str:
"""Escape any value for HTML text or a quoted attribute."""
if value is None:
return ""
return html.escape(str(value), quote=True)
def _esc_text(value: object) -> str:
"""Escape free-form text after redacting secrets embedded inside it."""
if value is None:
return ""
return _esc(scrub_text(str(value)))
def _bool_badge(
value: bool, *, true_label: str = "yes", false_label: str = "no"
) -> str:
css = "badge-ok" if value else "badge-blocked"
label = true_label if value else false_label
return f'<span class="badge {css}">{_esc(label)}</span>'
def _source_badge(source: SourceStatus) -> str:
css = "badge-ok" if source.available else "badge-blocked"
badge = f'<span class="badge {css}">{_esc(source.status)}</span>'
if source.detail:
badge += f' <span class="muted">{_esc_text(source.detail)}</span>'
return badge
def _notes_block(snapshot: RestartConsoleSnapshot) -> str:
if not snapshot.notes:
return ""
items = "".join(f"<li>{_esc_text(note)}</li>" for note in snapshot.notes)
return f"<ul class='reasons'>{items}</ul>"
def _impact_section(snapshot: RestartConsoleSnapshot) -> str:
head = (
"<section class='health-card'>"
f"<h3>Impact preview {_source_badge(snapshot.impact_source)}</h3>"
)
impact = snapshot.impact
if impact is None:
return (
head
+ "<p class='muted'>No impact preview is available, so the blast "
"radius of a restart is unknown. Treat this as unsafe.</p></section>"
)
counts = impact.get("counts") or {}
verdict = str(impact.get("verdict") or "unknown")
verdict_css = "badge-ok" if verdict == "safe" else "badge-blocked"
rows = "".join(
f"<tr><th>{_esc(key.replace('_', ' '))}</th><td>{_esc(value)}</td></tr>"
for key, value in sorted(counts.items())
)
reasons = "".join(
f"<li>{_esc_text(reason)}</li>" for reason in (impact.get("reasons") or [])
)
incomplete = ""
if not impact.get("inventory_complete", False):
detail = "; ".join(str(r) for r in (impact.get("incomplete_reasons") or []))
incomplete = (
"<p class='error'><strong>Inventory incomplete:</strong> "
f"{_esc_text(detail or 'unspecified')}. The coordinator fails "
"closed on an incomplete inventory.</p>"
)
sessions = impact.get("affected_sessions") or []
session_rows = "".join(
"<tr>"
f"<td><code>{_esc(s.get('session_id'))}</code></td>"
f"<td>{_esc(s.get('role'))}</td>"
f"<td>{_esc(s.get('pid'))}</td>"
f"<td>{_bool_badge(bool(s.get('live')), true_label='live', false_label='idle')}</td>"
f"<td>{_bool_badge(not s.get('heartbeat_stale'), true_label='fresh', false_label='stale')}</td>"
"</tr>"
for s in sessions[:50]
)
session_table = (
"<h4>Sessions a restart would terminate</h4>"
"<div class='table-scroll'><table class='registry'><thead><tr>"
"<th>Session</th><th>Role</th><th>PID</th><th>State</th>"
"<th>Heartbeat</th></tr></thead><tbody>"
f"{session_rows}</tbody></table></div>"
if session_rows
else "<p class='muted'>No affected sessions reported.</p>"
)
truncated = (
f"<p class='muted'>Showing the first 50 of {_esc(len(sessions))} "
"affected sessions.</p>"
if len(sessions) > 50
else ""
)
return (
head
+ "<p class='health-headline'>Verdict "
f"<span class='badge {verdict_css}'>{_esc(verdict)}</span> · "
f"blast radius <code>{_esc(impact.get('blast_radius'))}</code> · "
f"class <code>{_esc(impact.get('restart_class'))}</code></p>"
+ incomplete
+ (f"<ul class='reasons'>{reasons}</ul>" if reasons else "")
+ (f"<table class='registry'><tbody>{rows}</tbody></table>" if rows else "")
+ session_table
+ truncated
+ "</section>"
)
def _drain_section(snapshot: RestartConsoleSnapshot) -> str:
head = (
"<section class='health-card'>"
f"<h3>Drain proof {_source_badge(snapshot.drain_source)}</h3>"
)
drain = snapshot.drain
if drain is None:
return (
head
+ "<p class='muted'>No drain proof has been presented to this view. "
"The #661 gate authorizes a restart only against a valid, unexpired, "
"clean proof, so the absence of one is a denial, not a pass.</p>"
"</section>"
)
reasons = "".join(
f"<li>{_esc_text(reason)}</li>" for reason in (drain.get("reasons") or [])
)
return (
head
+ "<table class='registry'><tbody>"
f"<tr><th>Valid</th><td>{_bool_badge(bool(drain.get('valid')))}</td></tr>"
f"<tr><th>Clean</th><td>{_bool_badge(bool(drain.get('clean')))}</td></tr>"
f"<tr><th>Expired</th><td>{_bool_badge(not drain.get('expired'), true_label='no', false_label='yes')}</td></tr>"
f"<tr><th>Tampered</th><td>{_bool_badge(not drain.get('tampered'), true_label='no', false_label='yes')}</td></tr>"
f"<tr><th>Proof id</th><td><code>{_esc(drain.get('proof_id'))}</code></td></tr>"
"</tbody></table>"
+ (f"<ul class='reasons'>{reasons}</ul>" if reasons else "")
+ "</section>"
)
def _reconcile_section(snapshot: RestartConsoleSnapshot) -> str:
head = (
"<section class='health-card'>"
f"<h3>Post-restart reconcile {_source_badge(snapshot.reconcile_source)}</h3>"
)
proof = snapshot.reconcile
if proof is None:
return (
head
+ "<p class='muted'>No post-restart completion proof is recorded. "
"Until one is, the last restart's recovery state is unproven.</p>"
"</section>"
)
items = "".join(
"<tr>"
f"<td>{_esc(item.get('dimension'))}</td>"
f"<td>{_esc(item.get('status'))}</td>"
f"<td>{_esc_text(item.get('summary'))}</td>"
f"<td>{_bool_badge(not item.get('follow_up_required'), true_label='no', false_label='yes')}</td>"
"</tr>"
for item in (proof.get("items") or [])
)
return (
head
+ "<p class='health-headline'>Status "
f"<code>{_esc(proof.get('overall_status'))}</code> · mode "
f"<code>{_esc(proof.get('mode'))}</code> · resolved "
f"{_esc(proof.get('resolved_count'))} · unresolved "
f"{_esc(proof.get('unresolved_count'))}</p>"
+ (
"<div class='table-scroll'><table class='registry'><thead><tr>"
"<th>Dimension</th><th>Status</th><th>Summary</th>"
"<th>Follow-up required</th></tr></thead><tbody>"
f"{items}</tbody></table></div>"
if items
else "<p class='muted'>No reconcile dimensions reported.</p>"
)
+ "</section>"
)
def _class_matrix_section(snapshot: RestartConsoleSnapshot) -> str:
rows = "".join(
"<tr>"
f"<td><code>{_esc(view.restart_class)}</code></td>"
f"<td><code>{_esc(view.required_permission)}</code></td>"
f"<td>{_esc(view.expected_blast_radius)}</td>"
f"<td>{_esc(view.drain_requirement)}</td>"
f"<td>{_esc(view.approval_requirement)}</td>"
f"<td>{_bool_badge(view.viewer_may_request)}</td>"
f"<td>{_bool_badge(view.viewer_may_execute)}</td>"
"</tr>"
for view in snapshot.restart_classes
)
return (
"<section class='health-card'>"
"<h3>Restart classes</h3>"
"<p class='muted'>The least-privilege matrix each restart request is "
"resolved against. &ldquo;You may request&rdquo; and &ldquo;you may "
"execute&rdquo; are computed for the current viewer role, not for a "
"generic operator.</p>"
"<div class='table-scroll'><table class='registry'><thead><tr>"
"<th>Class</th><th>Permission</th><th>Blast radius</th>"
"<th>Drain</th><th>Approval</th><th>You may request</th>"
"<th>You may execute</th></tr></thead><tbody>"
f"{rows}</tbody></table></div>"
"</section>"
)
def _approval_section(snapshot: RestartConsoleSnapshot) -> str:
rows = "".join(
"<tr>"
f"<td><code>{_esc(a.action_id)}</code></td>"
f"<td>{_esc(a.required_role)}</td>"
f"<td>{_bool_badge(a.allowed)}</td>"
f"<td>{_bool_badge(a.execution_enabled)}</td>"
f"<td><code>{_esc(a.reason_code)}</code></td>"
f"<td>{_esc_text(a.detail)}</td>"
"</tr>"
for a in snapshot.authorizations
)
return (
"<section class='health-card'>"
"<h3>Approval controls</h3>"
"<p class='muted'>Authorization is probed the way execution would probe "
"it, so &ldquo;execution enabled&rdquo; answers whether the action would "
"actually run — not merely whether this role outranks the requirement. "
"No control on this page performs the action.</p>"
"<div class='table-scroll'><table class='registry'><thead><tr>"
"<th>Action</th><th>Required role</th><th>Authorized</th>"
"<th>Execution enabled</th><th>Reason</th><th>Detail</th>"
"</tr></thead><tbody>"
f"{rows}</tbody></table></div>"
"</section>"
)
def _break_glass_section(snapshot: RestartConsoleSnapshot) -> str:
bg = snapshot.break_glass
if not bg.viewer_is_privileged:
return (
"<section class='health-card'>"
"<h3>Break-glass</h3>"
"<p class='muted'>Break-glass status is visible to operator-class "
"roles only. Your role does not carry that authority, so no "
"emergency surface is shown.</p>"
"</section>"
)
return (
"<section class='health-card'>"
"<h3>Break-glass "
f"{_bool_badge(bg.available, true_label='available', false_label='unavailable')}"
"</h3>"
f"<p class='muted'>{_esc_text(bg.reason)}</p>"
f"<p class='meta'>Tracked by issue #{_esc(bg.issue)}.</p>"
"</section>"
)
def render_restart_console_page(snapshot: RestartConsoleSnapshot) -> str:
"""Render the whole read-only restart console body."""
return (
"<h2>Restart status and impact</h2>"
f"<p class='meta'>Generated <code>{_esc(snapshot.generated_at)}</code> · "
f"viewer role <code>{_esc(snapshot.viewer_role)}</code> · "
f"authenticated {_bool_badge(snapshot.viewer_authenticated)} · "
f"read-only {_bool_badge(snapshot.read_only)}</p>"
+ _notes_block(snapshot)
+ _impact_section(snapshot)
+ _drain_section(snapshot)
+ _reconcile_section(snapshot)
+ _class_matrix_section(snapshot)
+ _approval_section(snapshot)
+ _break_glass_section(snapshot)
)