pyDE1 0.6.0

Controller for Decent Espresso DE1

pyDE1

License

Copyright © 2021 Jeff Kletsky. All Rights Reserved.

License for this software, part of the pyDE1 package, is granted under

GNU General Public License v3.0 only

SPDX-License-Identifier: GPL-3.0-only

Overview

This represents work-in-progress to an API-first implementation of core software for a controller for the DE1.

The extent of functionality is sufficient to upload profiles and pull shots, flush the group, steam, and draw hot water through the API, with stop-at-time, -volume, and -mass. Continuous updates of flow parameters, and state transitions are provided over MQTT. Firmware upload is supported, though not yet revealed in the API.

Profiles and real-time data are captured into a SQLite3 database that allows multiple, concurrent access. An example program is provided that generates legacy-style, "shot files" that are compatible with Visualizer and John Weiss' shot-plotting programs.

A "worked example" is available at examples/find_first_and_load.py that

• Initializes and starts an MQTT listener, then, through the API
• Determines if a DE1 and scale are connected
• If not, connects to the first-found
• Waits until the DE1 is "ready" (self-initializes without API intervention)
• Uploads a profile
• Sets the stop-at-weight target and disables stop-at-time and stop-at-volume
• Optionally disconnects the DE1 and scale

The APIs are under semantic versioning. The REST-like, HTTP-transport versions can be retrieved from version at the document root, and also include the Python and package versions installed. Each of the JSON-formatted, MQTT packets contains a version key:value for that payload.

Consumers of these APIs should only need to understand high-level actions, such as "Here is a profile blob, please load it." The operations and choice of connectivity to the devices is "hidden" behind the APIs.

Revision History

See also CHANGELOG.md

• 2021-07-25 – 0.6.0 adds database store
• 2021-07-14 – 0.5.0, "worked example" description
• 2021-07-03 – Updated for release 0.4.0, see also CHANGELOG.md
• 2021-06-26 – Content and organizational updates for release 0.3.0
• 2021-06-22 – Updated for release 0.2.0
• 2021-06-11 – Updated for release 0.1.0
• 2021-06-08 – Initial release

Support and Discussion

Support and discussion is active at DecentForum.com, on Discord in the Decent Espresso server and, to some extent, on the Espresso Aficianados server in the Manufacturers: decent channel. Support is, unfortunately, not available through Decent Diaspora on Basecamp.

Thanks to all that have been trying this out and providing valuable feedback!

See also https://github.com/jeffsf/pyDE1 where the alpha branch is current.

What's New

Please see CHANGELOG.md for more details

New

A SQLite3 database now saves all profiles uploaded to the DE1, as well as capturing virtually all real-time data during all flow sequences, including a brief set of data from before the state transition.

Profiles are unique by the content of their "raw source" and also have a "fingerprint" that is common across all profiles that produce the same "program" for the DE1. Changing a profile's name alone does not change this fingerprint. Changing the frames in a profile without changing the name changes both the ID of the profile, as well as its fingerprint. These both are calculated using SHA1 from the underlying data, so should be consistent across installs for the same source data or frame set.

Profiles can also be searched by the customary metadata:

• Title
• Author
• Notes
• Beverage type
• Date added

aiosqlite and its dependencies are now required.

Legacy-style shot data can be extracted from the database by an application other that that which is running the DE1. Creating a Visualizer-compatible "file" for upload can be done in around 80-100 ms on a RPi 3B. If written to a physical file, it is also compatible with John Weiss' shot-plotting programs. See pyDE1/shot_file/legacy.py

The database retains the last-known profile uploaded to the DE1. If a flow sequence beings prior to uploading a profile, it is used as the "most likely" profile and identified in the database with the profile_assumed flag.

NB: The database needs to be manually initialized prior to use.

One approach is

sudo -u <user> sqlite3 /var/lib/pyDE1/pyDE1.sqlite3 \
< path/to/pyDE1/src/pyDE1/database/schema/schema.001.sql 

The directory also needs to be writable by the user running the code as the database is set for WAL mode to permit concurrent access from multiple threads, processes and programs.

Deprecated

Profile.from_json_file() as it is no longer needed with the API able to upload profiles. If needed within the code base, read the file, and pass to Profile.from_json() to ensure that the profile source and signatures are properly updated.

DE1._recorder_active and the contents of shot_file.py have been superseded by database logging.

Requirements

Python 3.8 or later.

Available through pip:

• bleak
• aiologger
• aiosqlite
• asyncio-mqtt
• paho-mqtt
• requests

An MQTT broker compatible with MQTT 5 clients, such as mosquitto 2.0 (see below)

The Raspberry Pi version of Debian Buster ships with Python 3.7, which does not support named asyncio.Task() The "walrus operator" is also used.

Python 3.9 is expected to be part of Debian "next". Until that time, https://github.com/pyenv/pyenv can be used to install a version of your choice. On a RPi 3B, a complete build too under 15 minutes.

Development work is being done on Buster with Python 3.9.5 on a RPi 3B at this time.

The bleak library is supported on macOS, Linux, and Windows. Some development has also been done under macOS.

Short-Term Priorities

• API to query profile database and load DE1 directly with a former profile
• Manage unexpected disconnects and reconnects
• Abort long-running actions, such as uploading firmware
• Daemonize and provide Debian-compatible service script

Known Gaps

• Timeouts on certain locks and await actions
• Single-command read of the DE1 debug register
• Clean, descale, transport
• Clean up the imports
• More doc strings and typing
• Stand-alone documentation
• Quick-start guide

Other Work

• Onboard, unattended sleep timeout with override (GUI or HA can provide complex "scheduler")
• Background firmware update
• MQTT will and MQTT 5 message expiry time

Status — Alpha

This code is work in progress and is neither feature-complete nor fully tested.

Although most features are working, as described in Section 15 and elsewhere of the GPLv3.0 LICENSE:

THERE IS NO WARRANTY FOR THE PROGRAM, TO THE EXTENT PERMITTED BY APPLICABLE LAW. EXCEPT WHEN OTHERWISE STATED IN WRITING THE COPYRIGHT HOLDERS AND/OR OTHER PARTIES PROVIDE THE PROGRAM "AS IS" WITHOUT WARRANTY OF ANY KIND, EITHER EXPRESSED OR IMPLIED, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE. THE ENTIRE RISK AS TO THE QUALITY AND PERFORMANCE OF THE PROGRAM IS WITH YOU. SHOULD THE PROGRAM PROVE DEFECTIVE, YOU ASSUME THE COST OF ALL NECESSARY SERVICING, REPAIR OR CORRECTION.

Some Older Notes of Explanatory Value

Please see CHANGELOG.md for newer details

0.2.0

Inbound Control and Query API

An inbound API has been provided using a REST-like interface over HTTP. The API should be reasonably complete in its payload and method definitions and comments are welcomed on its sufficiency and completeness.

Both the inbound and outbound APIs run in separate processes to reduce the load on the controller itself.

GET should be available for the registered resources. See, in src/pyDE1/dispatcher

• resource.py for the registered resources, and
• mapping.py for the elements they contain, the expected value types, and how they nest.

None or null are often used to me "no value", such as for stop-at limits. As a result, though similar, this is not an RFC7368 JSON Merge Patch.

In Python notation, Optional[int] means an int or None. Where float is specified, a JSON value such as 20 is permitted.

GET presently returns "unreadable" values to be able to better show the structure of the JSON. When a value is unreadable, math.nan is used internally, which is output as the JSON NaN token.

GET also returns empty nodes to illustrate the structure of the document. This can be controlled with the PRUNE_EMPTY_NODES variable in implementation.py

Although PATCH has been implemented for most payloads, PUT is not yet enabled. PUT will be the appropriate verb forDE1_PROFILE and DE1_FIRMWARE as, at this time, in-place modification of these is not supported. The API mechanism for starting a firmware upload as not been determined, as it should be able to abort as it runs in the background, as well as notify when complete. Profile upload is likely to be similar, though it occurs on a much faster time scale.

The Python http.server module is used. It is not appropriate for exposed use.

There is no security to the control and query API at this time.

See further https://docs.python.org/3/library/http.server.html

It is likely that the server, itself, will be moved to a uWSGI (or similar) process.

With either the present HTTP implementation or a future uWSGI one, use of a webserver, such as nginx, will be able to provide TLS, authentication, and authorization, as well as a more "production-ready" exposure.

Other Significant Changes

• ShotSampleWithVolumeUpdates (v1.1.0) adds de1_time. de1_time and scale_time are preferred over arrival_time as, in a future version, these will be estimates that remove some of the jitter relative to packet-arrival time.

• To be able to keep cached values of DE1 variables current, a read-back is requested on each write.

• NoneSet and NONE_SET added to some enum.IntFlag to provide clearer representations

• Although is_read_once and is_stable have been roughed in, optimizations using them have not been done

• Disabled reads of CUUID.ReadFromMMR as it returns the request itself (which is not easily distinguishable from the data read. These two interpret their Length field differently, making it difficult to determine if 5 is an unexpected value or if it was just that 6 words were requested to be read.

• Scaling on MMR0x80LowAddr.TANK_WATER_THRESHOLD was corrected.

0.1.0

Outbound API

An outbound API (notifications) is provided in a separate process. The present implementation uses MQTT and provides timestamped, source-identified, semantically versioned JSON payloads for:

• DE1
  • Connectivity
  • State updates
  • Shot samples with accumulated volume
  • Water levels
• Scale
  • Connectivity
  • Weight and flow updates
• Flow sequencer
  • "Gate" clear and set
    • Sequence start
    • Flow begin
    • Expect drops
    • Exit preinfuse
    • Flow end
    • Flow-state exit
    • Last drops
    • Sequence complete
  • Stop-at-time/volume/weight
    • Enable, disable (with target)
    • Trigger (with target and value at trigger)

An example subscriber is provided in examples/monitor_delay.py. On a Raspberry Pi 3B, running Debian Buster and mosquitto 2.0 running on ::, median delays are under 10 ms from arrival_time of the triggering event to delivery of the MQTT packet to the subscriber.

Packets are being sent with retain True, so that, for example, the subscriber has the last-known DE1 state without having to wait for a state change. Checking the payload's arrival_time is suggested to determine if the data is fresh enough. The will feature of MQTT has not yet been implemented.

A good introduction to MQTT and MQTT 5 can be found at HiveMQ:

• https://www.hivemq.com/mqtt-essentials/
• https://www.hivemq.com/blog/mqtt5-essentials-part1-introduction-to-mqtt-5/

One good thing about MQTT is that you can have as many subscribers as you want without slowing down the controller. For example, you can have a live view on your phone, live view on your desktop, log to file, log to database, all at once.

Scan For And Use First DE1 And Skale Found

Though "WET" and needing to be "DRY", the first-found DE1 and Skale will be used. The Scale class has already been designed to be able to have each subclass indicate if it recognizes the advertisement. Once DRY, the scanner should be able to return the proper scale from any of the alternatives.

Refactoring of this is pending the formal release of BleakScanner.find_device_by_filter(filterfunc) from bleak PR#565

High Level Functionality

• Connect by address to DE1
• Read and decode BLE characteristics
• Encode and write BLE characteristics
• Read and decode MMR registers
• Encode and write MMR registers
• Upload firmware
• Parse JSON profile (v2) and upload
• Connect by address to SkaleII
• Scale processing for weight and flow, including period estimation
• Stop-at-time
• Stop-at-volume
• Stop-at-weight
• Enable/disable "shot" logging
• Outbound API over MQTT
• Basic connectivity tracking
• Bleutooth scanning
• Find and use first DE1 and Skale
• Inbound control and query API over HTTP
• Save profiles and real-time data into SQLite3 with concurrent access
• Provide legacy-style, "shot file" data for Miha Rekar's Visualizer and John Weiss' shot-plotting code

The main process runs under Python's native asyncio framework. There are many tutorials out there that make asynchronous programming look easy. "Hello world!" is always easy. For a better understanding, I found Lynn Root's asyncio: We Did It Wrong to be very insightful.

Installing Mosquitto 2.0

The example outbound API uses MQTT 5. If you don't already have a local MQTT 5 broker configured, there are some public test servers ("brokers"), such as https://test.mosquitto.org/, that can let you try things out quickly. A local broker is better from both from a security standpoint and for delay. The preferred configuration is to have a broker running on the same machine as this code on a loopback interface. Unfortunately, the paho library does not support Unix domain sockets at this time.

The example outbound API does not use encryption as it runs over a socket local to the host, the data is not considered "sensitive", and there is no control over the DE1. Token-based authentication, such as password, should be done over an encrypted channel if can be "snooped" by others.

Mosquitto 2.0 is a MQTT broker that supports MQTT 5. Older distributions only supply 1.x versions, such as 1.5.7 on Debian Buster. Debian Bullseye is showing that it will support 2.0.10 at this time.

Mosquitto 2.0 can be installed onto Debian systems without needing to build from source using the Mosquitto Debian Repository. The usual caveats around making personal decisions about which sources you trust apply.

You likely will want both mosquitto (the broker) and mosquitto-clients.

Installing on RPi will enable the mosquitto.service using /etc/mosquitto/mosquitto.conf. If you've used v1.x in the past, I'd suggest reading the release announcement as well as the notes on migrating from 1.x to 2.0

Notes

The code is littered with TODOs and personal notes. Ray may find his name mentioned with some loose thoughts about changes. These are loose thoughts worthy of some future discussion, not blockers and not direct requests!