fix: recover client control connection safely

Dockerfile

@@ -7,6 +7,7 @@ RUN pip wheel --no-cache-dir --wheel-dir /wheels .
 FROM builder AS test
 RUN pip install --no-cache-dir /wheels/*.whl
 COPY tests ./tests
+COPY scripts ./scripts
 CMD ["python", "-m", "unittest", "discover", "-s", "tests", "-v"]
 
 FROM python:3.11-slim

README.md

@@ -77,6 +77,13 @@ four eager-mesh routes, transfer/eviction/unarchive workflows, and the
 adversarial matrix without live infrastructure or Telegram. See
 `e2e/README.md`.
 
+## Deployment Preflight Tests
+
+Check deployment configuration immediately before bringing a new node online.
+The reusable read-only preflight validates the deployed image's filesystem and
+API view against its Docker bind mounts. See details in
+[docs/deployment-preflight.md](docs/deployment-preflight.md).
+
 The complete cross-project design, protocol, workflow, safety, deployment,
 Telegram UX, and testing documentation is published under [`docs/`](docs/).
 

deploy/production/README.md

@@ -1,12 +1,15 @@
 # Production deployment
 
 These files are the non-secret, host-specific deployment manifests for the
-initial x1/x2/lithium topology.
+initial x1/x2/lithium topology and the later helium archive node.
 
 - Install the x1 and x2 files as
   `~/compose/ArchiveControl-cache/{compose.yaml,client.toml}`.
 - Install the lithium files as
   `~/compose/ArchiveControl-archive/{compose.yaml,client.toml}`.
+- The helium directory contains three isolated compose-project manifests. Install
+  them under `~/Repositories/compose/{qbittorrent-helium,syncthing-helium,ArchiveControl-archive}`
+  as described in `helium/README.md`.
 - Create sibling `state`, `backups`, and `secrets` directories owned by the
   configured container UID/GID.
 - Secret files are never committed. Each `secrets` directory contains
@@ -22,10 +25,17 @@ have different mount IDs and `link(2)` may return `EXDEV` across them. The
 client deliberately treats that case as copy-only and performs a full payload
 free-space check.
 
-When a Syncthing route is physically nested below the qB root, mount the qB
-root once and map the exact Syncthing API folder through it:
+Automatic routes always use the Syncthing API path `routes/<route-id>`. Bind
+that path from a dedicated directory beneath the qB data root, then map the
+same path through the qB client mount:
 
 ```yaml
+# Syncthing compose project
+volumes:
+  - /srv/syncthing-config:/var/syncthing
+  - /srv/downloads/.archive-control-routes:/var/syncthing/routes
+
+# Archive Control client compose project
 volumes:
   - /srv/downloads:/data/qb
   - /srv/syncthing-config:/data/sync
@@ -35,23 +45,30 @@ volumes:
 [syncthing]
 api_root = "/var/syncthing"
 local_root = "/data/sync"
-local_path_overrides = { "/var/syncthing/Downloads/Sync" = "/data/qb/Sync" }
+local_path_overrides = { "/var/syncthing/routes" = "/data/qb/.archive-control-routes" }
 ```
 
-Do **not** additionally mount `/srv/downloads/Sync` at a path beneath
-`/data/sync`. The override is the authoritative mapping for that folder and
-keeps qB source files and staging destinations in one mount namespace. Use
-the folder's normalized API-visible **path** as the override key (for example,
-Syncthing `~/Downloads/Sync` becomes `/var/syncthing/Downloads/Sync`); do not
-use the folder ID.
+Do **not** mount the route directory separately into the client. The override
+is the authoritative mapping and keeps qB source files and automatic route
+folders in one mount namespace. Existing manually configured Syncthing folders
+below the qB tree may retain their own exact overrides. Use API-visible paths,
+not Syncthing folder IDs, as override keys.
 
-Before starting a stack, validate it with:
+Before starting a stack, validate its config and then run the generic host
+preflight with that machine's own paths and container names:
 
 ```sh
 docker compose config
 docker compose run --rm archive-client --check-config
+python3 scripts/preflight-deployment.py --help
 ```
 
 The check is fail-fast and performs local permission, filesystem, sparse-file,
 hard-link, and reflink probes. Normal startup additionally probes the local
 qBittorrent and Syncthing APIs before registration.
+
+`scripts/preflight-deployment.py` is deliberately topology-agnostic: pass the
+host's `client.toml` and its client, qBittorrent, and Syncthing container names.
+It resolves Docker mounts rather than assuming project names or host paths. It
+also catches a common fatal error: a Syncthing `api_root` that names its config
+volume rather than the mounted shared-data tree.

deploy/production/helium/README.md

@@ -0,0 +1,34 @@
+# Helium archive-node deployment
+
+This directory contains the non-secret manifests for the dedicated helium
+archive node. Install the files into these separate compose projects:
+
+- `~/Repositories/compose/qbittorrent-helium/compose.yaml`
+- `~/Repositories/compose/syncthing-helium/compose.yaml`
+- `~/Repositories/compose/ArchiveControl-archive/{compose.yaml,client.toml}`
+
+The qBittorrent API and Syncthing GUI/API bind only to loopback. Syncthing
+transport/discovery ports are published normally. qBittorrent data and
+Syncthing route folders both live under `/media/Data2`; the Archive Control
+container therefore mounts `/media/Data2` once at `/data/storage` so archive
+placements can hardlink rather than make a full copy.
+
+`/media/Data2/Downloading` already contains unrelated files. The fresh qB
+instance must not import or manage them; only Archive Control-created torrents
+are managed.
+
+Before the first client start, run the repository preflight on the target host
+(replace container names if that host uses different compose project names):
+
+```sh
+python3 scripts/preflight-deployment.py \
+  --client-config ~/Repositories/compose/ArchiveControl-archive/client.toml \
+  --client-container archive-control-helium-archive-client-1 \
+  --syncthing-container helium-syncthing-syncthing-1 \
+  --qbittorrent-container helium-qbittorrent-qbittorrent-1
+```
+
+It is read-only: it checks the configured API roots resolve to the same host
+paths as the client roots, that both data roots share one client bind mount for
+hardlinks, secret-file permissions, filesystem capabilities, and the local qB
+and Syncthing APIs. It does not send the control token or modify any container.

deploy/production/helium/archive-control-compose.yaml

@@ -0,0 +1,19 @@
+name: archive-control-helium
+
+services:
+  archive-client:
+    image: sodium/archive-clients:v0.1.19
+    user: "1000:1000"
+    restart: unless-stopped
+    network_mode: host
+    command: ["--config", "/etc/archive-control/client.toml"]
+    environment:
+      QB_USER: admin
+    volumes:
+      - ./client.toml:/etc/archive-control/client.toml:ro
+      - ./secrets:/run/secrets:ro
+      - ./state:/var/lib/archive-control
+      - ./backups:/var/backups/archive-control
+      # Do not split this into separate qB and Sync bind mounts: hardlinks
+      # require both trees to resolve within this one mount namespace.
+      - /media/Data2:/data/storage

deploy/production/helium/client.toml

@@ -0,0 +1,42 @@
+client_id = "helium-archive"
+display_name = "helium archive"
+role = "archive"
+control_endpoint = "ws://bot.everdream.xyz:8766/archive_control"
+shared_token_file = "/run/secrets/archive_control_token"
+state_db = "/var/lib/archive-control/client.db"
+backup_dir = "/var/backups/archive-control"
+
+[connection]
+registration_timeout = "10s"
+heartbeat_interval = "15s"
+offline_timeout = "45s"
+reconnect_initial = "1s"
+reconnect_max = "60s"
+reconnect_reset_after = "60s"
+reconnect_jitter = true
+
+[jobs]
+stall_after = "30m"
+verification_timeout = "30m"
+poll_interval = "1s"
+free_space_reserve_bytes = 33554432
+
+[backup]
+interval = "6h"
+recent = 12
+daily = 14
+weekly = 8
+
+[qbittorrent]
+endpoint = "http://127.0.0.1:8081"
+username = "${QB_USER}"
+password_file = "/run/secrets/qb_password"
+api_root = "/downloads/Downloading"
+local_root = "/data/storage/Downloading"
+
+[syncthing]
+endpoint = "http://127.0.0.1:8384"
+api_key_file = "/run/secrets/syncthing_api_key"
+api_root = "/var/syncthing/Sync"
+local_root = "/data/storage/Sync"
+advertised_addresses = ["dynamic"]

deploy/production/helium/qbittorrent-compose.yaml

@@ -0,0 +1,24 @@
+name: helium-qbittorrent
+
+services:
+  qbittorrent:
+    image: sodium/qbittorrent-nox:5.2.3-lt2-1-proxy-listen-amd64
+    environment:
+      PUID: "1000"
+      PGID: "1000"
+      TZ: Etc/UTC
+      WEBUI_PORT: "8081"
+      TORRENTING_PORT: "6881"
+    command: ["--webui-port=8081", "--confirm-legal-notice"]
+    volumes:
+      - ./config:/config
+      # This is deliberately one parent mount. Archive Control maps both the
+      # qB and Syncthing subtrees through the same mount namespace.
+      - /media/Data2:/downloads
+    ports:
+      # Keep the published and internal WebUI ports identical: qBittorrent
+      # rejects a forwarded request whose Host header carries a different port.
+      - "127.0.0.1:8081:8081"
+      - "6881:6881/tcp"
+      - "6881:6881/udp"
+    restart: unless-stopped

deploy/production/helium/start-syncthing.sh

@@ -0,0 +1,18 @@
+#!/bin/sh
+set -eu
+
+home=/var/syncthing
+api_key=$(cat /run/secrets/syncthing_api_key)
+
+if [ ! -f "$home/config.xml" ]; then
+    syncthing generate --home="$home" --no-port-probing
+fi
+
+exec syncthing serve \
+    --home="$home" \
+    --gui-address=http://0.0.0.0:8384 \
+    --gui-apikey="$api_key" \
+    --no-browser \
+    --no-port-probing \
+    --no-restart \
+    --no-upgrade

deploy/production/helium/syncthing-compose.yaml

@@ -0,0 +1,19 @@
+name: helium-syncthing
+
+services:
+  syncthing:
+    image: syncthing/syncthing:2.1.2@sha256:4464f4161dd0251e20d46bb3aec83363db75d80cef1abdd5d5fd4054b04a004d
+    user: "1000:1000"
+    hostname: helium
+    entrypoint: ["/bootstrap/start-syncthing.sh"]
+    volumes:
+      - ./start-syncthing.sh:/bootstrap/start-syncthing.sh:ro
+      - ./secrets:/run/secrets:ro
+      - ./config:/var/syncthing
+      - /media/Data2/Sync:/var/syncthing/Sync
+    ports:
+      - "127.0.0.1:8384:8384"
+      - "22000:22000/tcp"
+      - "22000:22000/udp"
+      - "21027:21027/udp"
+    restart: unless-stopped

deploy/production/lithium/client.toml

@@ -39,4 +39,5 @@ endpoint = "http://syncthing:8384"
 api_key_file = "/run/secrets/syncthing_api_key"
 api_root = "/var/syncthing"
 local_root = "/data/sync"
+local_path_overrides = { "/var/syncthing/routes" = "/data/qb/.archive-control-routes" }
 advertised_addresses = ["dynamic"]

deploy/production/x1/client.toml

@@ -41,5 +41,5 @@ api_root = "/var/syncthing"
 local_root = "/data/sync"
 # This existing folder is physically inside the qB data tree. Map it through
 # that same client bind mount so source staging can use hardlinks.
-local_path_overrides = { "/var/syncthing/DownloadsSync" = "/data/qb/Sync" }
+local_path_overrides = { "/var/syncthing/DownloadsSync" = "/data/qb/Sync", "/var/syncthing/routes" = "/data/qb/.archive-control-routes" }
 advertised_addresses = ["dynamic"]

deploy/production/x2/client.toml

@@ -42,5 +42,5 @@ local_root = "/data/sync"
 # `DownloadsSync-X2` is ~/Downloads/Sync on the host, nested below the qB
 # root. Resolve it through /data/qb rather than a second nested bind mount so
 # source staging can hardlink it.
-local_path_overrides = { "/var/syncthing/Downloads/Sync" = "/data/qb/Sync" }
+local_path_overrides = { "/var/syncthing/Downloads/Sync" = "/data/qb/Sync", "/var/syncthing/routes" = "/data/qb/.archive-control-routes" }
 advertised_addresses = ["dynamic"]

docs/README.md

@@ -16,8 +16,9 @@ credentials and host-specific details.
 6. [Storage safety and recovery](storage-safety.md)
 7. [qBittorrent and Syncthing integration](service-apis.md)
 8. [Configuration, deployment, and usage](deployment-and-usage.md)
-9. [Telegram UX](telegram-ux.md)
-10. [Testing strategy](testing.md)
+9. [Deployment preflight tests](deployment-preflight.md)
+10. [Telegram UX](telegram-ux.md)
+11. [Testing strategy](testing.md)
 
 The independent protobuf source of truth lives in the
 `cabbage/archive-control-proto` repository. Generated bindings in this

docs/deployment-and-usage.md

@@ -95,6 +95,7 @@ backup_dir = "/var/backups/archive-control"
 registration_timeout = "10s" # First response must arrive within this window.
 heartbeat_interval = "15s" # Server may negotiate a different effective value.
 offline_timeout = "45s"
+outbound_enqueue_timeout = "15s" # Reconnect rather than wedging if sends stop draining.
 reconnect_initial = "1s"
 reconnect_max = "60s" # Retry forever, never wait longer than this.
 reconnect_reset_after = "60s"

docs/deployment-preflight.md

@@ -0,0 +1,121 @@
+# Deployment preflight tests
+
+Run the deployment preflight on every new node after its qBittorrent and
+Syncthing stacks are running, but before starting the Archive Control daemon
+for normal operation or creating any routes/jobs. It does not contact the
+control daemon, alter Syncthing/qBittorrent state, or print secret contents.
+It does create and remove two unique, zero-byte probe files while proving that
+the live client mount topology permits hard-link staging; no resource data or
+configuration is modified.
+
+The script is [`scripts/preflight-deployment.py`](../scripts/preflight-deployment.py).
+It deliberately takes paths and container names as arguments rather than
+assuming hostnames, compose project names, mount paths, credentials, or roles.
+It can therefore be used unchanged on cache and archive nodes.
+
+## Requirements
+
+- Run it on the Docker host being checked.
+- Use Python 3.8 or newer. The script reads the mounted configuration through
+  the client image, so it does not depend on the host Python TOML library.
+- Have Docker CLI access to the daemon.
+- Have the checkout containing the script available on that host, or copy only
+  this script into the node's compose directory.
+- Bring up qBittorrent and Syncthing first. Do not start the normal Archive
+  Control daemon yet.
+
+The disposable client container below uses the exact `archive-client` service
+image, mounts, environment, and network specified by that node's compose file.
+Consequently its qB/Syncthing API probes validate the same runtime environment
+the daemon will use, not the host shell's network namespace.
+
+## New-node procedure
+
+The example uses generic names. Replace paths, compose project directories,
+service names, and real container names for the node being deployed.
+
+```sh
+# 1. Bring up only the local dependencies.
+cd /srv/compose/syncthing-node && docker compose up -d
+cd /srv/compose/qbittorrent-node && docker compose up -d
+
+# 2. Create a temporary client from the exact production image/config, without
+#    running the daemon or registering it with control.
+cd /srv/compose/ArchiveControl-archive
+docker compose run -d --no-deps --name archive-control-preflight \
+  --entrypoint sleep archive-client infinity
+
+# 3. Discover the real dependency container names if needed.
+docker ps --format '{{.Names}}'
+
+# 4. Run the deployment checks. They include a disposable hard-link probe.
+python3 /path/to/archive-clients/scripts/preflight-deployment.py \
+  --client-config /srv/compose/ArchiveControl-archive/client.toml \
+  --client-container archive-control-preflight \
+  --syncthing-container syncthing-node-syncthing-1 \
+  --qbittorrent-container qbittorrent-node-qbittorrent-1
+
+# 5. Remove the temporary container, then start the real daemon only after a
+#    successful result.
+docker rm -f archive-control-preflight
+docker compose up -d archive-client
+```
+
+If the real client configuration is mounted somewhere other than the standard
+`/etc/archive-control/client.toml`, pass that in-container location explicitly:
+
+```sh
+python3 scripts/preflight-deployment.py ... \
+  --container-config /custom/path/client.toml
+```
+
+## Required arguments
+
+| Argument | Meaning |
+| --- | --- |
+| `--client-config` | Host path to the exact `client.toml` that will be mounted into the daemon. |
+| `--client-container` | Running disposable or already-running client container to inspect and probe. |
+| `--syncthing-container` | Running Syncthing container for this node. |
+| `--qbittorrent-container` | Running qBittorrent container for this node. |
+| `--container-config` | Optional `client.toml` path inside the client container; defaults to `/etc/archive-control/client.toml`. |
+
+## What it checks
+
+- The future automatic `routes/<route-id>` path resolves through Docker mounts
+  to the same host path as its client-side mapping.
+- That future route path and qBittorrent content root use one client bind
+  mount, so hard-link staging remains possible rather than silently falling
+  back to a space-consuming copy.
+- A real `link(2)` operation between a unique zero-byte file in the qB root
+  and one in the future automatic-route root. The probe verifies that both
+  names refer to the same inode and removes them unconditionally.
+- Every existing `archive-control:*` Syncthing folder beneath `routes/` still
+  has its local directory. This catches a route-root bind-mount migration that
+  would otherwise hide an already configured folder and later fail a job with
+  `No such file or directory`.
+- The token, qB password, and Syncthing API-key files are non-empty regular
+  files with no group/world permissions.
+- The client image can read its configuration and reports usable permissions,
+  sparse-file support, and filesystem capabilities for both roots.
+- qBittorrent authentication/version compatibility and Syncthing
+  authentication/device identity are healthy from the client container.
+
+In particular, if `syncthing.api_root` is the Syncthing configuration root,
+bind its `routes` subdirectory from a dedicated directory beneath qB's data
+root and add a `local_path_overrides` mapping for that exact API path. This
+prevents automatic route folders being created on a small configuration
+filesystem while the client expects to hardlink from the qB data mount.
+
+When converting an existing node, stop its client and Syncthing containers,
+move each existing `routes/<route-id>` directory from the old Syncthing config
+tree into the new qB-backed route-root directory, then recreate Syncthing and
+run this preflight. Do not discard existing route directories: they can contain
+route handshake state or an in-progress transfer namespace.
+
+## Failure handling
+
+Treat a nonzero exit status as a deployment blocker. Correct the compose
+mounts, `client.toml`, secret permissions, service credentials, or service
+image/version that the message identifies, then rerun the same command. Do not
+work around a mapping failure by creating route folders manually: the route
+provisioner relies on those API/local path mappings being exact.

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "archive-clients"
-version = "0.1.18"
+version = "0.1.19"
 requires-python = ">=3.11"
 dependencies = ["protobuf==7.35.1", "websockets==16.0"]
 

scripts/preflight-deployment.py

@@ -0,0 +1,327 @@
+#!/usr/bin/env python3
+"""Validate a host's Archive Control Docker deployment before it is used.
+
+This intentionally relies only on the config file and the named containers on
+the host where it runs. It never reads secret contents or calls a remote
+control daemon. The hard-link probe creates uniquely named, empty files in
+the configured qB and automatic-route directories and removes them afterward.
+"""
+
+from __future__ import annotations
+
+import argparse
+import json
+import os
+import stat
+import subprocess
+import sys
+from dataclasses import dataclass
+from pathlib import Path, PurePosixPath
+from typing import Any
+
+
+class CheckFailure(RuntimeError):
+    pass
+
+
+@dataclass(frozen=True)
+class Mapping:
+    source: Path
+    destination: PurePosixPath
+
+
+def docker_inspect(container: str) -> dict[str, Any]:
+    result = subprocess.run(
+        ["docker", "inspect", container], check=False, text=True,
+        capture_output=True,
+    )
+    if result.returncode:
+        raise CheckFailure(f"cannot inspect container {container!r}")
+    try:
+        value = json.loads(result.stdout)
+        return value[0]
+    except (json.JSONDecodeError, IndexError, TypeError) as exc:
+        raise CheckFailure(f"invalid Docker inspection for {container!r}") from exc
+
+
+def map_path(container: dict[str, Any], path: str) -> Mapping:
+    candidate = PurePosixPath(path)
+    matches: list[tuple[int, Mapping]] = []
+    for mount in container.get("Mounts", []):
+        if mount.get("Type") not in {"bind", "volume"}:
+            continue
+        source, destination = mount.get("Source"), mount.get("Destination")
+        if not isinstance(source, str) or not isinstance(destination, str):
+            continue
+        destination_path = PurePosixPath(destination)
+        try:
+            relative = candidate.relative_to(destination_path)
+        except ValueError:
+            continue
+        matches.append((len(destination_path.parts), Mapping(
+            Path(source).joinpath(*relative.parts), destination_path
+        )))
+    if not matches:
+        raise CheckFailure(f"{path} is not backed by a Docker bind/volume mount")
+    return max(matches, key=lambda item: item[0])[1]
+
+
+def container_user(container: dict[str, Any]) -> str | None:
+    value = container.get("Config", {}).get("User")
+    return value if isinstance(value, str) and value else None
+
+
+def require_same_path(label: str, left: Mapping, right: Mapping) -> None:
+    if left.source.resolve(strict=False) != right.source.resolve(strict=False):
+        raise CheckFailure(
+            f"{label} host paths differ: {left.source} != {right.source}"
+        )
+
+
+def require_regular_secret(path: Path) -> None:
+    try:
+        mode = path.stat().st_mode
+    except OSError as exc:
+        raise CheckFailure(f"secret is unavailable: {path}") from exc
+    if not stat.S_ISREG(mode) or stat.S_IMODE(mode) & 0o077 or path.stat().st_size == 0:
+        raise CheckFailure(f"secret must be a non-empty mode-0600 regular file: {path}")
+
+
+def route_local_path(syncthing: dict[str, Any]) -> str:
+    """Resolve the fixed, future automatic ``routes/`` path in the client."""
+
+    api_root = PurePosixPath(syncthing["api_root"])
+    candidate = api_root / "routes"
+    overrides = syncthing.get("local_path_overrides", {})
+    if not isinstance(overrides, dict):
+        raise CheckFailure("syncthing.local_path_overrides must be a table")
+    matches: list[tuple[int, PurePosixPath, str]] = []
+    for raw_api, raw_local in overrides.items():
+        if not isinstance(raw_api, str) or not isinstance(raw_local, str):
+            raise CheckFailure("syncthing.local_path_overrides entries are invalid")
+        root = PurePosixPath(raw_api)
+        try:
+            relative = candidate.relative_to(root)
+        except ValueError:
+            continue
+        matches.append((len(root.parts), relative, raw_local))
+    if matches:
+        _, relative, local_root = max(matches, key=lambda item: item[0])
+        return str(Path(local_root).joinpath(*relative.parts))
+    try:
+        relative = candidate.relative_to(api_root)
+    except ValueError as exc:
+        raise CheckFailure("future route path is outside syncthing.api_root") from exc
+    return str(Path(syncthing["local_root"]).joinpath(*relative.parts))
+
+
+def run_client_check(container: str, config: str, flag: str) -> None:
+    result = subprocess.run(
+        ["docker", "exec", container, "archive-client", "--config", config, flag],
+        check=False, text=True, capture_output=True,
+    )
+    if result.returncode:
+        detail = result.stderr.strip() or result.stdout.strip() or "failed"
+        raise CheckFailure(f"client {flag} failed: {detail}")
+    print(result.stdout.strip())
+
+
+def mounted_config(container: str, config: str) -> dict[str, Any]:
+    """Read normalized non-secret config through the client image itself."""
+
+    program = '''import json,sys
+from pathlib import Path
+from archive_clients.config import ClientConfig
+config=ClientConfig.load(Path(sys.argv[1]))
+value={"shared_token_file":str(config.shared_token_file)}
+value["qbittorrent"]={"api_root":str(config.qbittorrent.api_root),"local_root":str(config.qbittorrent.local_root),"password_file":str(config.qbittorrent.password_file)}
+value["syncthing"]={"api_root":str(config.syncthing.api_root),"local_root":str(config.syncthing.local_root),"api_key_file":str(config.syncthing.api_key_file),"local_path_overrides":{str(api):str(local) for api,local in config.syncthing.local_path_overrides}}
+print(json.dumps(value,sort_keys=True))'''
+    result = subprocess.run(
+        ["docker", "exec", container, "python", "-c", program, config],
+        check=False, text=True, capture_output=True,
+    )
+    if result.returncode:
+        detail = result.stderr.strip() or result.stdout.strip() or "failed"
+        raise CheckFailure(f"cannot load mounted client config: {detail}")
+    try:
+        value = json.loads(result.stdout)
+    except json.JSONDecodeError as exc:
+        raise CheckFailure("mounted client config output is invalid") from exc
+    if not isinstance(value, dict):
+        raise CheckFailure("mounted client config is invalid")
+    return value
+
+
+def run_service_check(container: str, config: str) -> None:
+    program = '''import json,sys
+from pathlib import Path
+from archive_clients.config import ClientConfig
+from archive_clients.probes import probe_root
+from archive_clients.services import probe_qbittorrent,probe_syncthing
+config=ClientConfig.load(Path(sys.argv[1]))
+sync_probe=probe_root(config.syncthing.local_root)
+probes=[probe_qbittorrent(config.qbittorrent),probe_syncthing(config.syncthing,sparse_supported=sync_probe.sparse_files)]
+print(json.dumps({"services":[{"service":p.service,"state":p.state,"detail":p.detail,"version":p.version,"device_id":p.device_id} for p in probes]},sort_keys=True))
+raise SystemExit(0 if all(p.state == 1 for p in probes) else 1)'''
+    result = subprocess.run(
+        ["docker", "exec", container, "python", "-c", program, config],
+        check=False, text=True, capture_output=True,
+    )
+    if result.returncode:
+        detail = result.stderr.strip() or result.stdout.strip() or "failed"
+        raise CheckFailure(f"client local-service probe failed: {detail}")
+    print(result.stdout.strip())
+
+
+def run_hardlink_probe(
+    container: str, qb_root: str, route_root: str, user: str | None = None,
+) -> None:
+    """Prove that the daemon can hard-link from qB data into automatic routes."""
+
+    program = r'''import os,sys,tempfile
+source_root,route_root=sys.argv[1:]
+source_path=None
+destination_path=None
+try:
+    source_fd,source_path=tempfile.mkstemp(prefix=".archive-control-preflight-",dir=source_root)
+    os.close(source_fd)
+    destination_fd,destination_path=tempfile.mkstemp(prefix=".archive-control-preflight-",dir=route_root)
+    os.close(destination_fd)
+    os.unlink(destination_path)
+    os.link(source_path,destination_path)
+    source_stat=os.stat(source_path)
+    destination_stat=os.stat(destination_path)
+    if source_stat.st_dev != destination_stat.st_dev or source_stat.st_ino != destination_stat.st_ino:
+        raise RuntimeError("link(2) did not produce the same filesystem inode")
+    print("hard-link staging probe passed")
+finally:
+    for path in (destination_path,source_path):
+        if path:
+            try:
+                os.unlink(path)
+            except FileNotFoundError:
+                pass
+'''
+    command = ["docker", "exec"]
+    if user:
+        command.extend(["--user", user])
+    command.extend([container, "python", "-c", program, qb_root, route_root])
+    result = subprocess.run(
+        command,
+        check=False, text=True, capture_output=True,
+    )
+    if result.returncode:
+        detail = result.stderr.strip() or result.stdout.strip() or "failed"
+        raise CheckFailure(f"hard-link staging probe failed: {detail}")
+    print(result.stdout.strip())
+
+
+def run_existing_route_directory_check(container: str, config: str) -> None:
+    """Reject a mount migration that hides an already configured route folder."""
+
+    program = r'''import json,os,sys
+from pathlib import Path,PurePosixPath
+from archive_clients.config import ClientConfig
+from archive_clients.syncthing import SyncthingHttp
+config=ClientConfig.load(Path(sys.argv[1]))
+transport=SyncthingHttp(config.syncthing)
+route_root=config.syncthing.api_root / "routes"
+missing=[]
+checked=[]
+for folder in transport.get_json("/rest/config").get("folders",[]):
+    if not isinstance(folder,dict) or not str(folder.get("label","")).startswith("archive-control:"):
+        continue
+    raw=folder.get("path")
+    if not isinstance(raw,str) or not raw:
+        continue
+    api=PurePosixPath(raw)
+    if api.parts and api.parts[0] == "~":
+        api=config.syncthing.api_root.joinpath(*api.parts[1:])
+    try:
+        api.relative_to(route_root)
+    except ValueError:
+        continue
+    local=config.syncthing.roots.api_to_local(api.as_posix())
+    checked.append(str(local))
+    if not local.is_dir():
+        missing.append({"folder_id":folder.get("id",""),"path":str(local)})
+print(json.dumps({"checked":checked,"missing":missing},sort_keys=True))
+raise SystemExit(1 if missing else 0)
+'''
+    result = subprocess.run(
+        ["docker", "exec", container, "python", "-c", program, config],
+        check=False, text=True, capture_output=True,
+    )
+    if result.returncode:
+        detail = result.stderr.strip() or result.stdout.strip() or "failed"
+        raise CheckFailure(
+            "an existing Archive Control Syncthing route directory is missing; "
+            f"complete the route-directory migration before starting jobs: {detail}"
+        )
+    print(result.stdout.strip())
+
+
+def main(argv: list[str] | None = None) -> int:
+    parser = argparse.ArgumentParser(
+        description="Read-only Archive Control Docker deployment preflight"
+    )
+    parser.add_argument("--client-config", type=Path, required=True,
+                        help="host path to client.toml")
+    parser.add_argument("--client-container", required=True)
+    parser.add_argument("--syncthing-container", required=True)
+    parser.add_argument("--qbittorrent-container", required=True)
+    parser.add_argument("--container-config", default="/etc/archive-control/client.toml",
+                        help="client.toml path inside the client container")
+    args = parser.parse_args(argv)
+    try:
+        client = docker_inspect(args.client_container)
+        syncthing = docker_inspect(args.syncthing_container)
+        qbittorrent = docker_inspect(args.qbittorrent_container)
+        configured_host_path = map_path(client, args.container_config).source
+        if configured_host_path.resolve(strict=False) != args.client_config.resolve(strict=False):
+            raise CheckFailure(
+                "--client-config does not match the file mounted into the client"
+            )
+        config = mounted_config(args.client_container, args.container_config)
+        qb = config["qbittorrent"]
+        sync = config["syncthing"]
+
+        client_qb = map_path(client, qb["local_root"])
+        client_route = map_path(client, route_local_path(sync))
+        syncthing_route = map_path(
+            syncthing, str(PurePosixPath(sync["api_root"]) / "routes")
+        )
+        qb_api = map_path(qbittorrent, qb["api_root"])
+        require_same_path(
+            "future Syncthing routes/local route mapping",
+            syncthing_route, client_route,
+        )
+        require_same_path("qBittorrent api_root/local_root", qb_api, client_qb)
+        if client_qb.destination != client_route.destination:
+            raise CheckFailure(
+                "qBittorrent and future route roots use separate client bind "
+                "mounts; hard-link staging would be unavailable"
+            )
+        run_hardlink_probe(
+            args.client_container, qb["local_root"], route_local_path(sync),
+            container_user(client),
+        )
+        run_existing_route_directory_check(
+            args.client_container, args.container_config
+        )
+        for key in ("shared_token_file",):
+            require_regular_secret(map_path(client, config[key]).source)
+        for service, key in ((qb, "password_file"), (sync, "api_key_file")):
+            require_regular_secret(map_path(client, service[key]).source)
+        run_client_check(args.client_container, args.container_config, "--check-config")
+        run_service_check(args.client_container, args.container_config)
+    except (CheckFailure, KeyError, OSError) as exc:
+        print(f"preflight failed: {exc}", file=sys.stderr)
+        return 1
+    print("preflight passed: bind mappings, hard-link staging, secrets, filesystem capabilities, and local APIs are healthy")
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())

src/archive_clients/bencode.py

@@ -168,7 +168,7 @@ def _v1_files(info: dict[bytes, Any]) -> list[MetaFile]:
         components = [name] + [_component(part) for part in raw_path]
         result.append(MetaFile(
             "/".join(components), _length(item.get(b"length")),
-            _padding(item),
+            _padding(item) or _bitcomet_padding_name(components[-1]),
         ))
     return result
 
@@ -223,3 +223,13 @@ def _length(value: Any) -> int:
 def _padding(value: dict[bytes, Any]) -> bool:
     attributes = value.get(b"attr", b"")
     return isinstance(attributes, bytes) and b"p" in attributes
+
+
+def _bitcomet_padding_name(component: str) -> bool:
+    """Recognize BitComet's legacy padding-file convention.
+
+    Such v1 torrents often omit the standard ``attr=p`` flag, but
+    qBittorrent/libtorrent still hides these synthetic entries from its file
+    API.  The exact reserved prefix is the interoperable marker.
+    """
+    return component.startswith("_____padding_file_")

src/archive_clients/config.py

@@ -85,6 +85,7 @@ class ConnectionConfig:
     registration_timeout: float = 10
     heartbeat_interval: float = 15
     offline_timeout: float = 45
+    outbound_enqueue_timeout: float = 15
     reconnect_initial: float = 1
     reconnect_max: float = 60
     reconnect_reset_after: float = 60
@@ -249,6 +250,7 @@ def _connection(value: Any) -> ConnectionConfig:
         raise ConfigError("connection must be a table")
     _keys(value, {
         "registration_timeout", "heartbeat_interval", "offline_timeout",
+        "outbound_enqueue_timeout",
         "reconnect_initial", "reconnect_max", "reconnect_reset_after",
         "reconnect_jitter",
     }, "connection")
@@ -256,6 +258,9 @@ def _connection(value: Any) -> ConnectionConfig:
         registration_timeout=_duration(value.get("registration_timeout", "10s")),
         heartbeat_interval=_duration(value.get("heartbeat_interval", "15s")),
         offline_timeout=_duration(value.get("offline_timeout", "45s")),
+        outbound_enqueue_timeout=_duration(
+            value.get("outbound_enqueue_timeout", "15s")
+        ),
         reconnect_initial=_duration(value.get("reconnect_initial", "1s")),
         reconnect_max=_duration(value.get("reconnect_max", "60s")),
         reconnect_reset_after=_duration(value.get("reconnect_reset_after", "60s")),
@@ -265,6 +270,8 @@ def _connection(value: Any) -> ConnectionConfig:
         raise ConfigError("reconnect_initial cannot exceed reconnect_max")
     if result.offline_timeout <= result.heartbeat_interval:
         raise ConfigError("offline_timeout must exceed heartbeat_interval")
+    if result.outbound_enqueue_timeout <= 0:
+        raise ConfigError("outbound_enqueue_timeout must be positive")
     if not isinstance(result.reconnect_jitter, bool):
         raise ConfigError("reconnect_jitter must be boolean")
     return result

src/archive_clients/daemon.py

@@ -229,16 +229,11 @@ class ArchiveClientDaemon:
                 # detects the missing application heartbeats and reconnects.
                 while True:
                     try:
-                        frame = await asyncio.wait_for(
-                            websocket.recv(),
-                            self.config.connection.offline_timeout,
+                        frame = await self._next_frame(
+                            websocket, writer
                         )
                     except ConnectionClosedOK:
                         return
-                    except asyncio.TimeoutError as exc:
-                        raise RuntimeError(
-                            "control heartbeat timed out"
-                        ) from exc
                     await self._handle(decode(frame), outbound, command_tasks)
             finally:
                 writer.cancel()
@@ -323,6 +318,63 @@ class ArchiveClientDaemon:
         while True:
             await websocket.send(await outbound.get())
 
+    async def _next_frame(self, websocket: Any, writer: asyncio.Task[None]):
+        """Receive one frame while supervising the connection writer.
+
+        The previous receive-only wait let a failed writer go unnoticed until
+        the remote side happened to close or a heartbeat timeout elapsed.
+        Waiting for both directions makes a failed send an immediate,
+        reconnectable connection failure.
+        """
+
+        receiver = asyncio.create_task(websocket.recv())
+        try:
+            done, _ = await asyncio.wait(
+                {receiver, writer},
+                timeout=self.config.connection.offline_timeout,
+                return_when=asyncio.FIRST_COMPLETED,
+            )
+            if writer in done:
+                if not receiver.done():
+                    receiver.cancel()
+                    await asyncio.gather(receiver, return_exceptions=True)
+                if writer.cancelled():
+                    raise RuntimeError("control writer was cancelled")
+                error = writer.exception()
+                if error is not None:
+                    raise RuntimeError("control writer failed") from error
+                raise RuntimeError("control writer stopped")
+            if receiver not in done:
+                receiver.cancel()
+                await asyncio.gather(receiver, return_exceptions=True)
+                raise RuntimeError("control heartbeat timed out")
+            return receiver.result()
+        except BaseException:
+            if not receiver.done():
+                receiver.cancel()
+                await asyncio.gather(receiver, return_exceptions=True)
+            raise
+
+    async def _enqueue(
+        self, outbound: asyncio.Queue[str], message: str
+    ) -> None:
+        """Bound connection-local backpressure so a dead writer cannot wedge I/O.
+
+        A job event is durable before it is sent, so dropping this particular
+        connection after a bounded wait is safe: reconciliation/redelivery on
+        the next session will recover it.  In contrast, indefinitely waiting
+        for a full queue can prevent heartbeat acknowledgements from being
+        read or sent, which prevents the reconnect supervisor from running.
+        """
+
+        try:
+            await asyncio.wait_for(
+                outbound.put(message),
+                timeout=self.config.connection.outbound_enqueue_timeout,
+            )
+        except asyncio.TimeoutError as exc:
+            raise RuntimeError("control outbound queue is blocked") from exc
+
     async def _handle(
         self,
         envelope: Any,
@@ -334,13 +386,20 @@ class ArchiveClientDaemon:
             response = new_envelope()
             response.correlation_id = envelope.message_id
             response.heartbeat_ack.sequence = envelope.heartbeat.sequence
-            await outbound.put(encode(response))
+            await self._enqueue(outbound, encode(response))
         elif payload == "command":
             await self._accept_command(envelope, outbound, command_tasks)
         elif payload == "protocol_error":
             logger.warning(
                 "control_reported_protocol_error",
-                extra={"error_code": envelope.protocol_error.error.code},
+                extra={
+                    "error_code": envelope.protocol_error.error.code,
+                    "error_detail": envelope.protocol_error.error.message,
+                    "offending_message_id": (
+                        envelope.protocol_error.offending_message_id
+                    ),
+                    "retryable": envelope.protocol_error.error.retryable,
+                },
             )
 
     async def _accept_command(
@@ -398,7 +457,7 @@ class ArchiveClientDaemon:
         response = new_envelope()
         response.correlation_id = envelope.message_id
         response.command_ack.CopyFrom(acknowledgement)
-        await outbound.put(encode(response))
+        await self._enqueue(outbound, encode(response))
 
         if (
             accepted is not None
@@ -419,7 +478,7 @@ class ArchiveClientDaemon:
                 job_snapshot.last_event_sequence = int(
                     row["last_event_sequence"]
                 )
-                await outbound.put(encode(snapshot))
+                await self._enqueue(outbound, encode(snapshot))
         elif (
             accepted is not None
             and accepted_for_execution
@@ -472,7 +531,6 @@ class ArchiveClientDaemon:
         outbound: asyncio.Queue[str],
         command_tasks: set[asyncio.Task[None]],
     ) -> None:
-        job_commands: list[control_pb2.Command] = []
         for row in await asyncio.to_thread(self.store.list_accepted_commands):
             acknowledgement = decode_message(
                 str(row["acknowledgement_json"]), control_pb2.CommandAck()
@@ -484,28 +542,14 @@ class ArchiveClientDaemon:
             )
             if command.WhichOneof("payload") == "ensure_route":
                 self._schedule_route_command(
-                    command, "", outbound, command_tasks
+                    command, "", outbound, command_tasks, restore_ready=True
                 )
-            elif (
-                command.WhichOneof("payload") in {
-                    "assign_job", "execute_step", "cancel_job"
-                }
-                and self.jobs is not None
-            ):
-                if command.WhichOneof("payload") == "cancel_job":
-                    self.jobs.request_cancel(command.cancel_job.job_id)
-                job_commands.append(command)
-        if job_commands:
-            task = asyncio.create_task(
-                self._resume_job_commands(job_commands, outbound),
-                name="resume-job-commands",
-            )
-            command_tasks.add(task)
-            task.add_done_callback(
-                lambda completed: self._command_finished(
-                    completed, command_tasks
-                )
-            )
+            # Job commands are intentionally not replayed here.  A command
+            # acknowledgement is durable on both sides; the control daemon
+            # redelivers an unacknowledged command, while registration
+            # reconciliation requests snapshots for active jobs.  Replaying
+            # every historical command re-emits overlapping event ranges and
+            # can flood the single connection's bounded outbound queue.
 
     async def _resume_route_commands(
         self,
@@ -604,10 +648,13 @@ class ArchiveClientDaemon:
                 response.correlation_id = correlation_id
                 response.job_event.CopyFrom(event)
                 future = asyncio.run_coroutine_threadsafe(
-                    outbound.put(encode(response)), loop
+                    self._enqueue(outbound, encode(response)), loop
                 )
                 try:
-                    future.result(timeout=30)
+                    future.result(
+                        timeout=self.config.connection.outbound_enqueue_timeout
+                        + 1
+                    )
                 except Exception:
                     # The event is already durable in the client DB and will
                     # be replayed after reconnect.
@@ -627,7 +674,7 @@ class ArchiveClientDaemon:
             response = new_envelope()
             response.correlation_id = correlation_id
             response.job_event.CopyFrom(event)
-            await outbound.put(encode(response))
+            await self._enqueue(outbound, encode(response))
 
     def _schedule_route_command(
         self,
@@ -635,12 +682,15 @@ class ArchiveClientDaemon:
         correlation_id: str,
         outbound: asyncio.Queue[str],
         command_tasks: set[asyncio.Task[None]],
+        restore_ready: bool = False,
     ) -> None:
         if command.command_id in self._active_route_commands:
             return
         self._active_route_commands.add(command.command_id)
         task = asyncio.create_task(
-            self._execute_route(command, correlation_id, outbound),
+            self._execute_route(
+                command, correlation_id, outbound, restore_ready=restore_ready
+            ),
             name=f"route-{command.ensure_route.route.route_id}",
         )
         command_tasks.add(task)
@@ -671,13 +721,14 @@ class ArchiveClientDaemon:
             response = new_envelope()
             response.correlation_id = correlation_id
             response.inventory_chunk.CopyFrom(chunk)
-            await outbound.put(encode(response))
+            await self._enqueue(outbound, encode(response))
 
     async def _execute_route(
         self,
         command: Any,
         correlation_id: str,
         outbound: asyncio.Queue[str],
+        restore_ready: bool = False,
     ) -> None:
         assert self.routes is not None
         spec = command.ensure_route.route
@@ -696,8 +747,21 @@ class ArchiveClientDaemon:
             response.route_update.CopyFrom(
                 decode_message(str(row["update_json"]), control_pb2.RouteUpdate())
             )
-            await outbound.put(encode(response))
-        if attempt["state"] in {"ready", "failed"}:
+            await self._enqueue(outbound, encode(response))
+        if attempt["state"] == "ready":
+            if restore_ready:
+                # Route readiness is durable, but this process-local lookup is
+                # not. Reconfigure (which validates the existing Syncthing
+                # folder and recreates its local directory if necessary) after
+                # a daemon restart, before replaying stored READY updates.
+                configured = await asyncio.to_thread(
+                    self.routes.configure,
+                    spec,
+                    time.monotonic() + spec.setup_timeout_seconds,
+                )
+                self._known_route_paths[spec.route_id] = configured.local_path
+            return
+        if attempt["state"] == "failed":
             return
 
         deadline = time.monotonic() + spec.setup_timeout_seconds
@@ -848,7 +912,7 @@ class ArchiveClientDaemon:
         response = new_envelope()
         response.correlation_id = correlation_id
         response.route_update.CopyFrom(update)
-        await outbound.put(encode(response))
+        await self._enqueue(outbound, encode(response))
 
     def _local_route(
         self, spec: route_pb2.EnsureRouteSpec, state: int

tests/test_daemon.py

@@ -110,6 +110,123 @@ class DaemonTransportTests(unittest.IsolatedAsyncioTestCase):
                 ):
                     await asyncio.wait_for(daemon._connection(), 1)
 
+    async def test_failed_writer_ends_connection_without_waiting_for_heartbeat(self):
+        """A send failure must immediately reach the reconnect supervisor."""
+
+        async def control(websocket):
+            registration = decode(await websocket.recv())
+            response = new_envelope()
+            response.correlation_id = registration.message_id
+            response.register_response.status = (
+                client_pb2.REGISTRATION_STATUS_ACCEPTED
+            )
+            response.register_response.negotiated_version.major = 1
+            await websocket.send(encode(response))
+            await websocket.wait_closed()
+
+        with tempfile.TemporaryDirectory() as directory:
+            root = Path(directory)
+            token = root / "token"
+            token.write_text("shared-secret", encoding="utf-8")
+            os.chmod(token, 0o600)
+            async with serve(control, "127.0.0.1", 0, ping_interval=None) as server:
+                port = server.sockets[0].getsockname()[1]
+                service = ServiceConfig("http://local", PurePosixPath("/api"), root)
+                config = ClientConfig(
+                    "cache-1", "Cache 1", "cache",
+                    f"ws://127.0.0.1:{port}", token,
+                    root / "state.db", root / "backups", service, service,
+                )
+                probe = FilesystemProbe(root, True, True, True, True, True)
+                daemon = ArchiveClientDaemon(config, [probe, probe], [])
+                await asyncio.to_thread(daemon.store.initialize)
+
+                async def failed_writer(websocket, outbound):
+                    raise OSError("simulated broken socket")
+
+                daemon._writer = failed_writer
+                with self.assertRaisesRegex(RuntimeError, "writer failed"):
+                    await asyncio.wait_for(daemon._connection(), 1)
+
+    async def test_full_outbound_queue_aborts_connection_instead_of_blocking_heartbeats(self):
+        """Bulk output cannot indefinitely block the receive/heartbeat loop."""
+
+        async def control(websocket):
+            registration = decode(await websocket.recv())
+            response = new_envelope()
+            response.correlation_id = registration.message_id
+            response.register_response.status = (
+                client_pb2.REGISTRATION_STATUS_ACCEPTED
+            )
+            response.register_response.negotiated_version.major = 1
+            await websocket.send(encode(response))
+            for sequence in range(1, 102):
+                heartbeat = new_envelope()
+                heartbeat.heartbeat.sequence = sequence
+                await websocket.send(encode(heartbeat))
+            await websocket.wait_closed()
+
+        with tempfile.TemporaryDirectory() as directory:
+            root = Path(directory)
+            token = root / "token"
+            token.write_text("shared-secret", encoding="utf-8")
+            os.chmod(token, 0o600)
+            async with serve(control, "127.0.0.1", 0, ping_interval=None) as server:
+                port = server.sockets[0].getsockname()[1]
+                service = ServiceConfig("http://local", PurePosixPath("/api"), root)
+                config = ClientConfig(
+                    "cache-1", "Cache 1", "cache",
+                    f"ws://127.0.0.1:{port}", token,
+                    root / "state.db", root / "backups", service, service,
+                    ConnectionConfig(outbound_enqueue_timeout=0.01),
+                )
+                probe = FilesystemProbe(root, True, True, True, True, True)
+                daemon = ArchiveClientDaemon(config, [probe, probe], [])
+                await asyncio.to_thread(daemon.store.initialize)
+
+                async def stopped_writer(websocket, outbound):
+                    await asyncio.Event().wait()
+
+                daemon._writer = stopped_writer
+                with self.assertRaisesRegex(RuntimeError, "outbound queue is blocked"):
+                    await asyncio.wait_for(daemon._connection(), 2)
+
+    async def test_reconnect_resume_skips_historical_job_commands(self):
+        """Registration reconciliation, not command replay, recovers job state."""
+
+        with tempfile.TemporaryDirectory() as directory:
+            root = Path(directory)
+            token = root / "token"
+            token.write_text("shared-secret", encoding="utf-8")
+            os.chmod(token, 0o600)
+            service = ServiceConfig("http://local", PurePosixPath("/api"), root)
+            config = ClientConfig(
+                "cache-1", "Cache 1", "cache", "ws://control", token,
+                root / "state.db", root / "backups", service, service,
+            )
+            probe = FilesystemProbe(root, True, True, True, True, True)
+            daemon = ArchiveClientDaemon(config, [probe, probe], [])
+            await asyncio.to_thread(daemon.store.initialize)
+            command = control_pb2.Command(command_id=str(uuid4()))
+            command.execute_step.job_id = str(uuid4())
+            command.execute_step.expected_last_event_sequence = 1
+            acknowledgement = control_pb2.CommandAck(
+                command_id=command.command_id,
+                status=control_pb2.COMMAND_ACK_STATUS_ACCEPTED,
+            )
+            await asyncio.to_thread(
+                daemon.store.accept_command,
+                command.command_id,
+                encode_message(command),
+                encode_message(acknowledgement),
+            )
+            daemon.jobs = Mock()
+            outbound = asyncio.Queue()
+            tasks = set()
+            await daemon._resume_commands(outbound, tasks)
+            self.assertEqual(tasks, set())
+            self.assertTrue(outbound.empty())
+
     async def test_eviction_assignment_and_steps_are_admitted(self):
         with tempfile.TemporaryDirectory() as directory:
             root = Path(directory)
@@ -280,6 +397,10 @@ class DaemonTransportTests(unittest.IsolatedAsyncioTestCase):
                 for _ in range(3)
             ]
             self.assertEqual([item.sequence for item in resumed], [1, 2, 3])
+            self.assertEqual(manager.configure.call_count, 2)
+            self.assertEqual(
+                restarted._route_path("route-1"), root / "routes/route-1"
+            )
 
     async def test_slow_inventory_does_not_block_heartbeat(self):
         with tempfile.TemporaryDirectory() as directory:

tests/test_deployment_preflight.py

@@ -0,0 +1,83 @@
+import importlib.util
+import sys
+import unittest
+from unittest import mock
+from pathlib import Path
+
+
+_SCRIPT = Path(__file__).parents[1] / "scripts" / "preflight-deployment.py"
+_SPEC = importlib.util.spec_from_file_location("deployment_preflight", _SCRIPT)
+assert _SPEC is not None and _SPEC.loader is not None
+preflight = importlib.util.module_from_spec(_SPEC)
+sys.modules[_SPEC.name] = preflight
+_SPEC.loader.exec_module(preflight)
+
+
+class DeploymentPreflightTests(unittest.TestCase):
+    def test_maps_nested_container_path_to_host_source(self):
+        container = {
+            "Mounts": [
+                {"Type": "bind", "Source": "/srv/data", "Destination": "/data"},
+                {"Type": "bind", "Source": "/srv/sync", "Destination": "/data/Sync"},
+            ]
+        }
+        mapping = preflight.map_path(container, "/data/Sync/routes/route-1")
+        self.assertEqual(mapping.source, Path("/srv/sync/routes/route-1"))
+        self.assertEqual(str(mapping.destination), "/data/Sync")
+
+    def test_rejects_api_and_client_roots_from_different_host_paths(self):
+        with self.assertRaisesRegex(preflight.CheckFailure, "host paths differ"):
+            preflight.require_same_path(
+                "Syncthing api_root/local_root",
+                preflight.Mapping(Path("/srv/syncthing-config/routes"), Path("/var/syncthing")),
+                preflight.Mapping(Path("/srv/sync/routes"), Path("/data/storage")),
+            )
+
+    def test_rejects_unmounted_container_path(self):
+        with self.assertRaisesRegex(preflight.CheckFailure, "not backed"):
+            preflight.map_path({"Mounts": []}, "/missing")
+
+    def test_maps_future_route_through_syncthing_override(self):
+        config = {
+            "api_root": "/var/syncthing",
+            "local_root": "/data/sync",
+            "local_path_overrides": {
+                "/var/syncthing/routes": "/data/qb/.archive-control-routes",
+            },
+        }
+        self.assertEqual(
+            preflight.route_local_path(config),
+            "/data/qb/.archive-control-routes",
+        )
+
+    def test_runs_hardlink_probe_with_qb_and_route_roots(self):
+        completed = __import__("subprocess").CompletedProcess(
+            args=[], returncode=0, stdout="hard-link staging probe passed\n", stderr=""
+        )
+        with mock.patch.object(preflight.subprocess, "run", return_value=completed) as run:
+            preflight.run_hardlink_probe(
+                "client", "/data/qb", "/data/qb/routes", "1001:1001"
+            )
+        args = run.call_args.args[0]
+        self.assertEqual(args[:6], ["docker", "exec", "--user", "1001:1001", "client", "python"])
+        self.assertEqual(args[-2:], ["/data/qb", "/data/qb/routes"])
+
+    def test_hardlink_probe_failure_is_actionable(self):
+        completed = __import__("subprocess").CompletedProcess(
+            args=[], returncode=1, stdout="", stderr="[Errno 18] Invalid cross-device link"
+        )
+        with mock.patch.object(preflight.subprocess, "run", return_value=completed):
+            with self.assertRaisesRegex(preflight.CheckFailure, "hard-link staging probe failed"):
+                preflight.run_hardlink_probe("client", "/data/qb", "/data/qb/routes")
+
+    def test_existing_route_directory_failure_is_actionable(self):
+        completed = __import__("subprocess").CompletedProcess(
+            args=[], returncode=1, stdout='{"missing":[{"folder_id":"route-1"}]}', stderr=""
+        )
+        with mock.patch.object(preflight.subprocess, "run", return_value=completed):
+            with self.assertRaisesRegex(preflight.CheckFailure, "route directory is missing"):
+                preflight.run_existing_route_directory_check("client", "/config.toml")
+
+
+if __name__ == "__main__":
+    unittest.main()

tests/test_resources.py

@@ -123,7 +123,7 @@ class ResourceTests(unittest.TestCase):
             b"files": [
                 {b"length": 3, b"path": [b"first.bin"]},
                 {
-                    b"attr": b"p", b"length": 5,
+                    b"length": 5,
                     b"path": [b"_____padding_file_5"],
                 },
                 {b"length": 7, b"path": [b"last.bin"]},