Garçon

Bot re-posting lunch posts from chosen Facebook pages on Slack.

How does it work?

The bot doesn't use Facebook's Graph API to get the posts as the Graph API doesn't allow accessing content of Facebook pages willy-nilly and that is exactly what we want to do. Instead, it scrapes necessary data directly from Facebook pages.

After extraction comes classification. To say whether a post is a lunch post or not the bot breaks it into a collection of words and searches for predefined keywords. To handle typos, misspellings, etc. the words are matched against keywords using Damerau-Levenshtein distance.

Each lunch post is then reposted to Slack using its API.

Fetched posts along with their classification, repost status, etc. are saved in a database to prevent same lunch offers from being reposted multiple times as well as allow the bot to be restarted without loosing data.

The whole procedure is repeated in regular intervals.

⚠️ I do not endorse scraping Facebook pages.

Stack

The service is written in Kotlin and uses the following stack:

• Kotlin 1.9 on Java 19 (docker image is based on JRE 21)
• Gradle 8.6 (with build script in Kotlin)
• Spring Boot 3.2
• Jooq for database access
• PostgreSQL 10+
• Kotest 5.8 and MockK 1.13 for tests
• ArchUnit 1.2 for architecture tests

Building & running locally

Always use the Gradle wrapper (./gradlew) to build the project from command line.

Useful commands:

• ./gradlew build - builds the project
• ./gradlew clean build - fully rebuilds the project
• ./gradlew test - runs all tests
• ./gradlew bootJar - build & package the service as a fat JAR
• ./gradlew bootRun - build & run the service locally
• ./gradlew generateJooq - (re)generate Jooq classes
• ./gradlew databaseUp - run a local, empty, fully migrated PostgreSQL database (convenient for testing the service locally or running integration tests from IDE)
• ./gradlew databaseDown - shut down local PostgreSQL database

During a build, a local, fully migrated PostgreSQL database is started and shut down after the build.

The service listens on HTTP port 8080 by default.

Installation

Slack application

Create a Slack app if you don't have one already:

1. Go to Slack Apps → Create New App.
2. Pick a name & workspace to which the app should belong.
3. Configure additional stuff like description & icon.

Configure permissions and Slash Commands for the app:

1. Go to Slack Apps → click on the name of your app.
2. Go to Slash Commands (under Features submenu) → Create New Command → Command: /lunch, Request URL: {BASE_URI}/commands/lunch where {BASE_URI} is the base URI under which the bot will be deployed/will handle requests → Save.
3. Go to OAuth & Permissions (under Features submenu) → Scopes section → Bot Token Scopes subsection → Add an OAuth Scope → select chat:write scope → confirm.
4. Go to OAuth & Permissions (under Features submenu) → OAuth Tokens for Your Workspace section → Take note of the Bot User OAuth Token (it starts with xoxb-). Set bot's LUNCH_SLACK_TOKEN environment variable to this value.

Install the app:

1. Go to Slack Apps → click on the name of your app.
2. Go to Install App (under Settings submenu) → Install to Workspace.
3. In Slack, go to the channel in which lunch notifications are to be received. Type /app and select Add apps to this channel. Select the Slack application created above.

Docker image

1. As described in Building & Running section create the fat JAR:

   ./gradlew bootJar
   

2. Build the docker image:

   docker build -t garcon .
   

3. Push built image to the docker registry of your choosing & deploy to your target environment.

PostgreSQL database

Create an empty PostgreSQL database for the bot with UTF-8 encoding to support emojis 😃. Take note of the credentials and make sure they allow DML & DDL queries as the bot will automatically migrate the database schema.

Environment variables

Name	Description	Required	Default/Example
PORT	HTTP port that will serve requests	✗	8080
ACTUATOR_PORT	HTTP port that will serve Actuator endpoints	✗	8081
JDBC_DATABASE_URL	JDBC URL to the database	✗	jdbc:postgresql://localhost:5432/garcon
JDBC_DATABASE_USERNAME	Username used to connect to the database	✗	garcon
JDBC_DATABASE_PASSWORD	Password used to connect to the database	✗	garcon
LUNCH_SYNC_INTERVAL	Interval between consecutive synchronizations of lunch posts.	✗	PT5M
LUNCH_CLIENT_USER_AGENT	User agent by which the client identifies itself when fetching lunch pages.	✗	Mozilla/5.0 (X11; Ubuntu; Linux x86_64; rv:80.0) Gecko/20100101 Firefox/80.0
LUNCH_CLIENT_TIMEOUT	Max time to wait for the lunch page to be fetched (expressed as ISO 8601 time duration).	✗	PT10S
LUNCH_CLIENT_RETRY_COUNT	Number of retries in case of failure.	✗	2
LUNCH_CLIENT_RETRY_MIN_JITTER	Min wait time between retries.	✗	PT0.05S
LUNCH_CLIENT_RETRY_MAX_JITTER	Max wait time between retries.	✗	PT3S
LUNCH_PAGES_<INDEX>_KEY, e.g. LUNCH_PAGES_0_KEY	Textual key of the lunch page, used as fallback for the page name when reposting. Should not change once assigned.	✓	PŻPS
LUNCH_PAGES_<INDEX>_URL, e.g. LUNCH_PAGES_0_URL	URL of the lunch page.	✓	https://www.facebook.com/1597565460485886/posts/
LUNCH_POST_LOCALE	Locale of text of posts used while extracting their keywords.	✗	Locale.ENGLISH
LUNCH_POST_KEYWORDS_<INDEX>_TEXT, e.g. LUNCH_POST_KEYWORDS_0_TEXT	The keyword that makes a post be considered as a lunch post, e.g. lunch or menu.	✗	lunch
LUNCH_POST_KEYWORDS_<INDEX>_EDIT_DISTANCE, e.g. LUNCH_POST_KEYWORDS_0_EDIT_DISTANCE	Maximum allowed Damerau-Levenshtein distance between any word from a post and the lunch keyword. Typically 1 or 2.	✗	1
LUNCH_SLACK_SIGNING_SECRET	Signing secret of the Slack app used for request verification. Request verification is disabled if the property is not set.	✗	******
LUNCH_SLACK_TOKEN	Token of the Slack app privileged to send and update reposts. Starts with xoxb-.	✓	xoxb-some-token
LUNCH_SLACK_CHANNEL	Channel ID (C1234567) or name (#random) to send reposts to.	✓	#random
LUNCH_REPOST_RETRY_INTERVAL	Interval between consecutive attempts to retry failed reposts.	✗	PT10M
LUNCH_REPOST_RETRY_BASE_DELAY	Base delay in the exponential backoff between consecutive retries of a failed repost.	✗	PT1M
LUNCH_REPOST_RETRY_MAX_ATTEMPTS	Max retry attempts for a failed repost.	✗	10

Slash commands

The following slash commands are supported:

• /lunch help - displays short help message listing supported slash commands
• /lunch or /lunch check - manually triggers checking for lunch posts
• /lunch log - displays tail of the synchronization log

Actuator

The service exposes Spring Boot Actuator endpoints under /internal prefix. By default, Actuator endpoints are available under a different port than the API - see ACTUATOR_PORT environment variable.

Metrics

The service exposes many metrics under a Prometheus scrape endpoint: /internal/prometheus.

Possible further work

• Slack configuration testing subcommand sending a test message
• Update/delete reposts based on upstream
• Custom business & technical metrics
• Adding verification of Slack request timestamps to prevent replay attacks
• Management / backoffice UI
• Instagram support

Checks

The repository contains definition of pre-commit hooks in .pre-commit-config.yaml. After installation, before each commit, it automatically runs Gitleaks on all staged changes.

To run these checks without making a commit:

• on staged files: pre-commit run,
• on all files: pre-commit run -a.