Engineering

This handbook page details processes specific to working with and within this department.

Team

Contact us

• To make a request of this department, create an issue and a team member will get back to you within one business day (If urgent, mention a team member in the #help-engineering Slack channel.
  • Any Fleet team member can view the kanban boards for this department, including pending tasks and the status of new requests.
  • Please use issue comments and GitHub mentions to communicate follow-ups or answer questions related to your request.

Responsibilities

The 🚀 Engineering department at Fleet is directly responsible for writing and maintaining the code for Fleet's core product and infrastructure.

Write a feature guide

We write guides for all new features. Feature guides are published before the feature is released so that our users understand how the feature is intended to work. A guide is a type of article, so the process for writing a guide and article is the same.

1. Review and follow the Fleet writing style guide.
2. Make a copy of a guide in the /articles directory and replace the content with your article. Make sure to maintain the same heading sizes and update the metadata tags at the bottom.
3. Open a new pull request containing your article into main and add the pull request to the milestone this feature will be shipped in. The pull request will automatically be assigned to the appropriate reviewer.

Create an engineering-initiated story

Engineering-initiated stories are types of user stories created by engineers to make technical changes to Fleet. Technical changes should improve the user experience or contributor experience. For example, optimizing SQL that improves the response time of an API endpoint improves user experience by reducing latency. A script that generates common boilerplate, or automated tests to cover important business logic, improves the quality of life for contributors, making them happier and more productive, resulting in faster delivery of features to our customers.

It is important to frame engineering-initiated user stories the same way we frame all user stories. Stay focused on how this technical change will drive value for our users.

To create an engineering-initiated user story, follow the user story drafting process. Once your user story is created using the new story template, make sure the ~engineering-initiated label is added, the :product label is removed, and the engineering output and architecture DRI (@lukeheath) is assigned.

What happens next? The engineering output and architecture DRI reviews and triages engineering-initiated stories weekly on the engineering board.

If there are product changes (i.e. interface, documentation, or dependency changes), leave the default labels in place so that it is triaged on the product drafting board.

If there are no product changes, and the DRI decides to prioritize the story, the story is added to the "Specified" column on the drafting board so that it can be estimated.

We prefer the term engineering-initiated stories over technical debt because the user story format helps keep us focused on our users and contributors.

Fix a bug

All bug fix pull requests should reference the issue they resolve with the issue number in the description. Please do not use any automated words since we don't want the issues to auto-close when the PR is merged.

Create a release candidate

All minor releases go through the release candidate process before they are published. A release candidate for the next minor release is created on the Tuesday before the release at 11:00 AM Pacific. A release candidate branch is created at minor-fleet-v4.x.x and no additional feature work is merged without EM and QA approval.

All bug fixes that are merged into main after the release candidate is created are merged into the release candidate by the engineer responsible for the fix.

Run the first step of the minor release section of the Fleet releases script to create the release candidate branch, the release QA issue, and announce the release candidate in Slack.

Deploy the release candidate to QA Wolf

During the release candidate period, the release candidate is deployed to our QA Wolf instance every morning instead of main to ensure that any new bugs reported by QA Wolf are in the upcoming release and need to be fixed before publishing the release.

Open the confidential repo environment variables page and update the QAWOLF_DEPLOY_TAG repository variable with the name of the release candidate branch.

Merge unreleased bug fixes into the release candidate

Only merge unreleased bug fixes during the release candidate period to minimize code churn and help ensure a stable release. To merge a bug fix into the release candidate:

1. Merge the fix into main.
2. git checkout the release candidate branch and create a new local branch.
3. git cherry-pick your commit from main into your new local branch.
4. Create a pull request from your new branch to the release candidate.

This process ensures your bug fix is included in main for future releases, as well as the release candidate branch for the pending release.

If there is partially merged feature work when the release candidate is created, the previously merged code must be reverted. If there is an exceptional, business-critical need to merge feature work into the release candidate, as determined by the release ritual DRI, the release candidate feature merge exception process may be followed.

Request release candidate feature merge exception

1. Notify product group EM that feature work will not merge into main before the release candidate is cut and requires a feature merge exception.
2. EM notifies QA lead for the product group and the release ritual DRI.
3. EM, QA lead, and release ritual DRI must all approve the feature work PR before it is merged into the release candidate branch.

This exception process should be avoided whenever possible. Any feature work merged into the release candidate will likely result in a significant release delay.

Confirm latest versions of dependencies

Before kicking off release QA, confirm that we are using the latest versions of dependencies we want to keep up-to-date with each release. Currently, those dependencies are:

1. Go: Latest minor release

• Check the Go version specified in Fleet's go.mod file (go 1.XX.YY).
• Check the latest minor version of Go. For example, if we are using go1.19.8, and there is a new minor version go1.19.9, we will upgrade.
• If the latest minor version is greater than the version included in Fleet, file a bug and assign it to the release ritual DRI and the current oncall engineer. Add the ~release blocker label. We must upgrade to the latest minor version before publishing the next release.
• If the latest major version is greater than the version included in Fleet, create a story and assign it to the release ritual DRI and the current oncall engineer. This will be considered for an upcoming sprint. The release can proceed without upgrading the major version.

In Go versioning, the number after the first dot is the "major" version, while the number after the second dot is the "minor" version. For example, in Go 1.19.9, "19" is the major version and "9" is the minor version. Major version upgrades are assessed separately by engineering.

2. macadmins-extension: Latest release

• Check the latest version of the macadmins-extension.
• Check the version included in Fleet.
• If the latest stable version of the macadmins-extension is greater than the version included in Fleet, file a bug and assign it to the release ritual DRI and the current on-call engineer.
• Add the ~release blocker label.

Note: Some new versions of the macadmins-extension include updates that require code changes in Fleet. Make sure to note in the bug that the update should be checked for any changes, like new tables, that require code changes in Fleet.

Our goal is to keep these dependencies up-to-date with each release of Fleet. If a release is going out with an old dependency version, it should be treated as a critical bug to make sure it is updated before the release is published.

3. osquery: Latest release

• Check the latest version of osquery.
• Check the version included in Fleet.
• If the latest release of osquery is greater than the version included in Fleet, file a bug and assign it to the release ritual DRI and the current on-call engineer.
• Do not add the ~release blocker label.
• Update the bug description to note that changes to osquery command-line flags require updates to Fleet's flag validation and related documentation as shown in this pull request.

4. Vulnerability data sources

• Check the NIST National Vulnerability Database website for any announcements that might impact our NVD data feed.
• Check the CISA website for any news or announcements that might impact our CISA data feed.

If an announcement is found for either data source that may impact data feed availability, notify the current on-call engineer. Notify them that it is their responsibility to investigate and file a bug or take further action as necessary.

5. Fleetd components

• Check for code changes to Orbit or Desktop since the last orbit-* tag was published.
• Check for code changes to the fleetd-chrome extension since the last fleetd-chrome-* tag was published.

If code changes are found for any fleetd components, create a new release QA issue to update fleetd. Delete the top section for Fleet core, and retain the bottom section for fleetd. Populate the necessary version changes for each fleetd component.

Indicate your product group is release-ready

Once a product group completes its QA process during the release candidate period, its QA lead moves the smoke testing ticket to the "Ready for release" column on their ZenHub board. They then notify the release ritual DRI by tagging them in a comment, indicating that their group is prepared for release. The release ritual DRI starts the release process after all QA leads have made these updates and confirmed their readiness for release.

Prepare Fleet release

Documentation on completing the Fleet release process can be found here.

Prepare fleetd agent release

Documentation on completing the fleetd agent release process can be found here.

Deploy a new release to dogfood

After each Fleet release, the new release is deployed to Fleet's "dogfood" (internal) instance. To avoid interruptions to sales demos using this instance, deploys should occur outside of the business hours of 7am - 5pm Pacific time Monday - Friday. If a deployment is necessary during business hours, coordinate with the Sales department in the #g-sales Slack channel.

How to deploy a new release to dogfood:

1. Head to the Tags page on the fleetdm/fleet Docker Hub: https://hub.docker.com/r/fleetdm/fleet/tags
2. In the Filter tags search bar, type in the latest release (ex. v4.19.0).
3. Locate the tag for the new release and copy the image name. An example image name is "fleetdm/fleet:v4.19.0".
4. Head to the "Deploy Dogfood Environment" action on GitHub: https://github.com/fleetdm/fleet/actions/workflows/dogfood-deploy.yml
5. Select Run workflow and paste the image name in the The image tag wished to be deployed. field.

Note that this action will not handle down migrations. Always deploy a newer version than is currently deployed.

Note that "fleetdm/fleet:main" is not an image name, instead use the commit hash in place of "main".

Conclude current milestone

Immediately after publishing a new release, we close out the associated GitHub issues and milestones.

1. Rename current milestone: In GitHub, change the current milestone name from 4.x.x-tentative to 4.x.x. 4.37.0-tentative becomes 4.37.0.

2. Update product group boards: In ZenHub, go to each product group board tracking the current release. Usually, these are #g-endpoint-ops and #g-mdm.

3. Remove milestone from unfinished items: If you see any items in columns other than "Ready for release" tagged with the current milestone, remove that milestone tag. These items didn't make it into the release.

4. Prep release items: Make sure all items in the "Ready for release" column have the current milestone and sprint tags. If not, select all items in the column and apply the appropriate tags.

5. Move user stories to drafting board: Select all items in "Ready for release" that have the story label. Apply the :product label and remove the :release label. These items will move back to the product drafting board.

6. Confirm and close: Make sure that all items with the story label have left the "Ready for release" column. Select all remaining items in the "Ready for release" column and move them to the "Closed" column. This will close the related GitHub issues.

7. Confirm and celebrate: Now, head to the Drafting board. Find all story issues with the current milestone (these are the ones you just moved). Move them to the "Confirm and celebrate" column. Product will close the issues during their confirm and celebrate ritual.

8. Close GitHub milestone: Visit GitHub's milestone page and close the current milestone.

9. Create next milestone: Create a new milestone for the next versioned release, 4.x.x-tentative.

10. Announce that the release milestone has been closed in #help-engineering.

11. Visit the confidential repo variables page and update the QAWOLF_DEPLOY_TAG repository variable to main so that the latest code is deployed to QA Wolf every morning.

Update the Fleet releases calendar

The Fleet releases Google calendar is kept up-to-date by the release ritual DRI. Any change to targeted release dates is reflected on this calendar.

Handle process exceptions for non-released code

Some of our code does not go through a scheduled release process, but is released immediately via GitHub workflows. This includes:

• The fleetdm/nvd repository.
• The fleetdm/vulnerabilities repository.
• Our website directory.

In these cases there are two differences in our process:

• QA is done before merging the code change to the main branch.
• Tickets are not moved to "Ready for release". Bug are closed, and user stories are moved to the product drafting board's "Confirm and celebrate" column.

Review a community pull request

If you're assigned a community pull request for review, it is important to keep things moving for the contributor. The goal is to not go more than one business day without following up with the contributor.

A PR should be merged if:

• It's a change that is needed and useful.
• The CI is passing.
• Tests are in place.
• Documentation is updated.
• Changes file is created.

For PRs that aren't ready to merge:

• Thank the contributor for their hard work and explain why we can't merge the changes yet.
• Encourage the contributor to reach out in the #fleet channel of osquery Slack to get help from the rest of the community.
• Offer code review and coaching to help get the PR ready to go (see note below).
• Keep an eye out for any updates or responses.

Sometimes (typically for Fleet customers), a Fleet team member may add tests and make any necessary changes to merge the PR.

If everything is good to go, approve the review.

For PRs that will not be merged:

• Thank the contributor for their effort and explain why the changes won't be merged.
• Close the PR.

Merge a community pull request

When merging a pull request from a community contributor:

• Ensure that the checklist for the submitter is complete.
• Verify that all necessary reviews have been approved.
• Merge the PR.
• Thank and congratulate the contributor.
• Share the merged PR with the team in the #help-promote channel of Fleet Slack to be publicized on social media. Those who contribute to Fleet and are recognized for their contributions often become great champions for the project.

Close a stale community issue

If a community member opens an issue that we can't reproduce leave a comment asking the author for more context. After one week with no reply, close the issue with a comment letting them know they are welcome to re-open it with any updates.

Schedule developer on-call workload

Engineering Managers are asked to be aware of the on-call rotation and reduce estimate capacity for each sprint accordingly. While it varies week to week considerably, the on-call responsibilities can sometimes take up a substantial portion of the engineer's time.

On-call engineers are available during the business hours of 9am - 5pm Pacific. The on-call support SLA requires a 1-hour response time during business hours to any @oncall mention.

Assume developer on-call alias

The on-call developer is responsible for:

• Knowing the on-call rotation.
• Performing the on-call responsibilities.
• Escalating community questions and issues.
• Successfully transferring the on-call persona to the next developer.
• Working on an engineering-initiated story.

To provide full-time focus to the role, the on-call engineer is not expected to work on sprint issues during their on-call assignment.

Notify stakeholders when a user story is pushed to the next release

User stories are intended to be completed in a single sprint. When a user story selected for a release has not merged into main by the time the release candidate is created, it is the product group EM's responsibility to notify stakeholders:

1. Add the ~pushed label to the user story.
2. Update the user story's milestone to the next minor version milestone.
3. Comment on the GitHub issue and at-mention the PM and anyone listed in the requester field.
4. If customer- labels are applied to the user story, at-mention the VP of Customer Success.

Run Fleet locally for QA purposes

To try Fleet locally for QA purposes, run fleetctl preview, which defaults to running the latest stable release.

To target a different version of Fleet, use the --tag flag to target any tag in Docker Hub, including any git commit hash or branch name. For example, to QA the latest code on the main branch of fleetdm/fleet, you can run: fleetctl preview --tag=main.

To start a preview without starting the simulated hosts, use the --no-hosts flag (e.g., fleetctl preview --no-hosts).

For each bug found, please use the bug report template to create a new bug report issue.

For unreleased bugs in an active sprint, a new bug is created with the ~unreleased bug label. The :release label and associated product group label is added, and the engineer responsible for the feature is assigned. If QA is unsure who the bug should be assigned to, it is assigned to the EM. Fixing the bug becomes part of the story.

Interview a developer candidate

Ensure the interview process follows these steps in order. This process must follow creating a new position through receiving job applications. Once the position is approved manage this process per candidate in a hiring pipeline

1. Reach out: Send an email or LinkedIn message introducing yourself. Include the URL for the position, your Calendly URL, and invite the candidate to schedule a 30-minute introduction call.
2. Conduct screening call: Discuss the requirements of the position with the candidate, and answer any questions they have about Fleet. Look for alignment with Fleet's values and technical expertise necessary to meet the requirements of the role.
3. Deliver technical assessment: Download the zip of the code challenge and ask them to complete and send their project back within 5 business days.
4. Test technical assessment: Verify the code runs and completes the challenge correctly. Check the code for best practices, good style, and tests that meet our standards.
5. Schedule technical interview: Send the candidate a calendly link for 1hr to talk to a Software Engineer on your team where the goal is to understand the thechnical capabilities of the candidate. An additional Software Engineer can optionally join if available. Share the candidate's project with the Software Engineers and ask them to review in advance so they are prepared with questions about the candidate's code.
6. Schedule HOPD interview: Send the candidate a calendly link for 30m talk to the Head of Product Design @noahtalerman.
7. Schedule CTO interview: Send the candidate a calendly link for 30m talk with our CTO @lukeheath.

If the candidate passes all of these steps then continue with hiring a new team member.

Perform an incident postmortem

Conduct a postmortem meetings for every service or feature outage and every critical bug, whether it's a customer's environment or on fleetdm.com.

1. Copy this postmortem template document and pre-populate where possible.
2. Invite stakeholders. Typically the EM, PM, QA, and engineers involved. If a customer incident, include the CSM.
3. Follow and populate document topic by topic. Determine the root cause (why it happened), as well as why our controls did not catch it before release.
4. Assign each action item an owner that who is responsible for creating a Github issue promptly and working with with the relevant PM/EM to prioritize.

Example Finished Document

Provide same-day support for major version macOS releases

Beginning with macOS 16, Fleet offers same-day support for all major version macOS releases.

1. Install major version macOS beta release on test devices.
2. Create a new QA release issue with the new major version in the issue title.
3. Complete all manual smoke tests in the issue and confirm they are passing.
4. Confirm all automated tests are passing.
5. File bugs with a P1 label and assign to the appropriate product group.
6. When all bugs are fixed, follow the writing a feature guide process to publish an article announcing Fleet same-day support for the new major release.

Fix flaky Go tests

Sometimes automated tests fail intermittently, causing PRs to become blocked and engineers to become sad and vengeful. Debugging a "flaky" or "rando" test failure typically involves:

• Adding extra logs to the test and/or related code to get more information about the failure.
• Running the test multiple times to reproduce the failure.
• Implementing an attempted fix to the test (or the related code, if there's an actual bug).
• Running the test multiple times to try and verify that the test no longer fails.

To aid in this process, we have the Stress Test Go Test action (aka the RandoKiller™). This is a Github Actions workflow that can be used to run one or more Go tests repeatedly until they fail (or until they pass a certain number of times). To use the RandoKiller:

• Create a branch whose name ends with -randokiller (for example sgress454/enqueue-mdm-command-randokiller).
• Modify the .github/workflows/config/randokiller.json file to your specifications (choosing the packages and tests to run, the mysql matrix, and the number of runs to do).
• Push up the branch with whatever logs/changes you need to help diagnose or fix the flaky test.
• Monitor the Stress Test Go Test workflow for your branch.
• Repeat until the stress test passes! Every push to your branch will trigger a new run of the workflow.

Record engineering KPIs

We track the effectiveness of our processes by observing issue throughput and identifying where buildups (and therefore bottlenecks) are occurring.

The metrics are:

• Number of bugs opened this week
• Total # bugs open
• Bugs in each state (inbox, acknowledged, reproduced)
• Number of bugs closed this week

Each week these are tracked and shared in the weekly KPI sheet by Luke Heath.

Edit a DNS record

We use Cloudflare to manage the DNS records of fleetdm.com and our other domains. To make DNS changes in Cloudflare:

1. Log into your Cloudflare account and select the "Fleet" account.
2. Select the domain you want to change and go to the DNS panel on that domain's dashboard.
3. To add a record, click the "Add record" button, select the record's type, fill in the required values, and click "Save". If you're making changes to an existing record, you only need to click on the record, update the record's values, and save your changes.

If you need access to Fleet's Cloudflare account, please ask the DRI Luke Heath in Slack for an invitation.

Accept new Apple developer account terms

Engineering is responsible for managing third-party accounts required to support engineering infrastructure. We use the official Fleet Apple developer account to notarize installers we generate for Apple devices. Whenever Apple releases new terms of service, we are unable to notarize new packages until the new terms are accepted.

When this occurs, we will begin receiving the following error message when attempting to notarize packages: "You must first sign the relevant contracts online." To resolve this error, follow the steps below.

1. Visit the Apple developer account login page.

2. Log in using the credentials stored in 1Password under "Apple developer account".

3. Contact the Head of Digital Experience to determine which phone number to use for 2FA.

4. Complete the 2FA process to log in.

5. Accept the new terms of service.

Renew MDM certificate signing request (CSR)

The certificate signing request (CSR) certificate expires every year. It needs to be renewed prior to expiring. This is notified to the team by the MDM calendar event IMPORTANT: Renew MDM CSR certificate

Steps to renew the certificate:

1. Visit the Apple developer account login page.
2. Log in using the credentials stored in 1Password under Apple developer account.
3. Verify you are using the Enterprise subaccount for Fleet Device Management Inc.
4. Generate a new certificate following the instructions in MicroMDM.
5. Note: mdmctl (a micromdm command for MDM vendors) will generate a VendorPrivateKey.key and VendorCertificateRequest.csr using an appropriate shared email relay and a passphrase (suggested generation method with pwgen available in brew / apt / yum pwgen -s 32 -1vcy)
6. Uploading VendorCertificateRequest.csr to Apple you will download a corresponding mdm.cer file
7. Convert the downloaded cert to PEM with openssl x509 -inform DER -outform PEM -in mdm.cer -out server.crt.pem
8. Update the Config vars in Heroku:

• Update sails_custom__mdmVendorCertPem with the results from step 7 server.crt.pem
• Update sails_custom__mdmVendorKeyPassphrase with the passphrase used in step 4
• Update sails_custom__mdmVendorKeyPem with VendorPrivateKey.key from step 4

9. Store updated values in Confidential 1Password Vault
10. Verify by logging into a normal apple account (not billing@...) and Generate a new Push Certificate following our setup MDM steps and verify the Expiration date is 1 year from today.
11. Adjust calendar event to be between 2-4 weeks before the next expiration.

Maintain TUF repo for secure agent updates

Instructions for creating and maintaining a TUF repo are available on our TUF handbook page.

Rituals

Stubs

The following stubs are included only to make links backward compatible.