While you could think this is an angry cloud, it’s actually a fierce and determined one 😃.
The cloud team owns all work that is necessary to build, secure, scale, and operate our multi-tenant hosted version of Sourcegraph for customers that do not want to deploy Sourcegraph on-premise.
The cloud team is also responsible for all backend-infrastructure areas of ownership. A lot of the work for Sourcegraph Cloud is in direct support of scaling and performance of backend-infrastructure. Work will be balanced as needed to support both efforts.
Areas of ownership
- Authorization and authentication
- Repository management (gitserver, repo-updater, src-expose)
- Data storage and access libraries
- GraphQL API
- License generation and enforcement
- Admin and user settings
- Make it work: Build the backend blocking work, and expose it (even in a roughly usable way) to the Sourcegraph team. We will be able to quickly surface any glaring issues and will have more thoughts around usability. Take shortcuts where possible (this is currently due to the team having fewer frontend resources).
- Where possible, making it work and usable should be combined (avoid duplicate efforts), but if it’s possible to separate the two in order to move things forward, we should!
- Make it usable: The experience has been designed and thought through. We feel good about putting this in front of users, and they will find it valuable!
- Make it fast: Now that users can try it, make sure the experience is fast for them (but it’s better to have a slow working feature than a fast not-working one).
- Make it scale: Make it work at large scale. Up until now we have been starting to grow awareness of the feature, so the number of users is starting to matter. It is better to have high demand and need to surge on scalability than to make an infinitely scalable unused feature.
User configured public code on Sourcegraph Cloud
Problem: Sourcegraph Cloud proactively indexes and refreshes only a fixed set of public repositories (e.g., top 100k public repositories on GitHub.com). Code outside of this set is lazily updated when a user visits that repository, and is not stored in our search index. This does not allow individual users to search over all the code that they might care about (e.g., as user might have one or more of their own repositories that are not in the top 100k on GitHub.com).
- The Sourcegraph organization and team members can add public code from GitHub.com, GitLab.com, and Bitbucket Cloud.
- Code is indexed.
- Any user can add public code from GitHub.com, GitLab.com, and Bitbucket Cloud.
- Connecting their first code host and adding their own public repos is intuitive and easy for new and existing users.
- System activity and progress is easy to understand and doesn’t take too long.
- Repo syncing
- Repo indexing
Private code on Sourcegraph Cloud
Problem: Customers who want to use Sourcegraph with their private code must setup and run a Sourcegraph instance on their own compute infrastructure. There are customers who want to use Sourcegraph but don’t want to have to deploy and operate their own Sourcegraph instance and associated compute infrastructure. To host private code on Sourcegraph Cloud, we need to not only ensure the security of our product (which is important for on-premise deployments too), but we also have to ensure the security of our Sourcegraph Cloud infrastructure.
- The Sourcegraph organization and team members can add private code to Sourcegraph Cloud.
- No plaintext tokens or secrets anywhere (including gitserver .git/config remotes).
- All private repository content are only decryptable by Sourcegraph services.
- Searcher and code intel caches.
- Any other service that needs access to repository contents needs to be able to decrypt what it needs on the fly.
- Searching the user’s or organization’s private code versus searching all Cloud code is intuitive.
- DEPENDENCY: We first need to finish user configured public code on Sourcegraph Cloud.
- Any user or organization can add private code to Sourcegraph Cloud for free before it’s GA.
- Authorization from code hosts is enforced (e.g., organizations, teams).
- Repository visibility and permissions on Sourcegraph is intuitive.
- Adding private repositories is part of the same flows as adding public repositories.
- DEPENDENCY: We need to have visibility into our attack surface.
- Sourcegraph Cloud is Generally Available (GA).
- Abuse protection: API rate limiting, DDoS mitigation, limiting user accounts.
- Scalable syncing of permissions, repos, changesets.
- High availability, SLOs, etc.
- Billing and subscriptions.
- Confidence in our security model.
Backend infrastructure goals are ad hoc as requests come up from customers or other teams. The Cloud team is responsible for scheduling and prioritizing these requests as they come up. See the roadmap for planned items.
- #cloud channel or @cloud-team in Slack
- team/cloud label in GitHub.
Planning & grooming
On the 7th day of the cycle, we create the next cycle’s tracking issue and ask all team-mates to add work they expect or want to do to it, while ensuring it has an estimate, enough context to be picked up and is labelled appropriately (i.e. async grooming).
The engineering manager also reviews all GitHub issues with the
team/cloud label and assigns anything deemed important to teammates’ workload for next cycle (e.g. new bugs, customer issues, tech debt).
On the 10th day of the cycle, we host our planning meeting. In this meeting, we:
- Revise the work previously added to the cycle makes sense in the context of our team’s milestones.
- Ensure everyone has enough slack time (~20%).
- Check assignments take into account knowledge sharing and pairing.
- Validate estimations as a group (i.e. smoke test approaches thought of by those who groomed)
- Write a Plan section on the tracking issue in prose, describing what we’re doing this cycle and why we chose to do those things.
We do regular updates to communicate our progress to members of the team, and to external stakeholders.
Daily personal Slack updates
Collaborating across timezones requires regular communication to keep each other updated on progress, and coordinate work handoff if needed. We use daily Slack updates to achieve this.
Every day, Slackbot will post a reminder in the #cloud channel for you to write your daily update.
At the end of each working day, you should post your update as a threaded response to the Slackbot message.
You should include in your update:
- What you worked on during your day.
- Whether you’re blocked on anything to make progress (a code review, input in an RFC or in a GitHub issue…).
- What you plan on tackling next.
At the beginning of each working day, you should read the updates thread for the previous working day, to learn what your teammates have been working on, and check if they need your help.
Weekly goal updates in the tracking issue
We use weekly progress updates in the tracking issue to inform external stakeholders of the progress of the team on the iteration goals.
Every Friday, Slackbot will post a reminder in #cloud for us to write weekly progress updates for each goal in the iteration.
The teammates working on a goal are responsible for this update and should decide amongst themselves who writes this update each week.
The cloud team holds bi-weekly retrospectives.
The meeting notes can be found in this doc.
We aim to:
- Understand whether we accomplished the goals we set at the beginning of the iteration. If we didn’t accomplish them, reflect on the reasons why.
- Discuss things that didn’t go well in the iteration, and identify action items to improve on these in the next iteration.
- Discuss things that went well in the past iteration, and that we should do more of / invest more into.
The cloud team holds bi-weekly syncs.
The meeting notes can be found in this doc.
Before team syncs, teammates and stakeholders should write down under “Discussion items” in the meeting notes document anything that they’d like to bring up for discussion with the whole team.
By default, members of the team will provide a brief update about the goals they’re working on (one update per iteration goal) — this will be a short version of the written update in the tracking issue.
- We’re hiring a Product Manager for this role. Christina Forney is involved in the meantime.
- Quinn Keast (Product Designer)
- Tomás Senart (Engineering Manager)
There continues to be a need for work to happen on other areas of our backend infrastructure and the Cloud team is the best team to own that kind of work today. When this Cloud team reaches 6 engineers, we plan to split the team in half to form a new Backend Infrastructure team. When this happens, Tomás will be the manager of both teams and will look to identify new managers so we are able to continue growing both teams.
Each of these teams should not grow larger than 8 engineers. In the mean-time, Tomás will be the manager of both teams, but we’re looking to hire Engineering Managers for both down the line.