lecture:
Software Architecture
---
title:
SmartHut.sm
---
author:
Claudio Maggioni
---
# Getting started
You will use [Markdown](https://www.markdownguide.org/cheat-sheetplan), [PlantUML](https://plantuml.com/), [architectural decision records](https://github.com/adr/madr), feature models and connector views to describe a software architecture model about your own project.
This document will grow during the semester as you sketch and refine your software architecture model.
When you are done with each task, please push so we can give you feedback about your work.
We begin by selecting a suitable project domain.
# Ex - Domain Selection
{.instructions
Submit the name and brief description (about 100 words) of your domain using the following vision statement template:
```
For [target customers]
Who [need/opportunity/problem]
The [name your project]
Is [type of project]
That [major features, core benefits, compelling reason to buy]
Unlike [current reality or competitors]
Our Project [summarize main advantages over status quo, unique selling point]
```
Please indicate if your choice is:
* a project you have worked on in the past (by yourself or with a team)
* a project you are going to work on this semester in another lecture (which one?)
* a new project you plan to build in the future
* some existing open source project you are interested to contribute to
The chosen domain should be unique for each student.
Please be ready to give a 2 minute presentation about it (you can use one slide but it's not necessary)
Hint: to choose a meaningful project look at the rest of the modeling tasks which you are going to perform in the context of your domain.
}
Project Name: SmartHut.sm
Project Type: Website / Web application / IoT
Vision Statement:
- For:
- people owning smart devices or other smart home IoT equipment;
- Who:
- wish to manage such equipment from a unified web interface and API;
- The:
- SmartHut.sm application;
- Is:
- A website and REST API enabled engine;
- That:
- is able to direcly control smart devices, group them in rooms, configure and apply pre-made settings to groups, and trigger user-defined actions based on user-defined conditions;
- Unlike:
- Other smart home solutions aims to be free and open source software able to be hosted on commodity hardware owned by the user;
- Our project:
- Aims to perform most features implemented by competitors and more, allowing for ease of expandability in terms of features and supported devices.
Additional Information: This project is from the Spring semester 2020 edition of _Software Atelier 4: Software engineering project_. The original project does not support any actual smart home device, and instead mocks device state and readouts in its backend. This poses as a potential opportunity to expand the existing architecture of the project to include device support as well. More information on the project can be found [here](https://maggioni.xyz/portfolio/SmartHut.html). Source code of the existing project can be found [here](https://git.maggioni.xyz/SmartHut).
# Ex - Architectural Decision Records
{.instructions
Software architecture is about making design decisions that will impact the quality of the software you plan to build.
Let's practice how to describe an architectural decision. We will keep using ADRs to document architectural decisions in the rest of the model.
Use the following template to capture one or more architectural design decisions in the context of your project domain
Pass: 1 ADR
Good: 2 ADR
Exceed: >2 ADR
}
# Ex - Quality Attribute Scenario
{.instructions
1. Pick a scenario for a specific quality attribute. Describe it with natural language.
2. Refine the scenario using the following structure:
```puml
@startuml
skinparam componentStyle rectangle
skinparam monochrome true
skinparam shadowing false
rectangle Environment {
[Source] -> [System] : Stimulus
[System] -> [Measure] : Response
}
@enduml
```
*Stimulus*: condition affecting the system
*Source*: entity generating the stimulus
*Environment*: context under which stimulus occurred (e.g., build, test, deployment, startup, normal operation, overload, failure, attack, change)
*Response*: observable result of the stimulus
*Measure*: benchmark or target value defining a successful response
Pass: 3 scenarios
Good: >3 scenarios
Exceed: >6 scenarios using challenging qualities
}
## Scenario \#1: Live *device refresh* parameter
Quality: _Configurability_
Scenario: When deploying its own instance of the SmartHut software, the user shall be able to change the refresh timeout for checking device status without restarting the software.
```puml
@startuml
skinparam componentStyle rectangle
skinparam monochrome true
skinparam shadowing false
rectangle "When user deploys instance of SmartHut" {
rectangle "User (Admin of its own instance)" as Source
rectangle "System not restarted" as Measure
Source -> [System] : "Changes devies refresh timeout"
[System] -> [Measure] : "No restarts required"
}
@enduml
```
## Scenario \#2: Time to market for first MVP
Quality: _Time to market_
Scenario: The first MVP aimed at gathering feedback from power users shall be shipped in less than 3 months.
```puml
@startuml
skinparam componentStyle rectangle
skinparam monochrome true
skinparam shadowing false
rectangle "Development" {
rectangle "SmartHut team" as Source
rectangle "max 3 months" as Measure
Source -> [System] : "Develops"
[System] -> [Measure] : "first MVP ready"
}
@enduml
```
## Scenario \#3: Authorized access to camera live stream
Quality: _Authorization_
Scenario: Access to camera live streams shall be granted only to users explicity authorized by the user owning the camera.
```puml
@startuml
skinparam componentStyle rectangle
skinparam monochrome true
skinparam shadowing false
rectangle "Normal operation" {
rectangle "Alice" as Source
rectangle "Access granted" as Measure
Source -> [System] : "Shares camera live stream with Bob"
[System] -> [Measure] : "Bob attempts to watch camera live stream"
}
@enduml
```
```puml
@startuml
skinparam componentStyle rectangle
skinparam monochrome true
skinparam shadowing false
rectangle "Normal operation" {
rectangle "Alice" as BSource
rectangle "Access denied" as BMeasure
BSource -> [System] : "Shares camera live stream with Bob"
[System] -> [BMeasure] : "Eve attempts to watch camera live stream"
}
@enduml
```
## Scenario \#4: State change of devices is processed within two seconds
Quality: _Performance (latency)_
Scenario: When a user changes the state of a smart device from the UI (e.g. switches on a light bulb), the change shall be applied to the smart device (e.g. light turns on) within 2 seconds
```puml
@startuml
skinparam componentStyle rectangle
skinparam monochrome true
skinparam shadowing false
rectangle "Normal operation" {
rectangle "User" as Source
rectangle "max 2 seconds" as Measure
Source -> [System] : "Requests device state change from UI"
[System] -> [Measure] : "Smart device changes state"
}
@enduml
```
## Scenario \#5: First time users are proficient at defining automations
Quality: _Usability_
Scenario: During user testing, at least 75% of users, with no prior training, should be able to define a device automation workflow within 20 minutes.
```puml
@startuml
skinparam componentStyle rectangle
skinparam monochrome true
skinparam shadowing false
rectangle "User testing" {
rectangle "Untrained user" as Source
rectangle "less than 20 minutes in 75% of cases" as Measure
Source -> [System] : "Define device automation flow"
[System] -> [Measure] : "Comply to user desire"
}
@enduml
```
## Scenario \#6: Docker compatibility
Quality: _Portability_
Scenario: Power users shall be able to install their own SmartHut instance if they have a functioning docker and docker-compose installation.
```puml
@startuml
skinparam componentStyle rectangle
skinparam monochrome true
skinparam shadowing false
rectangle "Deployment (personal instance of power user)" {
rectangle "User with docker install" as Source
rectangle "Correctly and without additional tweaking" as Measure
Source -> [System] : "Installs (via docker-compose specification)"
[System] -> [Measure] : "Is initialized and starts"
}
@enduml
```
## Scenario \#7: System is resilient w.r.t. unresponsive devices
Quality: _Resilience_
Scenario: The SmartHut UI is able to notify if a smart device becomes unresponsive without crashing and within 1 minute from the onset of the malfunction.
```puml
@startuml
skinparam componentStyle rectangle
skinparam monochrome true
skinparam shadowing false
rectangle "Normal operation" {
rectangle "Smart device" as Source
rectangle "Without crashing and within 1 minute" as Measure
Source -> [System] : "Becomes unresponsive"
[System] -> [Measure] : "Copes with failure and notifies user"
}
@enduml
```
# Ex - Quality Attribute Tradeoff
{.instructions
Pick a free combination of two qualities on the [map](https://usi365.sharepoint.com/:x:/s/MSDE2023-SoftwareArchitecture/EbK1lRTVOUZJhQoz0XdUBwIBd5vd5yQblOaOwYze4ovbuA?e=6aexs6) and write your name to claim it.
Then write a short text giving an example for the tradeoff in this assignment.
Pass: 1 unique trade-off
Good: 2 trade-offs
Exceed: >2 trade-offs
}
## Feasibility vs. Configurability
Developing an highly configurable product may be beneficial as the product may be easily reusable (e.g. via white labelling), but it may be more expensive or time consuming than building an ad hoc solution. Developing an product with less configuration options may be more feasible w.r.t. deliverability to the first client. However, since configurability is impacted, future requests from that customer or other customers wishing to use the same product may be more costly and/or time consuming.
## Ease of support vs. Portability
An highily portable product is inherently more difficult to provide support for due to the possible differences between the supported platforms. In turn, a
product with high ease of support may have official documentation procedures and troubleshooting domain knowledge bound to a particular platform only because
customers typically use that platform or that platform is the only one officially supported "by contract".
## Performance vs. Modularity
An highly modular system may have lower performance than an equivalent monolith implementation due to the cost that the interfaces between microservices pose (e.g., if microservices communicate via HTTP, a monolith may be more performant than the equivalent microservice architecture due to the absence of these calls). Conversely, an highly performant architecture may be optimized in such a way that may impact modularity (e.g. functionally distinct components and abstractions may be combined for the sake of performance, even if the resulting code is less modular and more difficult to maintain - consider C++ friend classes).
# Ex - Feature Modeling
{.instructions
In the context of your chosen project domain, describe your domain using a feature model.
The feature model should be correctly visualized using the following template:
If possible, make use of all modeling constructs.
Pass: Include at least 4 non-trivial features
Good: Include at least 6 non-trivial features, which are all implemented by your project
Exceed: Include more than 8 non-trivial features, indicate which are found in your project and which belong to one competitor
}
# Ex - Context Diagram
{.instructions
Prepare a context diagram to define the design boundary for your project.
Here is a PlantUML/C4 example to get started.
Make sure to include all possible user personas and external dependencies you may need.
Pass: 1 User and 1 Dependency
Good: >1 User and >1 Dependency
Exceed: >1 User and >1 Dependency, with both incoming and outgoing dependencies
}
# Ex - Component Model: Top-Down
{.instructions
Within the context of your project domain, represent a model of your modular software architecture decomposed into components.
The number of components in your logical view should be between 6 and 9:
- At least one component should be further decomposed into sub components
- At least one component should already exist. You should plan how to reuse it, by locating it in some software repository and including in your model the exact link to its specification and its price.
- At least one component should be stateful.
The logical view should represent provide/require dependencies that are consistent with the interactions represented in the process view.
The process view should illustrate how the proposed decomposition is used to satisfy the main use case given by your domain model.
You can add additional process views showing how other use cases can be satisfied by the same set of components.
This assignment will focus on modularity-related decisions, we will worry about deployment and the container view later.
Here is a PlantUML example logical view and process view.
```puml
@startuml
skinparam componentStyle rectangle
!include
title Example Logical View
interface " " as MPI
interface " " as SRI
interface " " as CDI
interface " " as PSI
[Customer Database <$database{scale=0.33}>] as CDB
[Music Player] as MP
[User Interface] as UI
[Payment Service] as PS
[Songs Repository] as SR
MP - MPI
CDI - CDB
SRI -- SR
PSI -- PS
MPI )- UI
UI --( SRI
UI -( CDI
MP --( SRI
CDB --( PSI
skinparam monochrome true
skinparam shadowing false
skinparam defaultFontName Courier
@enduml
```
```puml
@startuml
title Example Process View: Buy and Play the Song
participant "User Interface" as UI
participant "Music Player" as MP
participant "Songs Repository" as SR
participant "Customer Database" as CDB
participant "Payment Service" as PS
UI -> SR: Browse Songs
UI -> CDB: Buy Song
CDB -> PS: Charge Customer
UI -> MP: Play Song
MP -> SR: Get Music
skinparam monochrome true
skinparam shadowing false
skinparam defaultFontName Courier
@enduml
```
Hint: How to connect sub-components to other external components? Use this pattern.
```puml
@startuml
component C {
component S
component S2
S -(0- S2
}
interface I
S - I
component C2
I )- C2
skinparam monochrome true
skinparam shadowing false
skinparam defaultFontName Courier
@enduml
```
Pass: 6 components (1 decomposed), 1 use case/process view
Good: 6 components (1 decomposed), 2 use case/process view
Exceed: >6 components (>1 decomposed) and >2 use case/process view
}
## Logical View
## Process Views
Use Case:
# Ex - Component Model: Bottom-Up
{.instructions
Within the context of your project domain, represent a model of your modular software architecture decomposed into components.
To design this model you should attempt to buy and reuse as many components as possible.
In addition to the logical and process views, you should give a precise list to all sources and prices of the components you have selected to be reused.
Write an ADR to document your component selection process (indicating which alternatives were considered).
Pass: Existing design with at least 1 reused components (1 Logical View, 1 Process View)
Good: Existing design with at least 3 reused components (1 Logical View, 1 Process View, 1 ADR)
Exceed: Redesign based on >3 reused components (1 Logical View, >1 Process View, >1 ADR)
}
# Ex - Interface/API Specification
{.instructions
In this iteration, we will detail your previous model to specify the provided interface of all components based on their interactions found in your existing process views.
1. choose whether to use the top down or bottom up model. If you specify the interfaces of the bottom up model, your interface descriptions should match what the components you reuse already offer.
2. decide which interface elements are operations, properties, or events.
Get started with one of these PlantUML templates, or you can come up with your own notation to describe the interfaces, as long as it includes all the necessary details.
The first template describes separately the provided/required interfaces of each component.
The second template annotates the logical view with the interface descriptions: less redundant, but needs the logical dependencies to be modeled to show which are the required interfaces.
Pass: define interfaces of all outer-level components
Good: Define interfaces of all outer-level components. Does your architecture publish a Web API? If not, extend it so that it does.
Exceed: Also, document the Web API using the OpenAPI language. You can use the [OpenAPI-to-Tree](http://api-ace.inf.usi.ch/openapi-to-tree/) tool to visualize the structure of your OpenAPI description.
}
# Ex - Connector View
{.instructions
Extend your existing models introducing the connector view
For every pair of connected components (logical view), pick the most suitable connector. Existing components can play the role of connector, or new connectors may need to be introduced.
Make sure that the interactions shown in the process views reflect the primitives of the selected connector
Pass: model existing connectors based on previous model decisions
Good: model existing connectors based on previous model decisions, write an ADR about the choice of one connector
Exceed: introduce a new type of connector and update your existing process view
(sequence diagram) to show the connector primitives in action
}
# Ex - Adapters and Coupling
{.instructions
1. Highlight the connectors (or components) in your existing bottom-up design playing the role of adapter. (We suggest to use the bottom-up design since when dealing with externally sourced components, their interfaces can be a source of mismatches).
2. Which kind of mismatch** are they solving?
3. Introduce a wrapper in your architecture to hide one of the previously highlighted adapters
4. Where would standard interfaces play a role in your architecture? Which standards could be relevant in your domain?
5. Explain how one or more pairs of components are coupled according to different coupling facets
6. Provide more details on how each adapter solves the mismatches identified using pseudo-code or the actual code
7. How can you improve your architectural model to minimize coupling between components? (Include a revised logical/connector view with your solution)
Pass: 1-5 (with one adapter)
Good: 1-6 (with at least two adapters)
Exceed: 1-7 (with at least two adapters)
** If you do not find any mismatch in your existing design we suggest to introduce one artificially.
## Hints
* (1) Should we find cases where two components cannot communicate (and are doing it wrongly) and highlight they would need an adapter?, or cases where we have already a "component playing the role of adapter in the view" and highlight only the adapter?
*Both are fine. We assumed that if you draw a dependency (or a connector) the interfaces match, but if you detect that the components that should communicate cannot communicate then of course introduce an adapter to solve the mismatch*
* (2) Please show the details about the two interfaces which do not match (e.g., names of parameters, object structures) so that it becomes clear why an adapter is needed and what the adapter should do to bridge the mismatch
* (5-6) These questions are about the implications on coupling based on the decisions you documented in the connector view.
Whenever you have a connector you couple together the components and different connectors will have different forms of coupling
For example, if you use calls everywhere, do you really need them everywhere? is there some pair of components where you could use a message queue instead?
Regarding the coupling facets mentioned in question 5. You do not have to answer all questions related to "discovery", "session", "binding", "interaction", "timing", "interface" and "platform" (p.441, Coupling Facets). Just the ones that you think are relevant for your design and by answering them you can get ideas on how to do question 6.
}
# Ex - Physical and Deployment Views
{.instructions
a. Extend your architectural model with the following viewpoints:
1. Physical or Container View
2. Deployment View
Your model should be non-trivial: include more than one physical device/virtual container (or both). Be ready to discuss which connectors are found at the device/container boundaries.
b. Write an ADR about which deployment strategy you plan to adopt. The alternatives to be considered are: big bang, blue/green, shadow, pilot, gradual phase-in, canary, A/B testing.
c. (Optional) Prepare a demo of a basic continuous integration and delivery pipeline for your architectural documentation so that you can obtain a single, integrated PDF with all the viewpoints you have modeled so far.
For example:
- configure a GitHub webhook to be called whenever you push changes to your documentation
- setup a GitHub action (or similar) to build and publish your documentation on a website
Pass: 1 physical view, 1 deployment view, 1 ADR (b.)
Good: >1 physical view, >1 deployment view, 1 ADR (b.)
Exceed: 1 physical view, 1 deployment view, 1 ADR (b.) + 1 demo (c.)
}
# Ex - Availability and Services
{.instructions
The goal of this week is to plan how to deliver your software as a service with high availability.
1. If necessary, change your deployment design so that your software is hosted on a server (which could be running as a Cloud VM). Your SaaS architecture should show how your SaaS can be remotely accessed from a client such as a Web browser, or a mobile app
2. Sketch your software as a service pricing model (optional)
3. How would you define the availability requirements in your project domain? For example, what would be your expectation for the duration of planned/unplanned downtimes or the longest response time tolerated by your clients?
4. Which strategy do you adopt to monitor your service's availability? Extend your architecture with a watchdog or a heartbeat monitor and motivate your choice with an ADR.
5. What happens when a stateless component goes down? model a sequence diagram to show what needs to happen to recover one of your critical stateless components
6. How do you plan to recover stateful components? write an ADR about your choice of replication strategy and whether you prefer consistency vs. availability. Also, consider whether event sourcing would help in your context.
7. How do you plan to avoid cascading failures? Be ready to discuss how the connectors (modeled in your connector view) impact the reliability of your architecture.
8. How did you mitigate the impact of your external dependencies being not available? (if applicable)
Pass: 1, 3, 4, one of: 5, 6, 7, 8
Good: 1, 2, 3, 4, two of: 5, 6, 7, 8
Exceed: 1, 2, 3, 4, 5, 6, 7, 8
}
# Ex - Flexibility
{.instructions
Only dead software stops changing. You just received a message from your customer, they have an idea. Is your architecture ready for it?
1. Pick a new use case scenario. Precisely, what exactly do you need to change of your existing architecture so that it can be supported? Model the updated logical/process/deployment views.
2. Pick another use case scenario so that it can be supported without any major architectural change (i.e., while you cannot add new components, it is possible to extend the interface of existing ones or introduce new dependencies). Illustrate with a process view, how your previous design can satisfy the new requirement.
3. Change impact. One of your externally sourced component/Web service API has announced it will introduce a breaking change. What is the impact of such change? How can you control and limit the impact of such change? Update your logical view
4. Open up your architecture so that it can be extended with plugins by its end-users. Where would be a good extension point? Update your logical view and give at least one example of what a plugin would actually do.
5. Assuming you have a centralized deployment with all stateful components storing their state in the same database, propose a strategy to split the monolith into at least two different microservices. Model the new logical/deployment view as well as the interfaces of each microservice you introduce.
Pass: 1, one out of 2-5.
Good: 1, two out of 2-5.
Exceed: 1-5.
}