roborock.devices
Roborock Devices & Discovery
The devices module provides functionality to discover Roborock devices on the network. This section documents the full lifecycle of device discovery across Cloud and Network.
Usage TL;DR
- Discovery: Use
roborock.devices.device_manager.DeviceManagerto get device instances.- Call
create_device_manager(user_params)thenawait device_manager.get_devices().
- Call
- Control:
- Vacuums (V1): Use
device.v1_propertiesto access traits likestatusorconsumables.- Call
await trait.refresh()to update state. - Use
device.v1_properties.command.send()for raw commands (start/stop).
- Call
- Washers (A01): Use
device.a01_propertiesfor Dyad/Zeo devices.- Use
await device.a01_properties.query_values([...])to get state. - Use
await device.a01_properties.set_value(protocol, value)to control.
- Use
- Vacuums (V1): Use
Background: Understanding Device Protocols
The library supports three device protocol versions, each with different capabilities:
Protocol Summary
| Protocol | Device Examples | MQTT | Local TCP | Channel Type | Notes |
|---|---|---|---|---|---|
V1 (pv=1.0) |
Most vacuum robots (S7, S8, Q5, Q7, etc.) | ✅ | ✅ | V1Channel with RpcChannel |
Prefers local, falls back to MQTT |
A01 (pv=A01) |
Dyad, Zeo washers | ✅ | ❌ | MqttChannel + helpers |
MQTT only, DPS protocol |
B01 (pv=B01) |
Some newer models | ✅ | ❌ | MqttChannel + helpers |
MQTT only, DPS protocol |
Key Point: The DeviceManager automatically detects the protocol version and creates the appropriate channel type. You don't need to handle this manually.
Internal Architecture
The library is organized into distinct layers, each with a specific responsibility. Different device protocols use different channel implementations:
graph TB subgraph "Application Layer" User[Application Code] end subgraph "Device Management Layer" DM[DeviceManager<br/>Detects protocol version] end subgraph "Device Types by Protocol" V1Dev[V1 Devices<br/>pv=1.0<br/>Most vacuums] A01Dev[A01 Devices<br/>pv=A01<br/>Dyad, Zeo] B01Dev[B01 Devices<br/>pv=B01<br/>Some models] end subgraph "Traits Layer" V1Traits[V1 Traits<br/>Clean, Map, etc.] A01Traits[A01 Traits<br/>DPS-based] B01Traits[B01 Traits<br/>DPS-based] end subgraph "Channel Layer" V1C[V1Channel<br/>MQTT + Local] A01C[A01 send_decoded_command<br/>MQTT only] B01C[B01 send_decoded_command<br/>MQTT only] RPC[RpcChannel<br/>Multi-strategy] MC[MqttChannel<br/>Per-device wrapper] LC[LocalChannel<br/>TCP :58867] end subgraph "Session Layer" MS[MqttSession<br/>SHARED by all devices<br/>Idle timeout] LS[LocalSession<br/>Factory] end subgraph "Protocol Layer" V1P[V1 Protocol<br/>JSON RPC + AES] A01P[A01 Protocol<br/>DPS format] B01P[B01 Protocol<br/>DPS format] end subgraph "Transport Layer" MQTT[MQTT Broker<br/>Roborock Cloud] TCP[TCP Socket<br/>Direct to device] end User --> DM DM -->|pv=1.0| V1Dev DM -->|pv=A01| A01Dev DM -->|pv=B01| B01Dev V1Dev --> V1Traits A01Dev --> A01Traits B01Dev --> B01Traits V1Traits --> V1C A01Traits --> A01C B01Traits --> B01C V1C --> RPC RPC -->|Strategy 1| LC RPC -->|Strategy 2| MC A01C --> MC B01C --> MC MC --> MS LC --> LS MC --> V1P MC --> A01P MC --> B01P LC --> V1P MS --> MQTT LC --> TCP MQTT <--> TCP style User fill:#e1f5ff style DM fill:#fff4e1 style V1C fill:#ffe1e1 style RPC fill:#ffe1e1 style MS fill:#e1ffe1 style V1P fill:#f0e1ff style A01P fill:#f0e1ff style B01P fill:#f0e1ff
Layer Responsibilities
- Device Management Layer: Detects protocol version (
pvfield) and creates appropriate channels - Device Types: Different devices based on protocol version (V1, A01, B01)
- Traits Layer: Protocol-specific device capabilities and commands
- Channel Layer: Protocol-specific communication patterns
- V1: Full RPC channel with local + MQTT fallback
- A01/B01: Helper functions wrapping MqttChannel (MQTT only)
- MqttChannel: Per-device wrapper that uses shared
MqttSession
- Session Layer: Connection pooling and subscription management
- MqttSession: Shared single connection for all devices
- LocalSession: Factory for creating device-specific local connections
- Protocol Layer: Message encoding/decoding for different device versions
- Transport Layer: Low-level MQTT and TCP communication
Important: All MqttChannel instances share the same MqttSession, which maintains a single MQTT connection to the broker. This means:
- Only one TCP connection to the MQTT broker regardless of device count
- Subscription management is centralized with idle timeout optimization
- All devices communicate through device-specific MQTT topics on the shared connection
Protocol-Specific Architecture
| Protocol | Channel Type | Local Support | RPC Strategy | Use Case |
|---|---|---|---|---|
V1 (pv=1.0) |
V1Channel with RpcChannel |
✅ Yes | Multi-strategy (Local → MQTT) | Most vacuum robots |
A01 (pv=A01) |
MqttChannel + helpers |
❌ No | Direct MQTT | Dyad, Zeo washers |
B01 (pv=B01) |
MqttChannel + helpers |
❌ No | Direct MQTT | Some newer models |
Account Setup Internals
Login
- Login can happen with either email and password or email and sending a code. We currently prefer email with sending a code -- however the roborock no longer supports this method of login. In the future we may want to migrate to password if this login method is no longer supported.
- The Login API provides a
userDataobject with information on connecting to the cloud APIs - This
rriotdata contains per-session information, unique each time you login.- This contains information used to connect to MQTT
- You get an
-eusuffix in the API URLs if you are in the eu and-usif you are in the us
Home Data Internals
The HomeData includes information about the various devices in the home. We use v3
and it is notable that if devices don't show up in the home_data response it is likely
that a newer version of the API should be used.
products: This is a list of all of the products you have on your account. These objects are always the same (i.e. a s7 maxv is always the exact same.)- It only shows the products for devices available on your account
devicesandreceived_devices:- These both share the same objects, but one is for devices that have been shared with you and one is those that are on your account.
- The big things here are (MOST are static):
duid: A unique identifier for your device (this is always the same i think)name: The name of the device in your applocal_key: The local key that is needed for encoding and decoding messages for the device. This stays the same unless someone sets their vacuum back up.pv: the protocol version (i.e. 1.0 or A1 or B1)product_id: The id of the product from the above products list.device_status: An initial status for some of the data we care about, though this changes on each update.
rooms: The rooms in the home.- This changes if the user adds a new room or changes its name.
- We have to combine this with the room numbers from
GET_ROOM_MAPPINGon the device - There is another REST request
get_roomsthat will do the same thing. - Note: If we cache home_data, we likely need to use
get_roomsto get rooms fresh
Connection Implementation
Connection Flow by Protocol
The connection flow differs based on the device protocol version:
V1 Devices (Most Vacuums) - MQTT + Local
sequenceDiagram participant App as Application participant DM as DeviceManager participant V1C as V1Channel participant RPC as RpcChannel participant MC as MqttChannel participant LC as LocalChannel participant MS as MqttSession participant Broker as MQTT Broker participant Device as V1 Vacuum App->>DM: create_device_manager() DM->>MS: Create MQTT Session MS->>Broker: Connect Broker-->>MS: Connected App->>DM: get_devices() Note over DM: Detect pv=1.0 DM->>V1C: Create V1Channel V1C->>MC: Create MqttChannel V1C->>LC: Create LocalChannel (deferred) Note over V1C: Subscribe to device topics V1C->>MC: subscribe() MC->>MS: subscribe(topic, callback) MS->>Broker: SUBSCRIBE Note over V1C: Fetch network info via MQTT V1C->>RPC: send_command(GET_NETWORK_INFO) RPC->>MC: publish(request) MC->>MS: publish(topic, message) MS->>Broker: PUBLISH Broker->>Device: Command Device->>Broker: Response Broker->>MS: Message MS->>MC: callback(message) MC->>RPC: decoded message RPC-->>V1C: NetworkInfo Note over V1C: Connect locally using IP V1C->>LC: connect() LC->>Device: TCP Connect :58867 Device-->>LC: Connected Note over App: Commands prefer local App->>V1C: send_command(GET_STATUS) V1C->>RPC: send_command() RPC->>LC: publish(request) [Try local first] LC->>Device: Command via TCP Device->>LC: Response LC->>RPC: decoded message RPC-->>App: Status
A01/B01 Devices (Dyad, Zeo) - MQTT Only
sequenceDiagram participant App as Application participant DM as DeviceManager participant A01 as A01 Traits participant Helper as send_decoded_command participant MC as MqttChannel participant MS as MqttSession participant Broker as MQTT Broker participant Device as A01 Device App->>DM: create_device_manager() DM->>MS: Create MQTT Session MS->>Broker: Connect Broker-->>MS: Connected App->>DM: get_devices() Note over DM: Detect pv=A01 DM->>MC: Create MqttChannel DM->>A01: Create A01 Traits Note over A01: Subscribe to device topics A01->>MC: subscribe() MC->>MS: subscribe(topic, callback) MS->>Broker: SUBSCRIBE Note over App: All commands via MQTT App->>A01: set_power(True) A01->>Helper: send_decoded_command() Helper->>MC: subscribe(find_response) Helper->>MC: publish(request) MC->>MS: publish(topic, message) MS->>Broker: PUBLISH Broker->>Device: Command Device->>Broker: Response Broker->>MS: Message MS->>MC: callback(message) MC->>Helper: decoded message Helper-->>App: Result
Key Differences
| Aspect | V1 Devices | A01/B01 Devices |
|---|---|---|
| Protocols | V1 Protocol (JSON RPC) | DPS Protocol |
| Transports | MQTT + Local TCP | MQTT only |
| Channel Type | V1Channel with RpcChannel |
MqttChannel with helpers |
| Local Support | ✅ Yes, preferred | ❌ No |
| Fallback | Local → MQTT | N/A |
| Connection | Requires network info fetch | Direct MQTT |
| Examples | Most vacuum robots | Dyad washers, Zeo models |
MQTT Connection (All Devices)
- Initial device information must be obtained from MQTT
- For V1 devices, we set up the MQTT device connection before the local device connection
- The
NetworkingInfoneeds to be fetched to get additional information about connecting to the device (e.g., Local IP Address) - This networking info can be cached to reduce network calls
- MQTT is also the only way to get the device Map
- The
- Incoming and outgoing messages are decoded/encoded using the device
local_key - For A01/B01 devices, MQTT is the only transport
Local Connection (V1 Devices Only)
- We use the
ipfrom theNetworkingInfoto find the device - The local connection is preferred for improved latency and reducing load on the cloud servers to avoid rate limiting
- Connections are made using a normal TCP socket on port
58867 - Incoming and outgoing messages are decoded/encoded using the device
local_key - Messages received on the stream may be partially received, so we keep a running buffer as messages are partially decoded
- Not available for A01/B01 devices
RPC Pattern (V1 Devices)
V1 devices use a publish/subscribe model for both MQTT and local connections, with an RPC abstraction on top:
graph LR subgraph "RPC Layer" A[send_command] -->|1. Create request| B[Encoder] B -->|2. Subscribe for response| C[Channel.subscribe] B -->|3. Publish request| D[Channel.publish] C -->|4. Wait for match| E[find_response callback] E -->|5. Match request_id| F[Future.set_result] F -->|6. Return| G[Command Result] end subgraph "Channel Layer" C --> H[Subscription Map] D --> I[Transport] I --> J[Device] J --> K[Incoming Messages] K --> H H --> E end
Key Design Points:
- Temporary Subscriptions: Each RPC creates a temporary subscription that matches the request ID
- Subscription Reuse:
MqttSessionkeeps subscriptions alive for 60 seconds (or idle timeout) to enable reuse during command bursts - Timeout Handling: Commands timeout after 10 seconds if no response is received
- Multiple Strategies:
V1Channeltries local first, then falls back to MQTT if local fails
Class Design & Components
Current Architecture
The current design separates concerns into distinct layers:
classDiagram class Channel { <<abstract>> +subscribe(callback) Callable +publish(message) +is_connected() bool } class MqttChannel { -MqttSession session -duid: str -local_key: str +subscribe(callback) +publish(message) } class LocalChannel { -host: str -transport: Transport -local_key: str +connect() +subscribe(callback) +publish(message) +close() } class V1Channel { -MqttChannel mqtt_channel -LocalChannel local_channel -RpcChannel rpc_channel +send_command(method, params) +subscribe(callback) } class RpcChannel { -List~RpcStrategy~ strategies +send_command(method, params) } class RpcStrategy { +name: str +channel: Channel +encoder: Callable +decoder: Callable +health_manager: HealthManager } class MqttSession { -Client client -dict listeners -dict idle_timers +subscribe(topic, callback) +publish(topic, payload) +close() } Channel <|-- MqttChannel Channel <|-- LocalChannel Channel <|-- V1Channel MqttChannel --> MqttSession V1Channel --> MqttChannel V1Channel --> LocalChannel V1Channel --> RpcChannel RpcChannel --> RpcStrategy RpcStrategy --> Channel
Key Components
Channel Interface
The Channel abstraction provides a uniform interface for both MQTT and local connections:
subscribe(callback): Register a callback for incoming messagespublish(message): Send a message to the deviceis_connected: Check connection status
This abstraction allows the RPC layer to work identically over both transports.
MqttSession (Shared Across All Devices)
The MqttSession manages a single shared MQTT connection for all devices:
- Single Connection: Only one TCP connection to the MQTT broker, regardless of device count
- Per-Device Topics: Each device communicates via its own MQTT topics (e.g.,
rr/m/i/{user}/{username}/{duid}) - Subscription Pooling: Multiple callbacks can subscribe to the same topic
- Idle Timeout: Keeps subscriptions alive for 10 seconds after the last callback unsubscribes (enables reuse during command bursts)
- Reconnection: Automatically reconnects and re-establishes all subscriptions on connection loss
- Thread-Safe: Uses asyncio primitives for safe concurrent access
Efficiency: Creating 5 devices means 5 MqttChannel instances but only 1 MqttSession and 1 MQTT broker connection.
MqttChannel (Per-Device Wrapper)
Each device gets its own MqttChannel instance that:
- Wraps the shared
MqttSession - Manages device-specific topics (publish to
rr/m/i/.../duid, subscribe torr/m/o/.../duid) - Handles protocol-specific encoding/decoding with the device's
local_key - Provides the same
Channelinterface asLocalChannel
RpcChannel with Multiple Strategies (V1 Only)
The RpcChannel implements the request/response pattern over pub/sub channels and is only used by V1 devices:
# Example: V1Channel tries local first, then MQTT
strategies = [
RpcStrategy(name="local", channel=local_channel, ...),
RpcStrategy(name="mqtt", channel=mqtt_channel, ...),
]
rpc_channel = RpcChannel(strategies)
For each V1 command:
- Try the first strategy (local)
- If it fails, try the next strategy (MQTT)
- Return the first successful result
A01/B01 devices don't use RpcChannel. Instead, they use helper functions (send_decoded_command) that directly wrap MqttChannel.
Protocol-Specific Channel Architecture
| Component | V1 Devices | A01/B01 Devices |
|---|---|---|
| Channel Class | V1Channel |
MqttChannel directly |
| RPC Abstraction | RpcChannel with strategies |
Helper functions |
| Strategy Pattern | ✅ Multi-strategy (Local → MQTT) | ❌ Direct MQTT only |
| Health Manager | ✅ Tracks local/MQTT health | ❌ Not needed |
| Code Location | v1_channel.py |
a01_channel.py, b01_channel.py |
Health Management (V1 Only)
Each V1 RPC strategy can have a HealthManager that tracks success/failure:
- Exponential Backoff: After failures, wait before retrying
- Automatic Recovery: Periodically attempt to restore failed connections
- Network Info Refresh: Refresh local IP addresses after extended periods
A01/B01 devices don't need health management since they only use MQTT (no fallback).
Protocol Versions
Different device models use different protocol versions:
| Protocol | Devices | Encoding |
|---|---|---|
| V1 | Most vacuum robots | JSON RPC with AES encryption |
| A01 | Dyad, Zeo | DPS-based protocol |
| B01 | Some newer models | DPS-based protocol |
| L01 | Local protocol variant | Binary protocol negotiation |
The protocol layer handles encoding/decoding transparently based on the device's pv field.
Prior API Issues
Complex Inheritance Hierarchy: Multiple inheritance with classes like RoborockMqttClientV1 inheriting from both RoborockMqttClient and RoborockClientV1
Callback-Heavy Design: Heavy reliance on callbacks and listeners in RoborockClientV1.on_message_received and the ListenerModel system
Version Fragmentation: Separate v1 and A01 APIs with different patterns and abstractions
Mixed Concerns: Classes handle both communication protocols (MQTT/local) and device-specific logic
Complex Caching: The AttributeCache system with RepeatableTask adds complexity
Manual Connection Management: Users need to manually set up both MQTT and local clients as shown in the README example
Design Goals
Prefer a single unified client that handles both MQTT and local connections internally.
Home and device discovery (fetching home data and device setup) will be behind a single API.
Asyncio First: Everything should be asyncio as much as possible, with fewer callbacks.
The clients should be working in terms of devices. We need to detect capabilities for each device and not expose details about API versions.
Reliability issues: The current Home Assistant integration has issues with reliability and needs to be simplified. It may be that there are bugs with the exception handling and it's too heavy on the cloud APIs and could benefit from more seamless caching.
Migration from Legacy APIs
The library previously had:
- Separate
RoborockMqttClientV1andRoborockLocalClientV1classes - Manual connection management
- Callback-heavy design with
on_message_received - Complex inheritance hierarchies
The new design:
DeviceManagerhandles all connection managementV1Channelautomatically manages both MQTT and local- Asyncio-first with minimal callbacks
- Clear separation of concerns through layers
- Users work with devices, not raw clients
Note: Legacy APIs in version_1_apis/ and version_a01_apis/ are deprecated and will be removed.
Implementation Details
Code Organization
roborock/
├── devices/ # Device management and channels
│ ├── device_manager.py # High-level device lifecycle
│ ├── channel.py # Base Channel interface
│ ├── mqtt_channel.py # MQTT channel implementation
│ ├── local_channel.py # Local TCP channel implementation
│ ├── v1_channel.py # V1 protocol channel with RPC strategies
│ ├── a01_channel.py # A01 protocol helpers
│ ├── b01_channel.py # B01 protocol helpers
│ └── traits/ # Device-specific command traits
│ └── v1/ # V1 device traits
│ ├── __init__.py # Trait initialization
│ ├── clean.py # Cleaning commands
│ ├── map.py # Map management
│ └── ...
├── mqtt/ # MQTT session management
│ ├── session.py # Base session interface
│ └── roborock_session.py # MQTT session with idle timeout
├── protocols/ # Protocol encoders/decoders
│ ├── v1_protocol.py # V1 JSON RPC protocol
│ ├── a01_protocol.py # A01 protocol
│ ├── b01_protocol.py # B01 protocol
│ └── ...
└── data/ # Data containers and mappings
├── containers.py # Status, HomeData, etc.
└── v1/ # V1-specific data structures
Threading Model
The library is asyncio-only with no threads:
- All I/O is non-blocking using
asyncio - No thread synchronization needed (single event loop)
- Callbacks are executed in the event loop
- Use
asyncio.create_task()for background work
Error Handling
graph TD A[send_command] --> B{Local Available?} B -->|Yes| C[Try Local] B -->|No| D[Try MQTT] C --> E{Success?} E -->|Yes| F[Return Result] E -->|No| G{Timeout?} G -->|Yes| H[Update Health Manager] H --> D G -->|No| I{Connection Error?} I -->|Yes| J[Mark Connection Failed] J --> D I -->|No| D D --> K{Success?} K -->|Yes| F K -->|No| L[Raise RoborockException]
Exception Types:
RoborockException: Base exception for all library errorsRoborockConnectionException: Connection-related failuresRoborockTimeout: Command timeout (10 seconds)
Caching Strategy
To reduce API calls and improve reliability:
- Home Data: Cached on disk, refreshed periodically
- Network Info: Cached for 12 hours
- Device Capabilities: Detected once and cached
- MQTT Subscriptions: Kept alive for 60 seconds (idle timeout)
Testing
Test structure mirrors the python module structure. For example,
the module roborock.devices.traits.v1.maps is tested in the file
tests/devices/traits/v1/test_maps.py. Each test file corresponds to a python
module.
The test suite uses mocking extensively to avoid real devices:
MockandAsyncMockfor channels and sessions- Fake message generators (
mqtt_packet.gen_publish()) - Snapshot testing for complex data structures
- Time-based tests use small timeouts (10-50ms) for speed
Example test structure:
@pytest.fixture
def mock_mqtt_channel():
"""Mock MQTT channel that simulates responses."""
channel = AsyncMock(spec=MqttChannel)
channel.response_queue = []
async def publish_side_effect(message):
# Simulate device response
if channel.response_queue:
response = channel.response_queue.pop(0)
await callback(response)
channel.publish.side_effect = publish_side_effect
return channel