Fix replica-path leaks and availability semantics from live testing

- _remote_headers and /check used struct convenience properties that
  read the global database; they now use the SessionContext / the handed
  store (also fixes Remote-Credential carrying a struct repr instead of
  the UUID).
- Replica availability: TTL clock starts at disconnect, not at last
  message or failed reconnect; tight WS keepalive for prompt dead-peer
  detection.
- Proxy preserves repeated Set-Cookie via raw headers; sync endpoint
  does its own accept (wsutil decorator pre-accepts) and bypasses host
  dispatch (server-to-server; satellite may use an out-of-domain
  address).
- Admin-credential bootstrap warning skips remote domains.

Verified live with two instances (remote :4501, satellite :4402):
replica snapshot + events, 204 forward with Remote-* in <1ms, validate
write-behind landing on the remote, proxied logout with instant local
eviction, 503 after cache_ttl of disconnect, resync after remote
restart.

paskia/bootstrap.py

@@ -67,7 +67,10 @@ async def check_admin_credentials() -> bool:
         # Check first admin user for credentials on any configured domain
         admin_user = admin_users[0]
         reg = domains.registry()
-        configured = sorted(d.rp_id for d in reg.domains)
+        # Remote domains hold their credentials on the remote instance
+        configured = sorted(d.rp_id for d in reg.domains if d.remote is None)
+        if not configured:
+            return False
 
         if not any(admin_user.credential_ids_for(rp_id) for rp_id in configured):
             # Admin exists but has no credential on any domain

paskia/fastapi/api.py

@@ -189,13 +189,13 @@ async def check_user(
     data = _store(request)
     try:
         u = data.users[user_uuid]
-        role = u.role
-        org = role.org
+        role = data.roles[u.role_uuid]
+        org = data.orgs[role.org_uuid]
     except KeyError:
         raise HTTPException(status_code=404, detail="User not found")
 
     host = hostutil.normalize_host(request.headers.get("host"))
-    org_perm_uuids = {p.uuid for p in org.permissions}
+    org_perm_uuids = {p.uuid for p in data.permissions.values() if org.uuid in p.orgs}
 
     effective_perms = []
     for perm_uuid in role.permission_set:
@@ -236,7 +236,7 @@ def _remote_headers(ctx) -> dict[str, str]:
         "Remote-Session-Expires": (
             (ctx.session.validated + EXPIRES).isoformat().replace("+00:00", "Z")
         ),
-        "Remote-Credential": str(ctx.session.credential),
+        "Remote-Credential": str(ctx.credential.uuid),
     }
 
 

paskia/fastapi/dispatch.py

@@ -63,6 +63,11 @@ class DispatchMiddleware:
         host = _header(scope, "host")
         host_domain = registry.resolve(host)
         if host_domain is None:
+            # The sync endpoint is server-to-server and token-gated: the
+            # satellite may reach us via an address outside our domains.
+            if scope.get("path") == "/auth/api/sync/ws" and registry.domains:
+                await self._dispatch(scope, receive, send, registry.domains[0])
+                return
             await send({"type": "websocket.close", "code": _WS_CLOSE_POLICY_VIOLATION})
             return
 

paskia/fastapi/proxy.py

@@ -50,13 +50,11 @@ async def proxy_to_remote(request: Request, remote: RemoteConfig) -> Response:
         content=await request.body(),
         headers=headers,
     )
-    response_headers = {
-        k: v
-        for k, v in upstream.headers.multi_items()
-        if k.lower() not in _SKIP_RESPONSE_HEADERS
-    }
-    return Response(
-        content=upstream.content,
-        status_code=upstream.status_code,
-        headers=response_headers,
-    )
+    response = Response(content=upstream.content, status_code=upstream.status_code)
+    # Raw headers to preserve repeated Set-Cookie
+    response.raw_headers = [
+        (k, v)
+        for k, v in upstream.headers.raw
+        if k.decode().lower() not in _SKIP_RESPONSE_HEADERS
+    ]
+    return response

paskia/fastapi/sync.py

@@ -12,7 +12,6 @@ import msgspec
 from fastapi import FastAPI, WebSocket, WebSocketDisconnect
 
 from paskia import db, syncfeed
-from paskia.fastapi.wsutil import websocket_error_handler
 
 _logger = logging.getLogger(__name__)
 
@@ -64,7 +63,6 @@ async def _apply_client_message(message: dict) -> None:
 
 
 @app.websocket("/ws")
-@websocket_error_handler
 async def sync_websocket(ws: WebSocket):
     tokens = syncfeed.tokens_from_env()
     auth = ws.headers.get("authorization", "")
@@ -77,33 +75,40 @@ async def sync_websocket(ws: WebSocket):
     feed = syncfeed.feed
     await _send(ws, {"type": "hello", "generation": feed.generation, "seq": feed.seq})
 
-    # The client always speaks first: resume request (possibly null fields)
-    resume = msgspec.json.decode(await ws.receive_bytes())
-    queue = feed.subscribe()
     try:
-        replay = None
-        if (
-            resume.get("type") == "resume"
-            and resume.get("generation") == feed.generation
-            and isinstance(resume.get("seq"), int)
-        ):
-            replay = feed.replay_since(resume["seq"])
-        if replay is not None:
-            for event in replay:
-                await _send(ws, event)
-        else:
-            for chunk in _snapshot_messages():
-                await _send(ws, chunk)
-        await _send(ws, {"type": "ready", "seq": feed.seq})
-
-        sender = asyncio.create_task(_pump(ws, queue))
+        # The client always speaks first: resume request (possibly null fields)
+        resume = msgspec.json.decode(await ws.receive_bytes())
+        queue = feed.subscribe()
         try:
-            while True:
-                await _apply_client_message(msgspec.json.decode(await ws.receive_bytes()))
+            replay = None
+            if (
+                resume.get("type") == "resume"
+                and resume.get("generation") == feed.generation
+                and isinstance(resume.get("seq"), int)
+            ):
+                replay = feed.replay_since(resume["seq"])
+            if replay is not None:
+                for event in replay:
+                    await _send(ws, event)
+            else:
+                for chunk in _snapshot_messages():
+                    await _send(ws, chunk)
+            await _send(ws, {"type": "ready", "seq": feed.seq})
+
+            sender = asyncio.create_task(_pump(ws, queue))
+            try:
+                while True:
+                    await _apply_client_message(
+                        msgspec.json.decode(await ws.receive_bytes())
+                    )
+            finally:
+                sender.cancel()
         finally:
-            sender.cancel()
-    finally:
-        feed.unsubscribe(queue)
+            feed.unsubscribe(queue)
+    except WebSocketDisconnect:
+        pass
+    except Exception:
+        _logger.exception("Sync WebSocket failed")
 
 
 async def _pump(ws: WebSocket, queue: asyncio.Queue) -> None:

paskia/satellite.py

@@ -12,6 +12,7 @@ cache_ttl for fail-open behavior bounded by session expiry).
 """
 
 import asyncio
+import contextlib
 import logging
 import time
 from datetime import UTC, datetime
@@ -73,6 +74,7 @@ class RemoteReplica:
         self.generation: str | None = None
         self.seq = 0
         self.last_contact = 0.0  # monotonic time of last snapshot/event
+        self.connected = False
         self._pending_refresh: dict[str, dict] = {}
         self._refresh_signal = asyncio.Event()
         self._task: asyncio.Task | None = None
@@ -80,9 +82,16 @@ class RemoteReplica:
         self._stopped = True
 
     def available(self) -> bool:
-        return (
-            self.last_contact > 0
-            and time.monotonic() - self.last_contact <= self.remote.cache_ttl
+        """Synced and either connected now or within cache_ttl of silence.
+
+        The websockets keepalive drops a wedged connection, so a live
+        connection means events arrive within one round trip; after losing
+        it the replica remains trusted for cache_ttl.
+        """
+        if not self.last_contact:
+            return False
+        return self.connected or (
+            time.monotonic() - self.last_contact <= self.remote.cache_ttl
         )
 
     def refresh_session(self, key: str, validated, ip: str, user_agent: str) -> None:
@@ -114,7 +123,7 @@ class RemoteReplica:
         for task in (self._task, self._sweeper):
             if task:
                 task.cancel()
-                with asyncio.suppress(asyncio.CancelledError):
+                with contextlib.suppress(asyncio.CancelledError):
                     await task
 
     async def _sweep(self) -> None:
@@ -132,6 +141,11 @@ class RemoteReplica:
                 raise
             except Exception as e:
                 _logger.info("Sync to %s failed: %s", self.remote.url, e)
+            if self.connected:
+                # The TTL clock starts when the feed goes down, not at the
+                # last message — an idle connection is healthy.
+                self.connected = False
+                self.last_contact = time.monotonic()
             if not self._stopped:
                 await asyncio.sleep(_RECONNECT_DELAY)
 
@@ -143,7 +157,11 @@ class RemoteReplica:
         )
         resume = {} if full_resync else {"generation": self.generation, "seq": self.seq}
         async with websockets.connect(
-            ws_url, additional_headers={"Authorization": f"Bearer {self.remote.token}"}
+            ws_url,
+            additional_headers={"Authorization": f"Bearer {self.remote.token}"},
+            # Prompt dead-peer detection: availability semantics count on it
+            ping_interval=5,
+            ping_timeout=5,
         ) as ws:
             hello = msgspec.json.decode(await ws.recv())
             if hello.get("type") != "hello":
@@ -182,9 +200,10 @@ class RemoteReplica:
                             attach_stores()
                         self.generation = hello["generation"]
                         self.seq = message["seq"]
+                        self.connected = True
             finally:
                 sender.cancel()
-                with asyncio.suppress(asyncio.CancelledError):
+                with contextlib.suppress(asyncio.CancelledError):
                     await sender
 
     async def _send_loop(self, ws) -> None: