DEV Community: Pavel Kutáč

Dead code detection in Go monorepos with deadmono

Pavel Kutáč — Thu, 29 Jan 2026 11:48:18 +0000

Have you ever tried to clean up dead code in a Go monorepo with multiple services? Running deadcode on each service separately gives you conflicting results, and you can't safely delete anything. But there's a better way.

We maintain Go monorepo where multiple services share internal packages. Over time, code accumulates, and you want to clean up unused functions. But how do you know what's truly unused when you have about 10 services importing all different pacakges?

There are linters unused in GolangCI or unsedfunc in GoPLS. However those can detect only unexported unused functions within package. Go team also did a great tool called deadcode. But it accept single entrypoint and checks for unreachable code from single main function.

The problem with Simple Intersection

You might think: "Just run deadcode on each service and delete functions reported as dead in ALL services." This fails when services don't import the same packages.

Consider this monorepo structure:

├── services/
│   ├── service-a/     (imports pkg/cache, pkg/logging)
│   ├── service-b/     (imports pkg/cache, pkg/logging)
│   └── service-c/     (imports pkg/logging only)
└── pkg/
    ├── cache/
    │   ├── Get()      ✓ Used by service-a
    │   ├── Set()      ✓ Used by service-b
    │   └── Delete()   ✗ Actually dead
    └── logging/
        └── Info()

Running deadcode on each service:

service-a: Reports cache.Set(), cache.Delete() as dead
service-b: Reports cache.Get(), cache.Delete() as dead
service-c: Reports NOTHING about pkg/cache (doesn't import it!)

Simple intersection (dead in ALL services): EMPTY SET ❌

Why? service-c doesn't import pkg/cache at all, so it doesn't report anything about it. The intersection finds nothing, even though cache.Delete() is genuinely unused by all services!

The Solution: Package-Based Intersection

I created deadmono to solve this with package-based intersection:

Track which packages each service imports
Analyze each entrypoint separately
Intersect results per package (only among services that import it)
Report truly dead functions

Installation

# Install deadmono
go install github.com/arxeiss/deadmono/cmd/deadmono@latest

# Install the required deadcode tool
go install golang.org/x/tools/cmd/deadcode@latest

Usage

deadmono services/authn/main.go services/config/main.go services/healthcheck/main.go

This analyzes all three services and reports functions unused by all of them.

How It Works on the Example Above

For pkg/cache (imported by service-a, service-b):
  service-a reports: cache.Set(), cache.Delete() as dead
  service-b reports: cache.Get(), cache.Delete() as dead
  service-c: IGNORED (doesn't import pkg/cache)

  Intersection (service-a ∩ service-b):
    ✓ cache.Delete() - dead in BOTH services that import it

For pkg/logging (imported by all services):
  All services use logging.Info()
  Intersection: (empty - no dead functions)

Final Result:
  pkg/cache/cache.go:15:1: unreachable func: Delete

Now you can confidently delete cache.Delete() knowing it's truly unused.

Useful Flags

-test - Analyze test executables too
-generated - Include dead functions in generated files
-tags string - Comma-separated list of build tags
-filter string - Filter packages by regex (default: current module)
-json - Output in JSON format
-debug - Verbose debug output

Multiple Go Modules

By default, all entrypoints must be in the same module. To analyze across modules, you must provide -filter flag:

deadmono -filter "github.com/myorg/.*" \
  module1/services/api/main.go \
  module2/services/worker/main.go

If you maintain Go monorepos, give deadmono a try. It's open source and contributions are welcome!

Get GitHub PR from commit hash

Pavel Kutáč — Tue, 03 Jun 2025 12:38:54 +0000

Have you ever needed to get GitHub PR details from the commit hash in the master branch? With the Merge commits strategy, number of PR will be in the commit message. But what about Rebase and merge or Squash and merge strategy?

Lately, I have modified GitHub Action, which sends Slack message with the author of the last PR that was merged. But how to get the PR number and the author of that PR when I only have a log history?

Merge commit strategy

As I wrote above, when using the Merge commit strategy, GitHub creates a merge commit with a message that looks like this:

Merge pull request #270 from author/branch-name

fix: here is your original commit message

Here it is easy to extract the number. But before merging, the user can modify the message and the number of PR will be gone or different.

I have only commit hash

Browsing in git history with git log or similar, will not tell you much. But with GitHub API, you can ask which PR contains some commit hash.

And the best about that is, it works well in all scenarios! Including Rebase and merge and Squash and merge strategies, which create new commit. So the commit in the original PR and in the master branch have different hash.

The script below uses GitHub CLI which is pre-installed inside GitHub Actions, and by setting GH_TOKEN also authenticated.

jobs:
  your-action-name:
    runs-on: ubuntu-latest
    steps:
      - name: Checkout code
        uses: actions/checkout@v4
        with:
          fetch-depth: 10 # We are going to history a bit, so load little more to be sure

      - name: Get last PR author
        id: release-author
        env:
          GH_TOKEN: ${{ github.token }}
        run: |
          # Get last commit SHA
          # Don't use --no-merges if you use Merge strategy
          # Don't use --skip 1 unless really needed (we need that)
          last_commit_sha=$(git --no-pager log --format=%H --no-merges -n 1 --skip 1) || last_commit_sha=""
          author="unknown"

          if [ -n "$last_commit_sha" ]; then
            author=$(gh pr list --search $last_commit_sha --state merged --json author --jq '.[0].author.login') || author=""
          fi

          echo "author: $author"

I need to explain script above

git --no-pager log --format=%H --no-merges -n 1 --skip 1

git log operates on top of the git history
--no-pager turns off pagination - not necessary
--format=%H prints only full commit hash
--no-merges will skip merge commits - only if using Rebase and merge or Squash and merge
-n 1 --skip 1 instruct to return only 1 commit (-n) and --skip last one. This is important for us, as the last commit is created by Release please action. Remove skipping if not needed for you.

gh pr list --search $last_commit_sha --state merged --json author --jq '.[0].author.login'

gh pr list is using GitHub CLI to list PRs, already pre-installed in GitHub action
--search $last_commit_sha is a general search query, so we are searching by our commit hash
--state merged include only merged PRs
--json author --jq '.[0].author.login' tells the CLI tool to return JSON with the author field, which is then filtered by jq expression to give us the login of the first author

Is this bulletproof?

No, it is not, because it is using a generic search query. So if someone adds commit directly to comments, this will fail.

In that case, you can improve a query a bit to exclude comments from search --search "{$last_commit_sha} \!in:comments".

Or you can select author together with commits of PR, and then using JQ filter it out. It is however more complex, but also more reliable not to give false positives.

gh pr list --search $last_commit_sha --state=merged \
  --json author,commits \
  --jq '.[] | select(.commits[]?.oid == "'"$last_commit_sha"'") | .author.login'

If you spot another possible issue, share it with us in comments.

Automated code review requests with code owners

Pavel Kutáč — Tue, 11 Oct 2022 09:01:07 +0000

Tests, linters, build. All those and much more are very often automated in CI/CD. But you can go further and automate assigning developers to do a code review of PR. And you can specify different code owners of different parts of code as well.

Code owners are not a new feature, it was introduced on the GitHub blog already in the year 2017. But for small teams or individuals, this might not be so useful. So I learned about it just now.

What are Code owners and why you want to define them?

With code owners, you can define individual developer or the whole team, who is responsible for the code. When a new PR is created, it will automatically assign code owners to do a review, if some files owned by them are changed.

And later, in branch settings, you can specify Required review from code owners rule. And no one will be able to merge the code, until some owner approves PR.

The team as a code owner

You can specify also the whole team as code owner. Then the whole team is assigned to do a code review. But not an individual developer. To assign individual dev, you have to turn on this feature in team settings.

You can also select a member of the team, which is never assigned to do a review. In our team, it is the main architect. He is still code owner, because the team is code owner. But is not assigned to review.

Unless you specify him directly as code owner of some file. If that file is changed, he is assigned as reviewer. But not as a team member, but as code owner.

Define code owners

The config of code owners is in .github/CODEOWNERS file. Each line contains a file pattern followed by one or more developers or teams as owners. Later match takes precedence over the earlier rule.

File pattern is the same as in .gitignore file and is documented in Pattern format section of git documentation.

The final CODEOWNERS file might look like this including comments:

# Global rule set for everything. And special cases exclude later
* @mycompany/be-code-reviewers

# Pipeline files are owned by Octocat
.github/workflows/ @octocat

# Index and htaccess.
# This rules override previous one. Need to specify all of them if required.
/index.php @mycompany/be-code-reviewers @octocat
/.htaccess @mycompany/be-code-reviewers @octocat

# Dependencies
composer.json @arxeiss
composer.lock @arxeiss
/vendor/ @arxeiss

# You can also remove code owner if needed.
/CHANGELOG.md
/version.txt

Cover image is taken from GitHub blog.

Shell conditions in Gitlab CI

Pavel Kutáč — Tue, 06 Sep 2022 16:24:02 +0000

Scripts in Gitlab CI/CD might contain conditions, and there are multiple options for how to solve that. But which Shell is used and which syntax should be used then? This is not documented.

In gitlab-ci.yml file you specify the Docker image and also commands to execute. You can include conditions or other constructs in those commands. But used Docker image might contain multiple Shell implementations and it is necessary to know which Shell is selected and which syntax to use.

Manually specifying Shell in `*.sh` file

If the script is not written directly in .gitlab-ci.yml file, but in an external one, you can specify the path to Shell and the path to the file. Then you don't care much which Shell is selected in the executor. The pipeline might look like this:

deploy:
  image: registry.gitlab.com/pavel.kutac/docker-ftp-deployer:php81
  script:
    - /bin/bash /usr/local/bin/execute-deployment.sh

Condition directly inside `.gitlab-ci.yml` file

You can write conditions directly into YAML with Literal block style. Then the pipeline might look like this. Used script and Docker image are taken from Parallel incremental FTP Deployer. But now you need to know, which syntax you should use.

deploy:
  image: registry.gitlab.com/pavel.kutac/docker-ftp-deployer:php81
  script: |
    if [[ ${CI_COMMIT_MESSAGE,,} != "[skip-maintenance]"* ]]; then
      wget --no-verbose -O - https://kutac.cz/some/path/to/turn/on/maintenance-mode
    fi
    ./.ci/deploy/run-lftp-incremental-sync.sh

But how to get that info? I wasn't able to find it in the documentation. And mentioned Docker image contains Busybox sh and also bash. I had to dig into the source code of Gitlab runner.

And there it is clear, that Gitlab Runner first executes a short script with /bin/sh. BashDetectShellScript below checks different paths for different shell implementation and select one of them. The selected one is then used to execute all scripts from .gitlab-ci.yml file.

const BashDetectShellScript = `if [ -x /usr/local/bin/bash ]; then
    exec /usr/local/bin/bash $@
elif [ -x /usr/bin/bash ]; then
    exec /usr/bin/bash $@
elif [ -x /bin/bash ]; then
    exec /bin/bash $@
elif [ -x /usr/local/bin/sh ]; then
    exec /usr/local/bin/sh $@
elif [ -x /usr/bin/sh ]; then
    exec /usr/bin/sh $@
elif [ -x /bin/sh ]; then
    exec /bin/sh $@
elif [ -x /busybox/sh ]; then
    exec /busybox/sh $@
else
    echo shell not found
    exit 1
fi

`
// ...

func (b *BashShell) GetConfiguration(info common.ShellScriptInfo) (*common.ShellConfiguration, error) {
    // ...
    if info.Type == common.LoginShell {
        script.CmdLine += " -l"
        script.Arguments = []string{"-l"}
        script.DockerCommand = []string{"sh", "-c", strings.ReplaceAll(BashDetectShellScript, "$@", "-l")}
    } else {
        script.DockerCommand = []string{"sh", "-c", strings.ReplaceAll(BashDetectShellScript, "$@", "")}
    }
    // ...
}

Raspberry Pi as wireless audio receiver

Pavel Kutáč — Tue, 26 Apr 2022 09:01:14 +0000

How to turn old Hi-Fi tower into a modern audio speakers, with wireless connectivity? Use Raspberry Pi

My parents wanted to throw out old, but still functional, Hi-Fi tower. So I took it into office. However, only place where it fits is on the other side of room than tables are. I wanted to avoid laying cable on the ground and covering with duct tape. So I used Raspberry Pi, which was streaming fish in aquarium during Christmas time. And turned it into Bluetooth receiver.

Bluetooth, AirPlay and Spotify Connect

I used nice prepared script, which guides you through the process and install all required. I tested it only with Bluetooth, but it supports different types of connectivity. Script can be found on github.com/nicokaiser/rpi-audio-receiver.

After executing the script, I restarted Raspberry. Then I saw my Raspberry in the list of available Bluetooth devices on my laptot. The name of the device is the same as hostname, which might be changed during the script wizard as well.

USB Bluetooth dongle is required

In the first step, I tried to use internal onboard Bluetooth. The Readme of the repo discourage from using internal one, and I do as well. It most likely will not work, or have issues. I was able to connect, but wasn't able to play music. And second time, I wasn't able to connect anymore.

Later, I connected old USB dongle with Bluetooth v2.0. It was working immediately, but I had to manually pair devices. And it was causing some issues with my laptop, when I connected also wireless mouse. So I bought new USB dongle with Bluetooth v4.2. Since then, I don't need to manually pair devices and it works like a charm!.

Manually pairing devices

Sometimes you might be asked to enter PIN on the device, on the Raspberry. That is not easy, especially when you don't have monitor and keyboard connected. So I suggest to use SSH and connect remotely. Start Bluetooth interactive console with sudo bluetoothctl and do the steps below.

After starting console, you might see all MAC addresses that are trying to connect. Try to connect from your phone or laptop and you will see the address there. Copy it, you will need that.

# Type this after starting 'sudo bluetoothctl'
power on
agent on
pair 00:00:00:00:00:00 # Put here MAC address of your device
# Here you will be asked to enter PIN
trust 00:00:00:00:00:00 # Put here MAC address of your device as well

Gitlab runner on Raspberry Pi

Pavel Kutáč — Wed, 23 Mar 2022 12:01:09 +0000

There are many reasons why to run Gitlab runner on your own machine. And here is the tutorial how to do it with Raspberry Pi. Including installation of Docker, which is most likely needed.

The first step is to install Raspberry Pi OS to SD card. That is actually super easy. Just download and run Raspberry Pi Imager from raspberrypi.com/software. Select recommended version, target SD card, and start writing process. Then insert the SD card into Raspberry Pi and plug power. After Raspberry Pi OS will start, it suggests you to change password, connect to WiFi and install updates. I highly recommend to do all of that.

Install Docker and git

Docker will be most likely needed, while git is required. Luckily, git should be already preinstalled. Just try to run git --version to confirm that. The preferred way of installing Docker is via apt, however, that is not supported for Raspberry Pi. It is required to use the Convenience script. You can download and execute it with the command below. Because Docker requires root access, it is suggested to add the current user into the docker group. Also included in the script below.

# Downloads script and start installation
curl -fsSL https://get.docker.com -o get-docker.sh
sudo sh get-docker.sh
# Adds current user to docker group
sudo groupadd docker # Most likely returns an error, that group already exists
sudo usermod -aG docker ${USER}
groups ${USER} # List all groups, current user belongs to

After restarting, it is possible to test if Docker is running. Just execute docker run hello-world. It should print out some basic information.

Install Gitlab runner

Gitlab runner can be installed via script similarly to Docker. However, it ends up with an error, that the current distribution is not supported. Raspberry Pi OS is based on Debian but has its own ID. So it is required to print out the current OS version and pass some values into the script.

# Prints out current OS version
cat /etc/os-release
# Change os and dist variable with values from output above
curl -L "https://packages.gitlab.com/install/repositories/runner/gitlab-runner/script.deb.sh" | sudo os=debian dist=bullseye bash
# Now you can install Gitlab runner
sudo apt install gitlab-runner

Register Gitlab runner

The last step is to register Gitlab runner. One installation can handle multiple projects or project groups. The first step is to copy the registration token in CI/CD settings of project or group. It might be worth disabling shared runners too.

To start the registration process execute sudo gitlab-runner register command. It asks for multiple parameters and you should enter docker for executor parameter.

In newer version it might be necessary to run also sudo gitlab-runner install to install it as system service. Or use sudo gitlab-runner run to run until terminal is closed.

Issues with custom Docker image

If you are using in pipeline custom Docker image, you might be facing issues with incompatibility. Your computer is most likely amd64 architecture, while Raspberry is arm architecture. So your docker image will not work on Raspberry. You have to build docker image on Raspberry directly or use buildx extension. I did this for my Docker FTP Deployer repository. You can find more about Docker FTP Deployer in my article Parallel incremental FTP deploy in CI pipeline.

In Go nil is not equal to nil, sometimes

Pavel Kutáč — Tue, 22 Feb 2022 10:01:00 +0000

In some cases, Go returns false when comparing nil variable and nil. This might be confusing for beginners. But this leads to some benefits too.

The following code is a nice example when variable val is nil but the comparison in the function returns false. A similar situation is with readFile function, which returns error. You can run code in Go Playground to prove to yourself it is true.

func nilPrint(val interface{}) string {
    if val == nil {
        return "I got nil"
    }
    return fmt.Sprintf("I got %T type with value '%v'", val, val)
}

func readFile() error {
    var err *fs.PathError
    fmt.Println(err == nil)
    return err
}

func main() {
    // Prints: I got string type with value 'Some value'
    fmt.Println(nilPrint("Some value"))
    // Prints: I got nil
    fmt.Println(nilPrint(nil))
    // Prints: true
    // but then: I got *string type with value '<nil>'
    var val *string
    fmt.Println(val == nil) // Prints true
    fmt.Println(nilPrint(val))

    fmt.Println("\nWith error interface")
    // Prints: true (inside readFile)
    // but then: false (in main method)
    err := readFile()
    fmt.Println(err == nil)
}

The reason of this is interface

This behavior is caused by using interface{} as an input type. Which is something like Any. The second function returns error, which is also interface. The
comparison then returns false, because the value is null, but it has also data type. That is the reason why it is not equal to "raw" nil. If you are comparing variable with a known type, you are not affected by this behavior.

Solution with reflection

The way to fix the function which is returning error is simple. Do not set concrete error type, but stay with error type. So changing first line to this var err error. That is also best practise.

In the second case it might be little bit trickier. Changing data type is not so easy, because we want to accept anything. So the reflection package must be used here.

if val == nil || (reflect.ValueOf(val).Kind() == reflect.Ptr && reflect.ValueOf(val).IsNil()) {
    return "I got nil"
}

Calling methods on nil object

Go doesn't have methods like other OOP languages. In Go, we call it receivers, which acts similarly. But there is a main difference. If the variable of type is nil, you can still call its receivers, or methods if you want. The code below does not panic but works correctly. Check it on Go Playground.

type JSONParser struct {
    raw string
}

func (p *JSONParser) Length() int {
    if p == nil {
        return 0
    }
    return len(p.raw)
}

func main() {
    var parser *JSONParser
    fmt.Println(parser.Length()) // Prints: 0

    parser = &JSONParser{raw: `{"just": "show", "some_result": 255}`}
    fmt.Println(parser.Length()) // Prints: 36
}

Story of automated fish feeder with RaspberryPi

Pavel Kutáč — Tue, 01 Feb 2022 10:01:25 +0000

All started 2 years ago, when a colleague got a Beta fish and brought it to the office. By that day all of us started to learn how to care about aquarium. But our office building is closed every Christmas for almost 2 weeks. That might be too much for the fish without food.

Before Christmas 2020 we used our smaller tank and moved the fish to a colleague's flat. But during 2021 that tank became an Aquarium PC. So there was no spare tank to move the fish again. Automated fish feeder is must have. But the existing ones are either expensive or not compatible with our aquarium.

3D printed solution

On ThingiVerse.com we have found ideal solution. First, we had to do a small modification to properly attach stepper motor. The first prototype actually worked! But in the tank, we have also snail, not just fish. And the snail’s food was bigger than the bowl in the feeder. So we needed to do the next iteration and print the inner part again with bigger bowls.

Android Things vs RaspberryPi

I got an Android Things kit during Google Developer Days back in 2017 in Krakow, Poland. And I thought I could use it for feeder. However, Google stopped this project meanwhile. I thought it means, that only new features are not coming. But they removed also all images required to make the kit alive. I spent the whole weekend trying to find another solution to make it work, but unsuccessfully. So with colleagues, we decided to use the RaspberryPi which was unused in our office.

After we prepared RaspberryPi and we were able to control LED, we realized what will be the next challenge. The stepper motor driver needs input voltage in the range 8.2 - 45V and a logic voltage between 2.5V and 5.25V. But RaspberryPi needs 5V and has 3.3V GPIO pins. And we had an old laptop power adapter that produce 19.5V. The decision was made. We will use only 1 power supply and the feeder will contain also DC/DC step-down module to power Raspberry.

CRON with Python script and monitoring

RaspberryOS comes with some nice integration into Python, so controlling GPIO (General-purpose Inputs Outputs) is super easy. In some situations, I could call myself a control freak. I wanted to be sure, it will be working flawlessly. So besides some basic logging into the local file, I set up also monitoring with service Healthchecks.io. I was writing about it earlier.And for the worst case, we assigned a static IP address and set up port-forwarding to have access with SSH into the Raspberry.

The python script itself was executed regularly every hour by Linux Cron. And the last feeding time was controlled programmatically. So even during power strike, fish and snail would be fed after that. You can find the full script in our public Gitlab repository.

Visual feedback with non-stop stream to YouTube

Access with SSH, logging, and monitoring might check the software behavior. But as the feeder is a piece of hardware, it is much harder to monitor. So we decided to do a stream with a web camera on YouTube. We have used ffmpeg tool and the prepared script can be found again in our public Gitlab repository.

Everyones know, the developers are lazy. It is not a big surprise we kept everything to the last days. So we weren’t able to optimize streaming enough. The stream was in resolution 640x480px and 15 FPS and it still uses almost 95% of CPU time. So for safety reasons, we used another RapsberryPi just for streaming. If it would not survive, it wouldn’t affect our feeder.

Oh yes, we succeeded. But what about next year's improvements?

This article would be most likely private if the fish or the snail would die. And because you are reading it, they survived. Yaay! And we also have some baby snails. Yaay again! But there is still space for some improvements for Christmas 2022.

Food for the snail is a little bit bigger and sometimes it is stuck with fish food. So nothing or just a small portion fell into the water. It would be helpful to also control the direction of moving and do some shaking when food should fall down. And for sure we need to find a better streaming setting that would allow us to watch the stream in better quality.

Ignoring HTTP_PROXY environment vars in Go

Pavel Kutáč — Tue, 11 Jan 2022 12:01:07 +0000

Standard HTTP library in Go reads the environment variables HTTP_PROXY, HTTPS_PROXY, and NO_PROXY. But in some situations, those variables are ignored. On the other hand, the developer may want to disable the reading of those variables.

A proxy might be useful to go over the firewall of a company network when access to some sites is blocked. But a proxy is often used to debug and inspect communication during development. For example tools like HTTP Toolkit creates a local proxy server.

Every program in Go using net/http will read HTTP(S)_PROXY and NO_PROXY variables and acts accordingly. But what if the developer wants to be sure, users cannot use proxy at all? The code below is an example for explaining further issues.

func main() {
    fmt.Println(os.Getenv("HTTP_PROXY"))
    client := &http.Client{}
    resp, err := client.Get("http://localhost:8090/vm/1")
    if err != nil {
        log.Fatal(err)
    }
    body, err := ioutil.ReadAll(resp.Body)
    if err != nil {
        log.Fatal(err)
    }
    fmt.Println(string(body))
}

I set HTTP_PROXY and it is ignored

In the code above the content of HTTP_PROXY is printed out and then called GET request. But even if the value is set correctly, it is ignored. The main reason is the request is made to localhost which is ignored as well as 127.x.x.x. Luckily it is easy to go around this.

It is enough to set custom domain in /etc/hosts on Linux or C:\Windows\System32\drivers\etc\hosts on Windows by adding 127.0.0.1 localserver.loc and then change the code above.

resp, err := client.Get("http://localserver.loc:8090/vm/1")

This is documented behavior, but quite deeply. You can find it in source opensource.google/x/net/http/httpproxy/proxy.go line 118 and on line 181 is exact if which is responsible for that.

How to disable reading PROXY environment variables

You can use variable NO_PROXY to set which URLs will never go through a proxy. You can specify all with NO_PROXY=* or list URLs delimited by comma. But what to do if you as a developer wants to ignore all those variables? Then the own Transport of HTTP Client must be set.

transport := http.DefaultTransport
transport.(*http.Transport).Proxy = nil
client := &http.Client{
    Transport: transport,
}

Adding events to calendar automatically from email

Pavel Kutáč — Tue, 14 Dec 2021 12:01:14 +0000

Many email clients can show a small calendar widget with info about event sent in email. That one can also appear in the user's calendar, but only, if is well-formatted.

Method Request and Attendee

The email client can show info about the event based on attached iCalendar (*.ics) file. However, it must satisfy a few easy requirements:

Attach *.ics file with all information about the event.
Body of the *.ics file must contain property METHOD:REQUEST.
The Content-Type of the attachment must also contain method=REQUEST part.
The Attendee property must be part of the body of iCal file and must contain the email address of the recipient.

Below is the code in PHP for Laravel framework with spatie/icalendar-generator library. For more about sending emails in Laravel framework check the documentation.

use Carbon\Carbon;
use Illuminate\Notifications\Messages\MailMessage;
use Illuminate\Notifications\Notification;
use Spatie\IcalendarGenerator\Components\Calendar;
use Spatie\IcalendarGenerator\Components\Event;
use Spatie\IcalendarGenerator\Properties\TextProperty;

class EventCreatedNotification extends Notification
{
    // ...

    public function toMail(): MailMessage
    {
        $calendar = Calendar::create()
            ->productIdentifier('Kutac.cz')
            ->event(function (Event $event) {
                $event->name("Email with iCal 101")
                    ->attendee("attendee@gmail.com")
                    ->startsAt(Carbon::parse("2021-12-15 08:00:00"))
                    ->endsAt(Carbon::parse("2021-12-19 17:00:00"))
                    ->fullDay()
                    ->address('Online - Google Meet');
            });
        $calendar->appendProperty(TextProperty::create('METHOD', 'REQUEST'));        

        return (new MailMessage())
            ->subject("Invitation")
            ->markdown('mail.invite.created')
            ->attachData($calendar->get(), 'invite.ics', [
                'mime' => 'text/calendar; charset=UTF-8; method=REQUEST',
            ]);
    }
}

Issues are coming

Everything described above is enough to make it work. But every email client behaves differently. And that produces some issues. In the code I skipped the ->organizer() part, and on purpose. When the organizer is specified, email clients are sending emails when the user accepts or declines the invitation. So it can spam the organizer mailbox. Especially when the system is sending hundreds of emails.

However, web Outlook is sending those emails even if the organizer is not specified. Then the response is sent to the sender email. But what is even worse, Outlook will delete this email after accepting or declining! This can be changed in the settings, but may confuse less skilled users. Especially when the email contains tickets or payment information.

Automatic replies can be filtered out

It is possible to filter out those automatic replies. But it is not so easy, so most likely normal filter in the email client will not be enough. And probably some filter on the email server must be used.

The reply contains the same *.ics file. But instead of REQUEST method, it has method=REPLY in the Content-Type and METHOD:REPLY inside the body of the file.

More than an invitation

GMail supports much more than just a calendar widget. It can show information about your upcoming flights, hotel reservations, or Call to action button directly in the email list. But that is more complicated and you must register your application to have this available. But it is still possible and more can be found on Email markup page.

I would appreciate, if anyone solved some issues described here and would share them in comment.

Custom providers for Terraform

Pavel Kutáč — Tue, 16 Nov 2021 10:02:22 +0000

Terraform is a tool for managing infrastructure with config files. It supports many cloud providers like GCP, AWS but also Datadog, and many others. But if you are developing your own service, which should be manageable by Terraform, you have to write your own provider.

Manage infrastructure without Terraform means to log in to the Console of the provider and set up everything by clicking in the UI. However, it is hard to backup, version, and replicate for other environments like staging. And that is the reason why we have Terraform. If you don't know much about Terraform, watch Terraform in 100s video before we dive into writing own providers.

Behind the scenes

Before writing own provider, let's look at how Terraform works in the background. Terraform first reads all *.tf files and prepares dependency graph - the execution plan. Based on the plan Terraform then can perform CRUD operations. Those methods must be implemented in the provider which is responsible for calling backend API. So Terraform does not care if you use REST, gRPC, or SOAP gRPC API.

Terraform core just parses the file and performs the plan by calling the provider's CRUD functions. Theoretically, someone could write blog posts with Terraform.

Demo app

I created a sample project which contains a simple HTTP server with REST API. And also a custom provider which is communicating with that HTTP server. So you can easily play with the provider and Terraform. Full code can be found on my GitHub github.com/arxeiss/sample-terraform-provider

# Start the HTTP server
make start_server 

# Build Terraform provider
make terraform_build 

# All *.tf files are in config subfolder
cd provider/config 
terraform plan # to show plan which will be executed
terraform apply # to execute the plan

Cron monitoring for better sleep

Pavel Kutáč — Tue, 19 Oct 2021 09:01:20 +0000

Regular execution of jobs with Cron is common in almost every system. However, discovering the job stops being executed might be difficult and can take some time. So it is important to monitor it.

Bugs, which are happening just before the eyes of the user, someone will report eventually. But when background tasks are stopped, no one might notice it for even months. So monitoring is essential. And it might not be even so expensive or hard.

Healthchecks.io - Dead's man switch

In the previous article, I wrote about Cron-job service which might be good to trigger cron tasks. However, what happens, when the response code is 200 but nothing is executed? This is really hard to discover. So services like Healthchecks.io are here to help. And with a quite generous free tier.

Service is working on Dead's man switch principle. It is required to notify the service in set intervals, otherwise notifications are being sent. For example to email, SMS, phone call, Slack, Discord, Teams, Trello, WhatsApp, or one of another 16 services.

There are 2 check types. One is called Simple which accepts the interval in which the system is expecting pings. The second one is Cron, which supports basic Cron syntax to define times when to run. Both support also Grace time which defines the maximum possible delay.

Success Ping, Start, Fail, and Exit code

Only success Ping is required as a minimal starting point. By sending Ping you are telling that now the task was processed. If the service does not receive a ping at a given time or interval, notifications are sent. Endpoint Start is notifying that the task just started. Now the service is expecting Ping in defined limit since Start. This is good for detecting endless loops or stuck code.

Endpoints Fail and Exit code is there to notify about failure. With Exit code, you can specify the numeric exit code of the job.

It is possible to attach a log message with a size up to 10 kB to the requests. For more info see the documentation.

HTTP requests and retry

As documentation states, HTTP pings are not really reliable.

It is required to set reasonable timeouts and a retry policy. Here is a simplified helper class for the Laravel framework below. It would be better to use the Retry plugin, but simple for might be enough here.

<?php
namespace App\Services;

use GuzzleHttp\Client;
use Illuminate\Support\Facades\Config;
use Illuminate\Support\Facades\Log;

class HealthChecksService
{
    private $client;

    public function __construct()
    {
        $this->client = new Client([ 'base_uri' => Config::get('healthchecks.api_endpoint') ]);
    }

    private function makeRequest(string $action, string $checkName, string $log = ""): void
    {
        $options = [
            'timeout' => 2,
            'connect_timeout' => 2,
            'body' => $log,
        ];

        $exception = null;
        $i = 0;
        for (; $i < 3; $i += 1) {
            try {
                $this->client->request('POST', \rtrim("/{$action}", '/'), $options);
                $exception = null;

                break;
            } catch (\Throwable $e) {
                $exception = $e;
                \usleep(\pow(2, $i) * 500000);
            }
        }
        if ($exception) {
            Log::warning(
                'Cannot perform HealthChecks request',
                ['checkName' => $checkName, 'action' => $action, 'exception' => $exception, 'retry' => $i + 1],
            );
        }
    }

    public function start(string $checkName): void
    {
        $this->makeRequest('start', $checkName);
    }

    public function success(string $checkName, string $log = ""): void
    {
        $this->makeRequest('', $checkName, $log);
    }

    public function exitStatus(string $checkName, $exitStatus, string $log = ""): void
    {
        $this->makeRequest("$exitStatus", $checkName, $log);
    }
}